| Title: | The Time Series Modeling Companion to 'healthyR' |
|---|---|
| Description: | Hospital time series data analysis workflow tools, modeling, and automations. This library provides many useful tools to review common administrative time series hospital data. Some of these include average length of stay, and readmission rates. The aim is to provide a simple and consistent verb framework that takes the guesswork out of everything. |
| Authors: | Steven Sanderson [aut, cre, cph]
|
| Maintainer: | Steven Sanderson <[email protected]> |
| License: | MIT + file LICENSE |
| Version: | 0.3.1.9000 |
| Built: | 2024-12-12 06:32:05 UTC |
| Source: | https://github.com/spsanderson/healthyR.ts |
This function attempts to make a non-stationary time series stationary. This function attempts to make a given time series stationary by applying transformations such as differencing or logarithmic transformation. If the time series is already stationary, it returns the original time series.
auto_stationarize(.time_series)auto_stationarize(.time_series)
.time_series |
A time series object to be made stationary. |
If the input time series is non-stationary (determined by the Augmented Dickey-Fuller test), this function will try to make it stationary by applying a series of transformations:
It checks if the time series is already stationary using the Augmented Dickey-Fuller test.
If not stationary, it attempts a logarithmic transformation.
If the logarithmic transformation doesn't work, it applies differencing.
If the time series is already stationary, it returns the original time series. If a transformation is applied to make it stationary, it returns a list with two elements:
stationary_ts: The stationary time series.
ndiffs: The order of differencing applied to make it stationary.
Steven P. Sanderson II, MPH
Other Utility:
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
# Example 1: Using the AirPassengers dataset auto_stationarize(AirPassengers) # Example 2: Using the BJsales dataset auto_stationarize(BJsales)# Example 1: Using the AirPassengers dataset auto_stationarize(AirPassengers) # Example 2: Using the BJsales dataset auto_stationarize(BJsales)
This function is a helper function. It will take in a set of workflows and then
perform the modeltime::modeltime_calibrate() and modeltime::plot_modeltime_forecast().
calibrate_and_plot( ..., .type = "testing", .splits_obj, .data, .print_info = TRUE, .interactive = FALSE )calibrate_and_plot( ..., .type = "testing", .splits_obj, .data, .print_info = TRUE, .interactive = FALSE )
... |
The workflow(s) you want to add to the function. |
.type |
Either the training(splits) or testing(splits) data. |
.splits_obj |
The splits object. |
.data |
The full data set. |
.print_info |
The default is TRUE and will print out the calibration accuracy tibble and the resulting plotly plot. |
.interactive |
The defaults is FALSE. This controls if a forecast plot is interactive or not via plotly. |
This function expects to take in workflows fitted with training data.
The original time series, the simulated values and a some plots
Steven P. Sanderson II, MPH
Other Utility:
auto_stationarize(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
## Not run: suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(recipes)) suppressPackageStartupMessages(library(rsample)) suppressPackageStartupMessages(library(parsnip)) suppressPackageStartupMessages(library(workflows)) data <- ts_to_tbl(AirPassengers) %>% select(-index) splits <- timetk::time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_obj <- recipe(value ~ ., data = training(splits)) model_spec <- linear_reg( mode = "regression" , penalty = 0.1 , mixture = 0.5 ) %>% set_engine("lm") wflw <- workflow() %>% add_recipe(rec_obj) %>% add_model(model_spec) %>% fit(training(splits)) output <- calibrate_and_plot( wflw , .type = "training" , .splits_obj = splits , .data = data , .print_info = FALSE , .interactive = FALSE ) ## End(Not run)## Not run: suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(recipes)) suppressPackageStartupMessages(library(rsample)) suppressPackageStartupMessages(library(parsnip)) suppressPackageStartupMessages(library(workflows)) data <- ts_to_tbl(AirPassengers) %>% select(-index) splits <- timetk::time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_obj <- recipe(value ~ ., data = training(splits)) model_spec <- linear_reg( mode = "regression" , penalty = 0.1 , mixture = 0.5 ) %>% set_engine("lm") wflw <- workflow() %>% add_recipe(rec_obj) %>% add_model(model_spec) %>% fit(training(splits)) output <- calibrate_and_plot( wflw , .type = "training" , .splits_obj = splits , .data = data , .print_info = FALSE , .interactive = FALSE ) ## End(Not run)
Gets the upper 97.5% quantile of a numeric vector.
ci_hi(.x, .na_rm = FALSE)ci_hi(.x, .na_rm = FALSE)
.x |
A vector of numeric values |
.na_rm |
A Boolean, defaults to FALSE. Passed to the quantile function. |
Gets the upper 97.5% quantile of a numeric vector.
A numeric value.
Steven P. Sanderson II, MPH
Other Statistic:
ci_lo(),
ts_adf_test()
x <- mtcars$mpg ci_hi(x)x <- mtcars$mpg ci_hi(x)
Gets the lower 2.5% quantile of a numeric vector.
ci_lo(.x, .na_rm = FALSE)ci_lo(.x, .na_rm = FALSE)
.x |
A vector of numeric values |
.na_rm |
A Boolean, defaults to FALSE. Passed to the quantile function. |
Gets the lower 2.5% quantile of a numeric vector.
A numeric value.
Steven P. Sanderson II, MPH
Other Statistic:
ci_hi(),
ts_adf_test()
x <- mtcars$mpg ci_lo(x)x <- mtcars$mpg ci_lo(x)
8 Hex RGB color definitions suitable for charts for colorblind people.
color_blind()color_blind()
This function is used in others in order to help render plots for those that are color blind.
A vector of 8 Hex RGB definitions.
Steven P. Sanderson II, MPH
color_blind()color_blind()
This is a function that sits inside of the ts_time_event_analysis_tbl(). It
is only meant to be used there. This is an internal function.
internal_ts_backward_event_tbl(.data, .horizon)internal_ts_backward_event_tbl(.data, .horizon)
.data |
The date.frame/tibble that holds the data. |
.horizon |
How far do you want to look back or ahead. |
This is a helper function for ts_time_event_analysis_tbl() only.
A tibble.
Steven P. Sanderson II, MPH
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
This is a function that sits inside of the ts_time_event_analysis_tbl(). It
is only meant to be used there. This is an internal function.
internal_ts_both_event_tbl(.data, .horizon)internal_ts_both_event_tbl(.data, .horizon)
.data |
The date.frame/tibble that holds the data. |
.horizon |
How far do you want to look back or ahead. |
This is a helper function for ts_time_event_analysis_tbl() only.
A tibble.
Steven P. Sanderson II, MPH
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
This is a function that sits inside of the ts_time_event_analysis_tbl(). It
is only meant to be used there. This is an internal function.
internal_ts_forward_event_tbl(.data, .horizon)internal_ts_forward_event_tbl(.data, .horizon)
.data |
The date.frame/tibble that holds the data. |
.horizon |
How far do you want to look back or ahead. |
This is a helper function for ts_time_event_analysis_tbl() only.
A tibble.
Steven P. Sanderson II, MPH
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
This takes in a model fit and returns the method of the fit object.
model_extraction_helper(.fit_object)model_extraction_helper(.fit_object)
.fit_object |
A time-series fitted model |
Currently supports forecasting model of one of the following from the
forecast package:
workflow fitted models.
A model description
Steven P. Sanderson II, MPH
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
# NOT RUN ## Not run: suppressPackageStartupMessages(library(forecast)) # Create a model fit_arima <- auto.arima(AirPassengers) model_extraction_helper(fit_arima) ## End(Not run)# NOT RUN ## Not run: suppressPackageStartupMessages(library(forecast)) # Create a model fit_arima <- auto.arima(AirPassengers) model_extraction_helper(fit_arima) ## End(Not run)
step_ts_acceleration creates a a specification of a recipe
step that will convert numeric data into from a time series into its
acceleration.
step_ts_acceleration( recipe, ..., role = "predictor", trained = FALSE, columns = NULL, skip = FALSE, id = rand_id("ts_acceleration") )step_ts_acceleration( recipe, ..., role = "predictor", trained = FALSE, columns = NULL, skip = FALSE, id = rand_id("ts_acceleration") )
recipe |
A recipe object. The step will be added to the sequence of operations for this recipe. |
... |
One or more selector functions to choose which
variables that will be used to create the new variables. The
selected variables should have class |
role |
For model terms created by this step, what analysis role should they be assigned?. By default, the function assumes that the new variable columns created by the original variables will be used as predictors in a model. |
trained |
A logical to indicate if the quantities for preprocessing have been estimated. |
columns |
A character string of variables that will be
used as inputs. This field is a placeholder and will be
populated once |
skip |
A logical. Should the step be skipped when the recipe is baked by bake.recipe()? While all operations are baked when prep.recipe() is run, some operations may not be able to be conducted on new data (e.g. processing the outcome variable(s)). Care should be taken when using skip = TRUE as it may affect the computations for subsequent operations. |
id |
A character string that is unique to this step to identify it. |
Numeric Variables
Unlike other steps, step_ts_acceleration does not
remove the original numeric variables. recipes::step_rm() can be
used for this purpose.
For step_ts_acceleration, an updated version of recipe with
the new step added to the sequence of existing steps (if any).
Main Recipe Functions:
recipes::recipe()
recipes::prep()
recipes::bake()
Other Recipes:
step_ts_velocity()
suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(recipes)) len_out = 10 by_unit = "month" start_date = as.Date("2021-01-01") data_tbl <- tibble( date_col = seq.Date(from = start_date, length.out = len_out, by = by_unit), a = rnorm(len_out), b = runif(len_out) ) # Create a recipe object rec_obj <- recipe(a ~ ., data = data_tbl) %>% step_ts_acceleration(b) # View the recipe object rec_obj # Prepare the recipe object prep(rec_obj) # Bake the recipe object - Adds the Time Series Signature bake(prep(rec_obj), data_tbl) rec_obj %>% prep() %>% juice()suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(recipes)) len_out = 10 by_unit = "month" start_date = as.Date("2021-01-01") data_tbl <- tibble( date_col = seq.Date(from = start_date, length.out = len_out, by = by_unit), a = rnorm(len_out), b = runif(len_out) ) # Create a recipe object rec_obj <- recipe(a ~ ., data = data_tbl) %>% step_ts_acceleration(b) # View the recipe object rec_obj # Prepare the recipe object prep(rec_obj) # Bake the recipe object - Adds the Time Series Signature bake(prep(rec_obj), data_tbl) rec_obj %>% prep() %>% juice()
step_ts_velocity creates a a specification of a recipe
step that will convert numeric data into from a time series into its
velocity.
step_ts_velocity( recipe, ..., role = "predictor", trained = FALSE, columns = NULL, skip = FALSE, id = rand_id("ts_velocity") )step_ts_velocity( recipe, ..., role = "predictor", trained = FALSE, columns = NULL, skip = FALSE, id = rand_id("ts_velocity") )
recipe |
A recipe object. The step will be added to the sequence of operations for this recipe. |
... |
One or more selector functions to choose which
variables that will be used to create the new variables. The
selected variables should have class |
role |
For model terms created by this step, what analysis role should they be assigned?. By default, the function assumes that the new variable columns created by the original variables will be used as predictors in a model. |
trained |
A logical to indicate if the quantities for preprocessing have been estimated. |
columns |
A character string of variables that will be
used as inputs. This field is a placeholder and will be
populated once |
skip |
A logical. Should the step be skipped when the recipe is baked by bake.recipe()? While all operations are baked when prep.recipe() is run, some operations may not be able to be conducted on new data (e.g. processing the outcome variable(s)). Care should be taken when using skip = TRUE as it may affect the computations for subsequent operations. |
id |
A character string that is unique to this step to identify it. |
Numeric Variables
Unlike other steps, step_ts_velocity does not
remove the original numeric variables. recipes::step_rm() can be
used for this purpose.
For step_ts_velocity, an updated version of recipe with
the new step added to the sequence of existing steps (if any).
Main Recipe Functions:
recipes::recipe()
recipes::prep()
recipes::bake()
Other Recipes:
step_ts_acceleration()
suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(recipes)) len_out = 10 by_unit = "month" start_date = as.Date("2021-01-01") data_tbl <- tibble( date_col = seq.Date(from = start_date, length.out = len_out, by = by_unit), a = rnorm(len_out), b = runif(len_out) ) # Create a recipe object rec_obj <- recipe(a ~ ., data = data_tbl) %>% step_ts_velocity(b) # View the recipe object rec_obj # Prepare the recipe object prep(rec_obj) # Bake the recipe object - Adds the Time Series Signature bake(prep(rec_obj), data_tbl) rec_obj %>% prep() %>% juice()suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(recipes)) len_out = 10 by_unit = "month" start_date = as.Date("2021-01-01") data_tbl <- tibble( date_col = seq.Date(from = start_date, length.out = len_out, by = by_unit), a = rnorm(len_out), b = runif(len_out) ) # Create a recipe object rec_obj <- recipe(a ~ ., data = data_tbl) %>% step_ts_velocity(b) # View the recipe object rec_obj # Prepare the recipe object prep(rec_obj) # Bake the recipe object - Adds the Time Series Signature bake(prep(rec_obj), data_tbl) rec_obj %>% prep() %>% juice()
Perform an fft using stats::fft() and return a tidier style output list with plots.
tidy_fft( .data, .date_col, .value_col, .frequency = 12L, .harmonics = 1L, .upsampling = 10L )tidy_fft( .data, .date_col, .value_col, .frequency = 12L, .harmonics = 1L, .upsampling = 10L )
.data |
The data.frame/tibble you will pass for analysis. |
.date_col |
The column that holds the date. |
.value_col |
The column that holds the data to be analyzed. |
.frequency |
The frequency of the data, 12 = monthly for example. |
.harmonics |
How many harmonic waves do you want to produce. |
.upsampling |
The up sampling of the time series. |
This function will perform a few different things, but primarily it will
compute the Fast Discrete Fourier Transform (FFT) using stats::fft(). The
formula is given as:
There are many items returned inside of a list invisibly. There are four primary categories of data returned in the list. Below are the primary categories and the items inside of them.
data:
data
error_data
input_vector
maximum_harmonic_tbl
differenced_value_tbl
dff_tbl
ts_obj
plots:
harmonic_plot
diff_plot
max_har_plot
harmonic_plotly
max_har_plotly
parameters:
harmonics
upsampling
start_date
end_date
freq
model:
m
harmonic_obj
harmonic_model
model_summary
A list object returned invisibly.
Steven P. Sanderson II, MPH
Other Data Generator:
ts_brownian_motion(),
ts_brownian_motion_augment(),
ts_geometric_brownian_motion(),
ts_geometric_brownian_motion_augment(),
ts_random_walk()
suppressPackageStartupMessages(library(dplyr)) data_tbl <- AirPassengers %>% ts_to_tbl() %>% select(-index) a <- tidy_fft( .data = data_tbl, .value_col = value, .date_col = date_col, .harmonics = 3, .frequency = 12 ) a$plots$max_har_plot a$plots$harmonic_plotsuppressPackageStartupMessages(library(dplyr)) data_tbl <- AirPassengers %>% ts_to_tbl() %>% select(-index) a <- tidy_fft( .data = data_tbl, .value_col = value, .date_col = date_col, .harmonics = 3, .frequency = 12 ) a$plots$max_har_plot a$plots$harmonic_plot
Takes a numeric vector and will return the acceleration of that vector.
ts_acceleration_augment(.data, .value, .names = "auto")ts_acceleration_augment(.data, .value, .names = "auto")
.data |
The data being passed that will be augmented by the function. |
.value |
This is passed |
.names |
The default is "auto" |
Takes a numeric vector and will return the acceleration of that vector. The acceleration of a time series is computed by taking the second difference, so
This function is intended to be used on its own in order to add columns to a tibble.
A augmented tibble
Steven P. Sanderson II, MPH
Other Augment Function:
ts_growth_rate_augment(),
ts_velocity_augment()
suppressPackageStartupMessages(library(dplyr)) len_out = 10 by_unit = "month" start_date = as.Date("2021-01-01") data_tbl <- tibble( date_col = seq.Date(from = start_date, length.out = len_out, by = by_unit), a = rnorm(len_out), b = runif(len_out) ) ts_acceleration_augment(data_tbl, b)suppressPackageStartupMessages(library(dplyr)) len_out = 10 by_unit = "month" start_date = as.Date("2021-01-01") data_tbl <- tibble( date_col = seq.Date(from = start_date, length.out = len_out, by = by_unit), a = rnorm(len_out), b = runif(len_out) ) ts_acceleration_augment(data_tbl, b)
Takes a numeric vector and will return the acceleration of that vector.
ts_acceleration_vec(.x)ts_acceleration_vec(.x)
.x |
A numeric vector |
Takes a numeric vector and will return the acceleration of that vector. The acceleration of a time series is computed by taking the second difference, so
This function can be used on it's own. It is also the basis for the function
ts_acceleration_augment().
A numeric vector
Steven P. Sanderson II, MPH
Other Vector Function:
ts_growth_rate_vec(),
ts_velocity_vec()
suppressPackageStartupMessages(library(dplyr)) len_out = 25 by_unit = "month" start_date = as.Date("2021-01-01") data_tbl <- tibble( date_col = seq.Date(from = start_date, length.out = len_out, by = by_unit), a = rnorm(len_out), b = runif(len_out) ) vec_1 <- ts_acceleration_vec(data_tbl$b) plot(data_tbl$b) lines(data_tbl$b) lines(vec_1, col = "blue")suppressPackageStartupMessages(library(dplyr)) len_out = 25 by_unit = "month" start_date = as.Date("2021-01-01") data_tbl <- tibble( date_col = seq.Date(from = start_date, length.out = len_out, by = by_unit), a = rnorm(len_out), b = runif(len_out) ) vec_1 <- ts_acceleration_vec(data_tbl$b) plot(data_tbl$b) lines(data_tbl$b) lines(vec_1, col = "blue")
This function performs the Augmented Dickey-Fuller test to assess the
stationarity of a time series. The Augmented Dickey-Fuller (ADF) test is used
to determine if a given time series is stationary. This function takes a
numeric vector as input, and you can optionally specify the lag order with
the .k parameter. If .k is not provided, it is calculated based on the
number of observations using a formula. The test statistic and p-value are
returned.
ts_adf_test(.x, .k = NULL)ts_adf_test(.x, .k = NULL)
.x |
A numeric vector representing the time series to be tested for stationarity. |
.k |
An optional parameter specifying the number of lags to use in the ADF test (default is calculated). |
A list containing the results of the Augmented Dickey-Fuller test:
test_stat: The test statistic from the ADF test.
p_value: The p-value of the test.
Steven P. Sanderson II, MPH
Other Statistic:
ci_hi(),
ci_lo()
# Example 1: Using the AirPassengers dataset ts_adf_test(AirPassengers) # Example 2: Using a custom time series vector custom_ts <- rnorm(100, 0, 1) ts_adf_test(custom_ts)# Example 1: Using the AirPassengers dataset ts_adf_test(AirPassengers) # Example 2: Using a custom time series vector custom_ts <- rnorm(100, 0, 1) ts_adf_test(custom_ts)
Returns a list output of any n simulations of a user specified
ARIMA model. The function returns a list object with two sections:
data
plots
The data section of the output contains the following:
simulation_time_series object (ts format)
simulation_time_series_output (mts format)
simulations_tbl (simulation_time_series_object in a tibble)
simulations_median_value_tbl (contains the stats::median() value of the
simulated data)
The plots section of the output contains the following:
static_plot The ggplot2 plot
plotly_plot The plotly plot
ts_arima_simulator( .n = 100, .num_sims = 25, .order_p = 0, .order_d = 0, .order_q = 0, .ma = c(), .ar = c(), .sim_color = "steelblue", .alpha = 0.05, .size = 1, ... )ts_arima_simulator( .n = 100, .num_sims = 25, .order_p = 0, .order_d = 0, .order_q = 0, .ma = c(), .ar = c(), .sim_color = "steelblue", .alpha = 0.05, .size = 1, ... )
.n |
The number of points to be simulated. |
.num_sims |
The number of different simulations to be run. |
.order_p |
The p value, the order of the AR term. |
.order_d |
The d value, the number of differencing to make the series stationary |
.order_q |
The q value, the order of the MA term. |
.ma |
You can list the MA terms respectively if desired. |
.ar |
You can list the AR terms respectively if desired. |
.sim_color |
The color of the lines for the simulated series. |
.alpha |
The alpha component of the |
.size |
The size of the median line for the |
... |
Any other additional arguments for stats::arima.sim |
This function takes in a user specified arima model. The specification
is passed to stats::arima.sim()
A list object.
Steven P. Sanderson II, MPH
https://www.machinelearningplus.com/time-series/arima-model-time-series-forecasting-python/
Other Simulator:
ts_forecast_simulator()
output <- ts_arima_simulator() output$plots$static_plotoutput <- ts_arima_simulator() output$plots$static_plot
This is a boilerplate function to create automatically the following:
recipe
model specification
workflow
tuned model (grid ect)
calibration tibble and plot
ts_auto_arima( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_arima", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )ts_auto_arima( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_arima", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )
.data |
The data being passed to the function. The time-series object. |
.date_col |
The column that holds the datetime. |
.value_col |
The column that has the value |
.formula |
The formula that is passed to the recipe like |
.rsamp_obj |
The rsample splits object |
.prefix |
Default is |
.tune |
Defaults to TRUE, this creates a tuning grid and tuned model. |
.grid_size |
If |
.num_cores |
How many cores do you want to use. Default is 1 |
.cv_assess |
How many observations for assess. See |
.cv_skip |
How many observations to skip. See |
.cv_slice_limit |
How many slices to return. See |
.best_metric |
Default is "rmse". See |
.bootstrap_final |
Not yet implemented. |
This uses the modeltime::arima_reg() with the engine set to arima
A list
Steven P. Sanderson II, MPH
https://business-science.github.io/modeltime/reference/arima_reg.html
Other Boiler_Plate:
ts_auto_arima_xgboost(),
ts_auto_croston(),
ts_auto_exp_smoothing(),
ts_auto_glmnet(),
ts_auto_lm(),
ts_auto_mars(),
ts_auto_nnetar(),
ts_auto_prophet_boost(),
ts_auto_prophet_reg(),
ts_auto_smooth_es(),
ts_auto_svm_poly(),
ts_auto_svm_rbf(),
ts_auto_theta(),
ts_auto_xgboost()
library(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_aa <- ts_auto_arima( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 5, .cv_slice_limit = 2, .tune = FALSE ) ts_aa$recipe_infolibrary(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_aa <- ts_auto_arima( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 5, .cv_slice_limit = 2, .tune = FALSE ) ts_aa$recipe_info
This is a boilerplate function to create automatically the following:
recipe
model specification
workflow
tuned model (grid ect)
calibration tibble and plot
ts_auto_arima_xgboost( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_arima_boost", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )ts_auto_arima_xgboost( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_arima_boost", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )
.data |
The data being passed to the function. The time-series object. |
.date_col |
The column that holds the datetime. |
.value_col |
The column that has the value |
.formula |
The formula that is passed to the recipe like |
.rsamp_obj |
The rsample splits object |
.prefix |
Default is |
.tune |
Defaults to TRUE, this creates a tuning grid and tuned model. |
.grid_size |
If |
.num_cores |
How many cores do you want to use. Default is 1 |
.cv_assess |
How many observations for assess. See |
.cv_skip |
How many observations to skip. See |
.cv_slice_limit |
How many slices to return. See |
.best_metric |
Default is "rmse". See |
.bootstrap_final |
Not yet implemented. |
This uses the modeltime::arima_boost() with the engine set to xgboost
A list
Steven P. Sanderson II, MPH
https://business-science.github.io/modeltime/reference/arima_boost.html
Other Boiler_Plate:
ts_auto_arima(),
ts_auto_croston(),
ts_auto_exp_smoothing(),
ts_auto_glmnet(),
ts_auto_lm(),
ts_auto_mars(),
ts_auto_nnetar(),
ts_auto_prophet_boost(),
ts_auto_prophet_reg(),
ts_auto_smooth_es(),
ts_auto_svm_poly(),
ts_auto_svm_rbf(),
ts_auto_theta(),
ts_auto_xgboost()
library(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_auto_arima_xgboost <- ts_auto_arima_xgboost( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 5, .cv_slice_limit = 2, .tune = FALSE ) ts_auto_arima_xgboost$recipe_infolibrary(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_auto_arima_xgboost <- ts_auto_arima_xgboost( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 5, .cv_slice_limit = 2, .tune = FALSE ) ts_auto_arima_xgboost$recipe_info
This is a boilerplate function to create automatically the following:
recipe
model specification
workflow
tuned model (grid ect)
calibration tibble and plot
ts_auto_croston( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_croston", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )ts_auto_croston( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_croston", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )
.data |
The data being passed to the function. The time-series object. |
.date_col |
The column that holds the datetime. |
.value_col |
The column that has the value |
.formula |
The formula that is passed to the recipe like |
.rsamp_obj |
The rsample splits object |
.prefix |
Default is |
.tune |
Defaults to TRUE, this creates a tuning grid and tuned model. |
.grid_size |
If |
.num_cores |
How many cores do you want to use. Default is 1 |
.cv_assess |
How many observations for assess. See |
.cv_skip |
How many observations to skip. See |
.cv_slice_limit |
How many slices to return. See |
.best_metric |
Default is "rmse". See |
.bootstrap_final |
Not yet implemented. |
This uses the forecast::croston() for the parsnip engine. This
model does not use exogenous regressors, so only a univariate model of: value ~ date
will be used from the .date_col and .value_col that you provide.
A list
Steven P. Sanderson II, MPH
https://business-science.github.io/modeltime/reference/exp_smoothing.html#engine-details
https://pkg.robjhyndman.com/forecast/reference/croston.html
Other Boiler_Plate:
ts_auto_arima(),
ts_auto_arima_xgboost(),
ts_auto_exp_smoothing(),
ts_auto_glmnet(),
ts_auto_lm(),
ts_auto_mars(),
ts_auto_nnetar(),
ts_auto_prophet_boost(),
ts_auto_prophet_reg(),
ts_auto_smooth_es(),
ts_auto_svm_poly(),
ts_auto_svm_rbf(),
ts_auto_theta(),
ts_auto_xgboost()
Other exp_smoothing:
ts_auto_exp_smoothing(),
ts_auto_smooth_es(),
ts_auto_theta()
library(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_exp <- ts_auto_croston( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 5, .tune = FALSE ) ts_exp$recipe_infolibrary(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_exp <- ts_auto_croston( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 5, .tune = FALSE ) ts_exp$recipe_info
This is a boilerplate function to create automatically the following:
recipe
model specification
workflow
tuned model (grid ect)
calibration tibble and plot
ts_auto_exp_smoothing( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_exp_smooth", .tune = TRUE, .grid_size = 20, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )ts_auto_exp_smoothing( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_exp_smooth", .tune = TRUE, .grid_size = 20, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )
.data |
The data being passed to the function. The time-series object. |
.date_col |
The column that holds the datetime. |
.value_col |
The column that has the value |
.formula |
The formula that is passed to the recipe like |
.rsamp_obj |
The rsample splits object |
.prefix |
Default is |
.tune |
Defaults to TRUE, this creates a tuning grid and tuned model. |
.grid_size |
If |
.num_cores |
How many cores do you want to use. Default is 1 |
.cv_assess |
How many observations for assess. See |
.cv_skip |
How many observations to skip. See |
.cv_slice_limit |
How many slices to return. See |
.best_metric |
Default is "rmse". See |
.bootstrap_final |
Not yet implemented. |
This uses modeltime::exp_smoothing() under the hood with the engine
set to ets
A list
Steven P. Sanderson II, MPH
https://business-science.github.io/modeltime/reference/exp_smoothing.html#engine-details
https://pkg.robjhyndman.com/forecast/reference/ets.html
Other Boiler_Plate:
ts_auto_arima(),
ts_auto_arima_xgboost(),
ts_auto_croston(),
ts_auto_glmnet(),
ts_auto_lm(),
ts_auto_mars(),
ts_auto_nnetar(),
ts_auto_prophet_boost(),
ts_auto_prophet_reg(),
ts_auto_smooth_es(),
ts_auto_svm_poly(),
ts_auto_svm_rbf(),
ts_auto_theta(),
ts_auto_xgboost()
Other exp_smoothing:
ts_auto_croston(),
ts_auto_smooth_es(),
ts_auto_theta()
library(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_exp <- ts_auto_exp_smoothing( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 20, .tune = FALSE ) ts_exp$recipe_infolibrary(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_exp <- ts_auto_exp_smoothing( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 20, .tune = FALSE ) ts_exp$recipe_info
This is a boilerplate function to create automatically the following:
recipe
model specification
workflow
tuned model (grid ect)
calibration tibble and plot
ts_auto_glmnet( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_glmnet", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )ts_auto_glmnet( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_glmnet", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )
.data |
The data being passed to the function. The time-series object. |
.date_col |
The column that holds the datetime. |
.value_col |
The column that has the value |
.formula |
The formula that is passed to the recipe like |
.rsamp_obj |
The rsample splits object |
.prefix |
Default is |
.tune |
Defaults to TRUE, this creates a tuning grid and tuned model. |
.grid_size |
If |
.num_cores |
How many cores do you want to use. Default is 1 |
.cv_assess |
How many observations for assess. See |
.cv_skip |
How many observations to skip. See |
.cv_slice_limit |
How many slices to return. See |
.best_metric |
Default is "rmse". See |
.bootstrap_final |
Not yet implemented. |
This uses parsnip::linear_reg() and sets the engine to glmnet
A list
Steven P. Sanderson II, MPH
https://parsnip.tidymodels.org/reference/linear_reg.html
Other Boiler_Plate:
ts_auto_arima(),
ts_auto_arima_xgboost(),
ts_auto_croston(),
ts_auto_exp_smoothing(),
ts_auto_lm(),
ts_auto_mars(),
ts_auto_nnetar(),
ts_auto_prophet_boost(),
ts_auto_prophet_reg(),
ts_auto_smooth_es(),
ts_auto_svm_poly(),
ts_auto_svm_rbf(),
ts_auto_theta(),
ts_auto_xgboost()
library(dplyr) library(timetk) library(modeltime) library(glmnet) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_glmnet <- ts_auto_glmnet( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 5, .tune = FALSE ) ts_glmnet$recipe_infolibrary(dplyr) library(timetk) library(modeltime) library(glmnet) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_glmnet <- ts_auto_glmnet( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 5, .tune = FALSE ) ts_glmnet$recipe_info
This is a boilerplate function to create automatically the following:
recipe
model specification
workflow
calibration tibble and plot
ts_auto_lm( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_lm", .bootstrap_final = FALSE )ts_auto_lm( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_lm", .bootstrap_final = FALSE )
.data |
The data being passed to the function. The time-series object. |
.date_col |
The column that holds the datetime. |
.value_col |
The column that has the value |
.formula |
The formula that is passed to the recipe like |
.rsamp_obj |
The rsample splits object |
.prefix |
Default is |
.bootstrap_final |
Not yet implemented. |
This uses parsnip::linear_reg() and sets the engine to lm
A list
Steven P. Sanderson II, MPH
https://parsnip.tidymodels.org/reference/linear_reg.html
Other Boiler_Plate:
ts_auto_arima(),
ts_auto_arima_xgboost(),
ts_auto_croston(),
ts_auto_exp_smoothing(),
ts_auto_glmnet(),
ts_auto_mars(),
ts_auto_nnetar(),
ts_auto_prophet_boost(),
ts_auto_prophet_reg(),
ts_auto_smooth_es(),
ts_auto_svm_poly(),
ts_auto_svm_rbf(),
ts_auto_theta(),
ts_auto_xgboost()
library(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_lm <- ts_auto_lm( .data = data, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., ) ts_lm$recipe_infolibrary(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_lm <- ts_auto_lm( .data = data, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., ) ts_lm$recipe_info
This is a boilerplate function to create automatically the following:
recipe
model specification
workflow
tuned model (grid ect)
calibration tibble and plot
ts_auto_mars( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_mars", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )ts_auto_mars( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_mars", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )
.data |
The data being passed to the function. The time-series object. |
.date_col |
The column that holds the datetime. |
.value_col |
The column that has the value |
.formula |
The formula that is passed to the recipe like |
.rsamp_obj |
The rsample splits object |
.prefix |
Default is |
.tune |
Defaults to TRUE, this creates a tuning grid and tuned model. |
.grid_size |
If |
.num_cores |
How many cores do you want to use. Default is 1 |
.cv_assess |
How many observations for assess. See |
.cv_skip |
How many observations to skip. See |
.cv_slice_limit |
How many slices to return. See |
.best_metric |
Default is "rmse". See |
.bootstrap_final |
Not yet implemented. |
This uses the parsnip::mars() function with the engine set to earth.
A list
Steven P. Sanderson II, MPH
https://parsnip.tidymodels.org/reference/mars.html
Other Boiler_Plate:
ts_auto_arima(),
ts_auto_arima_xgboost(),
ts_auto_croston(),
ts_auto_exp_smoothing(),
ts_auto_glmnet(),
ts_auto_lm(),
ts_auto_nnetar(),
ts_auto_prophet_boost(),
ts_auto_prophet_reg(),
ts_auto_smooth_es(),
ts_auto_svm_poly(),
ts_auto_svm_rbf(),
ts_auto_theta(),
ts_auto_xgboost()
library(dplyr) library(timetk) library(modeltime) library(earth) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_auto_mars <- ts_auto_mars( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 20, .tune = FALSE ) ts_auto_mars$recipe_infolibrary(dplyr) library(timetk) library(modeltime) library(earth) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_auto_mars <- ts_auto_mars( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 20, .tune = FALSE ) ts_auto_mars$recipe_info
This is a boilerplate function to create automatically the following:
recipe
model specification
workflow
tuned model (grid ect)
calibration tibble and plot
ts_auto_nnetar( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_nnetar", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )ts_auto_nnetar( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_nnetar", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )
.data |
The data being passed to the function. The time-series object. |
.date_col |
The column that holds the datetime. |
.value_col |
The column that has the value |
.formula |
The formula that is passed to the recipe like |
.rsamp_obj |
The rsample splits object |
.prefix |
Default is |
.tune |
Defaults to TRUE, this creates a tuning grid and tuned model. |
.grid_size |
If |
.num_cores |
How many cores do you want to use. Default is 1 |
.cv_assess |
How many observations for assess. See |
.cv_skip |
How many observations to skip. See |
.cv_slice_limit |
How many slices to return. See |
.best_metric |
Default is "rmse". See |
.bootstrap_final |
Not yet implemented. |
This uses the modeltime::nnetar_reg() function with the engine set to nnetar.
A list
Steven P. Sanderson II, MPH
https://business-science.github.io/modeltime/reference/nnetar_reg.html
Other Boiler_Plate:
ts_auto_arima(),
ts_auto_arima_xgboost(),
ts_auto_croston(),
ts_auto_exp_smoothing(),
ts_auto_glmnet(),
ts_auto_lm(),
ts_auto_mars(),
ts_auto_prophet_boost(),
ts_auto_prophet_reg(),
ts_auto_smooth_es(),
ts_auto_svm_poly(),
ts_auto_svm_rbf(),
ts_auto_theta(),
ts_auto_xgboost()
library(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_nnetar <- ts_auto_nnetar( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 5, .tune = FALSE ) ts_nnetar$recipe_infolibrary(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_nnetar <- ts_auto_nnetar( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 5, .tune = FALSE ) ts_nnetar$recipe_info
This is a boilerplate function to create automatically the following:
recipe
model specification
workflow
tuned model (grid ect)
calibration tibble and plot
ts_auto_prophet_boost( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_prophet_boost", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )ts_auto_prophet_boost( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_prophet_boost", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )
.data |
The data being passed to the function. The time-series object. |
.date_col |
The column that holds the datetime. |
.value_col |
The column that has the value |
.formula |
The formula that is passed to the recipe like |
.rsamp_obj |
The rsample splits object |
.prefix |
Default is |
.tune |
Defaults to TRUE, this creates a tuning grid and tuned model. |
.grid_size |
If |
.num_cores |
How many cores do you want to use. Default is 1 |
.cv_assess |
How many observations for assess. See |
.cv_skip |
How many observations to skip. See |
.cv_slice_limit |
How many slices to return. See |
.best_metric |
Default is "rmse". See |
.bootstrap_final |
Not yet implemented. |
This uses the modeltime::prophet_boost() function with the engine
set to prophet_xgboost.
A list
Steven P. Sanderson II, MPH
https://business-science.github.io/modeltime/reference/prophet_boost.html
Other Boiler_Plate:
ts_auto_arima(),
ts_auto_arima_xgboost(),
ts_auto_croston(),
ts_auto_exp_smoothing(),
ts_auto_glmnet(),
ts_auto_lm(),
ts_auto_mars(),
ts_auto_nnetar(),
ts_auto_prophet_reg(),
ts_auto_smooth_es(),
ts_auto_svm_poly(),
ts_auto_svm_rbf(),
ts_auto_theta(),
ts_auto_xgboost()
Other prophet:
ts_auto_prophet_reg()
library(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_prophet_boost <- ts_auto_prophet_boost( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 5, .tune = FALSE ) ts_prophet_boost$recipe_infolibrary(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_prophet_boost <- ts_auto_prophet_boost( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 5, .tune = FALSE ) ts_prophet_boost$recipe_info
This is a boilerplate function to create automatically the following:
recipe
model specification
workflow
tuned model (grid ect)
calibration tibble and plot
ts_auto_prophet_reg( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_prophet_reg", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )ts_auto_prophet_reg( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_prophet_reg", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )
.data |
The data being passed to the function. The time-series object. |
.date_col |
The column that holds the datetime. |
.value_col |
The column that has the value |
.formula |
The formula that is passed to the recipe like |
.rsamp_obj |
The rsample splits object |
.prefix |
Default is |
.tune |
Defaults to TRUE, this creates a tuning grid and tuned model. |
.grid_size |
If |
.num_cores |
How many cores do you want to use. Default is 1 |
.cv_assess |
How many observations for assess. See |
.cv_skip |
How many observations to skip. See |
.cv_slice_limit |
How many slices to return. See |
.best_metric |
Default is "rmse". See |
.bootstrap_final |
Not yet implemented. |
This uses the modeltime::prophet_reg() function with the engine set to prophet.
A list
Steven P. Sanderson II, MPH
https://business-science.github.io/modeltime/reference/prophet_reg.html
Other Boiler_Plate:
ts_auto_arima(),
ts_auto_arima_xgboost(),
ts_auto_croston(),
ts_auto_exp_smoothing(),
ts_auto_glmnet(),
ts_auto_lm(),
ts_auto_mars(),
ts_auto_nnetar(),
ts_auto_prophet_boost(),
ts_auto_smooth_es(),
ts_auto_svm_poly(),
ts_auto_svm_rbf(),
ts_auto_theta(),
ts_auto_xgboost()
Other prophet:
ts_auto_prophet_boost()
library(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_prophet_reg <- ts_auto_prophet_reg( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 5, .tune = FALSE ) ts_prophet_reg$recipe_infolibrary(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_prophet_reg <- ts_auto_prophet_reg( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 5, .tune = FALSE ) ts_prophet_reg$recipe_info
Automatically builds generic time series recipe objects from a given tibble.
ts_auto_recipe( .data, .date_col, .pred_col, .step_ts_sig = TRUE, .step_ts_rm_misc = TRUE, .step_ts_dummy = TRUE, .step_ts_fourier = TRUE, .step_ts_fourier_period = 365/12, .K = 1, .step_ts_yeo = TRUE, .step_ts_nzv = TRUE )ts_auto_recipe( .data, .date_col, .pred_col, .step_ts_sig = TRUE, .step_ts_rm_misc = TRUE, .step_ts_dummy = TRUE, .step_ts_fourier = TRUE, .step_ts_fourier_period = 365/12, .K = 1, .step_ts_yeo = TRUE, .step_ts_nzv = TRUE )
.data |
The data that is going to be modeled. You must supply a tibble. |
.date_col |
The column that holds the date for the time series. |
.pred_col |
The column that is to be predicted. |
.step_ts_sig |
A Boolean indicating should the |
.step_ts_rm_misc |
A Boolean indicating should the following items be removed from the time series signature, default is TRUE.
|
.step_ts_dummy |
A Boolean indicating if all_nominal_predictors() should be dummied and with one hot encoding. |
.step_ts_fourier |
A Boolean indicating if |
.step_ts_fourier_period |
A number such as 365/12, 365/4 or 365 indicting the period of the fourier term. The numeric period for the oscillation frequency. |
.K |
The number of orders to include for each sine/cosine fourier series. More orders increase the number of fourier terms and therefore the variance of the fitted model at the expense of bias. See details for examples of K specification. |
.step_ts_yeo |
A Boolean indicating if the |
.step_ts_nzv |
A Boolean indicating if the |
This will build out a couple of generic recipe objects and return those items in a list.
Steven P. Sanderson II, MPH
suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) splits <- initial_time_split( data_tbl , prop = 0.8 ) ts_auto_recipe( .data = data_tbl , .date_col = date_col , .pred_col = value ) ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value )suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) splits <- initial_time_split( data_tbl , prop = 0.8 ) ts_auto_recipe( .data = data_tbl , .date_col = date_col , .pred_col = value ) ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value )
This is a boilerplate function to automatically create the following:
recipe
model specification
workflow
tuned model (grid ect)
calibration tibble and plot
ts_auto_smooth_es( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_smooth_es", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )ts_auto_smooth_es( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_smooth_es", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )
.data |
The data being passed to the function. The time-series object. |
.date_col |
The column that holds the datetime. |
.value_col |
The column that has the value |
.formula |
The formula that is passed to the recipe like |
.rsamp_obj |
The rsample splits object |
.prefix |
Default is |
.tune |
Defaults to TRUE, this creates a tuning grid and tuned model. |
.grid_size |
If |
.num_cores |
How many cores do you want to use. Default is 1 |
.cv_assess |
How many observations for assess. See |
.cv_skip |
How many observations to skip. See |
.cv_slice_limit |
How many slices to return. See |
.best_metric |
Default is "rmse". See |
.bootstrap_final |
Not yet implemented. |
This uses modeltime::exp_smoothing() and sets the parsnip::engine
to smooth_es.
A list
Steven P. Sanderson II, MPH
https://business-science.github.io/modeltime/reference/exp_smoothing.html#ref-examples
https://github.com/config-i1/smooth
Other Boiler_Plate:
ts_auto_arima(),
ts_auto_arima_xgboost(),
ts_auto_croston(),
ts_auto_exp_smoothing(),
ts_auto_glmnet(),
ts_auto_lm(),
ts_auto_mars(),
ts_auto_nnetar(),
ts_auto_prophet_boost(),
ts_auto_prophet_reg(),
ts_auto_svm_poly(),
ts_auto_svm_rbf(),
ts_auto_theta(),
ts_auto_xgboost()
Other exp_smoothing:
ts_auto_croston(),
ts_auto_exp_smoothing(),
ts_auto_theta()
library(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_smooth_es <- ts_auto_smooth_es( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 3, .tune = FALSE ) ts_smooth_es$recipe_infolibrary(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_smooth_es <- ts_auto_smooth_es( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 3, .tune = FALSE ) ts_smooth_es$recipe_info
This is a boilerplate function to automatically create the following:
recipe
model specification
workflow
tuned model (grid ect)
calibration tibble and plot
ts_auto_svm_poly( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_svm_poly", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )ts_auto_svm_poly( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_svm_poly", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )
.data |
The data being passed to the function. The time-series object. |
.date_col |
The column that holds the datetime. |
.value_col |
The column that has the value |
.formula |
The formula that is passed to the recipe like |
.rsamp_obj |
The rsample splits object |
.prefix |
Default is |
.tune |
Defaults to TRUE, this creates a tuning grid and tuned model. |
.grid_size |
If |
.num_cores |
How many cores do you want to use. Default is 1 |
.cv_assess |
How many observations for assess. See |
.cv_skip |
How many observations to skip. See |
.cv_slice_limit |
How many slices to return. See |
.best_metric |
Default is "rmse". See |
.bootstrap_final |
Not yet implemented. |
This uses parsnip::svm_poly() and sets the parsnip::engine
to kernlab.
A list
Steven P. Sanderson II, MPH
https://parsnip.tidymodels.org/reference/svm_poly.html
Other Boiler_Plate:
ts_auto_arima(),
ts_auto_arima_xgboost(),
ts_auto_croston(),
ts_auto_exp_smoothing(),
ts_auto_glmnet(),
ts_auto_lm(),
ts_auto_mars(),
ts_auto_nnetar(),
ts_auto_prophet_boost(),
ts_auto_prophet_reg(),
ts_auto_smooth_es(),
ts_auto_svm_rbf(),
ts_auto_theta(),
ts_auto_xgboost()
Other SVM:
ts_auto_svm_rbf()
library(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_auto_poly <- ts_auto_svm_poly( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 3, .tune = FALSE ) ts_auto_poly$recipe_infolibrary(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_auto_poly <- ts_auto_svm_poly( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 3, .tune = FALSE ) ts_auto_poly$recipe_info
This is a boilerplate function to automatically create the following:
recipe
model specification
workflow
tuned model (grid ect)
calibration tibble and plot
ts_auto_svm_rbf( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_svm_rbf", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )ts_auto_svm_rbf( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_svm_rbf", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )
.data |
The data being passed to the function. The time-series object. |
.date_col |
The column that holds the datetime. |
.value_col |
The column that has the value |
.formula |
The formula that is passed to the recipe like |
.rsamp_obj |
The rsample splits object |
.prefix |
Default is |
.tune |
Defaults to TRUE, this creates a tuning grid and tuned model. |
.grid_size |
If |
.num_cores |
How many cores do you want to use. Default is 1 |
.cv_assess |
How many observations for assess. See |
.cv_skip |
How many observations to skip. See |
.cv_slice_limit |
How many slices to return. See |
.best_metric |
Default is "rmse". See |
.bootstrap_final |
Not yet implemented. |
This uses parsnip::svm_rb() and sets the parsnip::engine
to kernlab.
A list
Steven P. Sanderson II, MPH
https://parsnip.tidymodels.org/reference/svm_rbf.html
Other Boiler_Plate:
ts_auto_arima(),
ts_auto_arima_xgboost(),
ts_auto_croston(),
ts_auto_exp_smoothing(),
ts_auto_glmnet(),
ts_auto_lm(),
ts_auto_mars(),
ts_auto_nnetar(),
ts_auto_prophet_boost(),
ts_auto_prophet_reg(),
ts_auto_smooth_es(),
ts_auto_svm_poly(),
ts_auto_theta(),
ts_auto_xgboost()
Other SVM:
ts_auto_svm_poly()
library(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_auto_rbf <- ts_auto_svm_rbf( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 3, .tune = FALSE ) ts_auto_rbf$recipe_infolibrary(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_auto_rbf <- ts_auto_svm_rbf( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 3, .tune = FALSE ) ts_auto_rbf$recipe_info
This is a boilerplate function to create automatically the following:
recipe
model specification
workflow
calibration tibble and plot
ts_auto_theta( .data, .date_col, .value_col, .rsamp_obj, .prefix = "ts_theta", .bootstrap_final = FALSE )ts_auto_theta( .data, .date_col, .value_col, .rsamp_obj, .prefix = "ts_theta", .bootstrap_final = FALSE )
.data |
The data being passed to the function. The time-series object. |
.date_col |
The column that holds the datetime. |
.value_col |
The column that has the value |
.rsamp_obj |
The splits object |
.prefix |
Default is |
.bootstrap_final |
Not yet implemented. |
This uses the forecast::thetaf() for the parsnip engine. This
model does not use exogenous regressors, so only a univariate model of: value ~ date
will be used from the .date_col and .value_col that you provide.
A list
Steven P. Sanderson II, MPH
https://business-science.github.io/modeltime/reference/exp_smoothing.html#engine-details
https://pkg.robjhyndman.com/forecast/reference/thetaf.html
Other Boiler_Plate:
ts_auto_arima(),
ts_auto_arima_xgboost(),
ts_auto_croston(),
ts_auto_exp_smoothing(),
ts_auto_glmnet(),
ts_auto_lm(),
ts_auto_mars(),
ts_auto_nnetar(),
ts_auto_prophet_boost(),
ts_auto_prophet_reg(),
ts_auto_smooth_es(),
ts_auto_svm_poly(),
ts_auto_svm_rbf(),
ts_auto_xgboost()
Other exp_smoothing:
ts_auto_croston(),
ts_auto_exp_smoothing(),
ts_auto_smooth_es()
library(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_theta <- ts_auto_theta( .data = data, .date_col = date_col, .value_col = value, .rsamp_obj = splits ) ts_theta$recipe_infolibrary(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_theta <- ts_auto_theta( .data = data, .date_col = date_col, .value_col = value, .rsamp_obj = splits ) ts_theta$recipe_info
This is a boilerplate function to create automatically the following:
recipe
model specification
workflow
tuned model (grid ect)
calibration tibble and plot
ts_auto_xgboost( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_xgboost", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )ts_auto_xgboost( .data, .date_col, .value_col, .formula, .rsamp_obj, .prefix = "ts_xgboost", .tune = TRUE, .grid_size = 10, .num_cores = 1, .cv_assess = 12, .cv_skip = 3, .cv_slice_limit = 6, .best_metric = "rmse", .bootstrap_final = FALSE )
.data |
The data being passed to the function. The time-series object. |
.date_col |
The column that holds the datetime. |
.value_col |
The column that has the value |
.formula |
The formula that is passed to the recipe like |
.rsamp_obj |
The rsample splits object |
.prefix |
Default is |
.tune |
Defaults to TRUE, this creates a tuning grid and tuned model. |
.grid_size |
If |
.num_cores |
How many cores do you want to use. Default is 1 |
.cv_assess |
How many observations for assess. See |
.cv_skip |
How many observations to skip. See |
.cv_slice_limit |
How many slices to return. See |
.best_metric |
Default is "rmse". See |
.bootstrap_final |
Not yet implemented. |
This uses the parsnip::boost_tree() with the engine set to xgboost
A list
Steven P. Sanderson II, MPH
Other Boiler_Plate:
ts_auto_arima(),
ts_auto_arima_xgboost(),
ts_auto_croston(),
ts_auto_exp_smoothing(),
ts_auto_glmnet(),
ts_auto_lm(),
ts_auto_mars(),
ts_auto_nnetar(),
ts_auto_prophet_boost(),
ts_auto_prophet_reg(),
ts_auto_smooth_es(),
ts_auto_svm_poly(),
ts_auto_svm_rbf(),
ts_auto_theta()
library(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_xgboost <- ts_auto_xgboost( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 5, .tune = FALSE ) ts_xgboost$recipe_infolibrary(dplyr) library(timetk) library(modeltime) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_xgboost <- ts_auto_xgboost( .data = data, .num_cores = 2, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., .grid_size = 5, .tune = FALSE ) ts_xgboost$recipe_info
Create a Brownian Motion Tibble
ts_brownian_motion( .time = 100, .num_sims = 10, .delta_time = 1, .initial_value = 0, .return_tibble = TRUE )ts_brownian_motion( .time = 100, .num_sims = 10, .delta_time = 1, .initial_value = 0, .return_tibble = TRUE )
.time |
Total time of the simulation. |
.num_sims |
Total number of simulations. |
.delta_time |
Time step size. |
.initial_value |
Integer representing the initial value. |
.return_tibble |
The default is TRUE. If set to FALSE then an object of class matrix will be returned. |
Brownian Motion, also known as the Wiener process, is a continuous-time random process that describes the random movement of particles suspended in a fluid. It is named after the physicist Robert Brown, who first described the phenomenon in 1827.
The equation for Brownian Motion can be represented as:
W(t) = W(0) + sqrt(t) * Z
Where W(t) is the Brownian motion at time t, W(0) is the initial value of the Brownian motion, sqrt(t) is the square root of time, and Z is a standard normal random variable.
Brownian Motion has numerous applications, including modeling stock prices in financial markets, modeling particle movement in fluids, and modeling random walk processes in general. It is a useful tool in probability theory and statistical analysis.
A tibble/matrix
Steven P. Sanderson II, MPH
Other Data Generator:
tidy_fft(),
ts_brownian_motion_augment(),
ts_geometric_brownian_motion(),
ts_geometric_brownian_motion_augment(),
ts_random_walk()
ts_brownian_motion()ts_brownian_motion()
Create a Brownian Motion Tibble
ts_brownian_motion_augment( .data, .date_col, .value_col, .time = 100, .num_sims = 10, .delta_time = NULL )ts_brownian_motion_augment( .data, .date_col, .value_col, .time = 100, .num_sims = 10, .delta_time = NULL )
.data |
The data.frame/tibble being augmented. |
.date_col |
The column that holds the date. |
.value_col |
The value that is going to get augmented. The last value of this column becomes the initial value internally. |
.time |
How many time steps ahead. |
.num_sims |
How many simulations should be run. |
.delta_time |
Time step size. |
Brownian Motion, also known as the Wiener process, is a continuous-time random process that describes the random movement of particles suspended in a fluid. It is named after the physicist Robert Brown, who first described the phenomenon in 1827.
The equation for Brownian Motion can be represented as:
W(t) = W(0) + sqrt(t) * Z
Where W(t) is the Brownian motion at time t, W(0) is the initial value of the Brownian motion, sqrt(t) is the square root of time, and Z is a standard normal random variable.
Brownian Motion has numerous applications, including modeling stock prices in financial markets, modeling particle movement in fluids, and modeling random walk processes in general. It is a useful tool in probability theory and statistical analysis.
A tibble/matrix
Steven P. Sanderson II, MPH
Other Data Generator:
tidy_fft(),
ts_brownian_motion(),
ts_geometric_brownian_motion(),
ts_geometric_brownian_motion_augment(),
ts_random_walk()
rn <- rnorm(31) df <- data.frame( date_col = seq.Date(from = as.Date("2022-01-01"), to = as.Date("2022-01-31"), by = "day"), value = rn ) ts_brownian_motion_augment( .data = df, .date_col = date_col, .value_col = value )rn <- rnorm(31) df <- data.frame( date_col = seq.Date(from = as.Date("2022-01-01"), to = as.Date("2022-01-31"), by = "day"), value = rn ) ts_brownian_motion_augment( .data = df, .date_col = date_col, .value_col = value )
Plot an augmented Geometric/Brownian Motion.
ts_brownian_motion_plot(.data, .date_col, .value_col, .interactive = FALSE)ts_brownian_motion_plot(.data, .date_col, .value_col, .interactive = FALSE)
.data |
The data you are going to pass to the function to augment. |
.date_col |
The column that holds the date |
.value_col |
The column that holds the value |
.interactive |
The default is FALSE, TRUE will produce an interactive plotly plot. |
This function will take output from either the ts_brownian_motion_augment()
or the ts_geometric_brownian_motion_augment() function and plot them. The
legend is set to "none" if the simulation count is higher than 9.
A ggplot2 object or an interactive plotly plot
Steven P. Sanderson II, MPH
Other Plot:
ts_event_analysis_plot(),
ts_qq_plot(),
ts_scedacity_scatter_plot()
library(dplyr) df <- ts_to_tbl(AirPassengers) %>% select(-index) augmented_data <- df %>% ts_brownian_motion_augment( .date_col = date_col, .value_col = value, .time = 144 ) augmented_data %>% ts_brownian_motion_plot(.date_col = date_col, .value_col = value)library(dplyr) df <- ts_to_tbl(AirPassengers) %>% select(-index) augmented_data <- df %>% ts_brownian_motion_augment( .date_col = date_col, .value_col = value, .time = 144 ) augmented_data %>% ts_brownian_motion_plot(.date_col = date_col, .value_col = value)
Takes in data that has been aggregated to the day level and makes a calendar heatmap.
ts_calendar_heatmap_plot( .data, .date_col, .value_col, .low = "red", .high = "green", .plt_title = "", .interactive = TRUE )ts_calendar_heatmap_plot( .data, .date_col, .value_col, .low = "red", .high = "green", .plt_title = "", .interactive = TRUE )
.data |
The time-series data with a date column and value column. |
.date_col |
The column that has the datetime values |
.value_col |
The column that has the values |
.low |
The color for the low value, must be quoted like "red". The default is "red" |
.high |
The color for the high value, must be quoted like "green". The default is "green" |
.plt_title |
The title of the plot |
.interactive |
Default is TRUE to get an interactive plot using |
The data provided must have been aggregated to the day level, if not funky output could result and it is possible nothing will be output but errors. There must be a date column and a value column, those are the only items required for this function to work.
This function is intentionally inflexible, it complains more and does less in order to force the user to supply a clean data-set.
A ggplot2 plot or if interactive a plotly plot
Steven P. Sanderson II, MPH
data_tbl <- data.frame( date_col = seq.Date( from = as.Date("2020-01-01"), to = as.Date("2022-06-01"), length.out = 365*2 + 180 ), value = rnorm(365*2+180, mean = 100) ) ts_calendar_heatmap_plot( .data = data_tbl , .date_col = date_col , .value_col = value , .interactive = FALSE )data_tbl <- data.frame( date_col = seq.Date( from = as.Date("2020-01-01"), to = as.Date("2022-06-01"), length.out = 365*2 + 180 ), value = rnorm(365*2+180, mean = 100) ) ts_calendar_heatmap_plot( .data = data_tbl , .date_col = date_col , .value_col = value , .interactive = FALSE )
Given a tibble/data.frame, you can get date from two different but comparative date ranges. Lets say you want to compare visits in one year to visits from 2 years before without also seeing the previous 1 year. You can do that with this function.
ts_compare_data(.data, .date_col, .start_date, .end_date, .periods_back)ts_compare_data(.data, .date_col, .start_date, .end_date, .periods_back)
.data |
The date.frame/tibble that holds the data |
.date_col |
The column with the date value |
.start_date |
The start of the period you want to analyze |
.end_date |
The end of the period you want to analyze |
.periods_back |
How long ago do you want to compare data too. Time units
are collapsed using
Arbitrary unique English abbreviations as in the |
Uses the timetk::filter_by_time() function in order to filter the date
column.
Uses the timetk::subtract_time() function to subtract time from the start date.
A tibble.
Steven P. Sanderson II, MPH
Other Time_Filtering:
ts_time_event_analysis_tbl()
suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(timetk)) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) ts_compare_data( .data = data_tbl , .date_col = date_col , .start_date = "1955-01-01" , .end_date = "1955-12-31" , .periods_back = "2 years" ) %>% summarise_by_time( .date_var = date_col , .by = "year" , visits = sum(value) )suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(timetk)) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) ts_compare_data( .data = data_tbl , .date_col = date_col , .start_date = "1955-01-01" , .end_date = "1955-12-31" , .periods_back = "2 years" ) %>% summarise_by_time( .date_var = date_col , .by = "year" , visits = sum(value) )
Plot out the data from the ts_time_event_analysis_tbl() function.
ts_event_analysis_plot( .data, .plot_type = "mean", .plot_ci = TRUE, .interactive = FALSE )ts_event_analysis_plot( .data, .plot_type = "mean", .plot_ci = TRUE, .interactive = FALSE )
.data |
The data that comes from the |
.plot_type |
The default is "mean" which will show the mean event change of the output from the analysis tibble. The possible values for this are: mean, median, and individual. |
.plot_ci |
The default is TRUE. This will only work if you choose one of the aggregate plots of either "mean" or "median" |
.interactive |
The default is FALSE. TRUE will return a plotly plot. |
This function will take in data strictly from the ts_time_event_analysis_tbl()
and plot out the data. You can choose what type of plot you want in the parameter
of .plot_type. This will give you a choice of "mean", "median", and "individual".
You can also plot the upper and lower confidence intervals if you choose one of the aggregate plots ("mean"/"median").
A ggplot2 object
Steven P. Sanderson II, MPH
Other Plot:
ts_brownian_motion_plot(),
ts_qq_plot(),
ts_scedacity_scatter_plot()
library(dplyr) df <- ts_to_tbl(AirPassengers) %>% select(-index) ts_time_event_analysis_tbl( .data = df, .horizon = 6, .date_col = date_col, .value_col = value, .direction = "both" ) %>% ts_event_analysis_plot() ts_time_event_analysis_tbl( .data = df, .horizon = 6, .date_col = date_col, .value_col = value, .direction = "both" ) %>% ts_event_analysis_plot(.plot_type = "individual")library(dplyr) df <- ts_to_tbl(AirPassengers) %>% select(-index) ts_time_event_analysis_tbl( .data = df, .horizon = 6, .date_col = date_col, .value_col = value, .direction = "both" ) %>% ts_event_analysis_plot() ts_time_event_analysis_tbl( .data = df, .horizon = 6, .date_col = date_col, .value_col = value, .direction = "both" ) %>% ts_event_analysis_plot(.plot_type = "individual")
Extract the fitted workflow from a ts_auto_ function.
ts_extract_auto_fitted_workflow(.input)ts_extract_auto_fitted_workflow(.input)
.input |
This is the output list object of a |
Extract the fitted workflow from a ts_auto_ function. This will
only work on those functions that are designated as Boilerplate.
A fitted workflow object.
Steven P. Sanderson II, MPH
## Not run: library(dplyr) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_lm <- ts_auto_lm( .data = data, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., ) ts_extract_auto_fitted_workflow(ts_lm) ## End(Not run)## Not run: library(dplyr) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_lm <- ts_auto_lm( .data = data, .date_col = date_col, .value_col = value, .rsamp_obj = splits, .formula = value ~ ., ) ts_extract_auto_fitted_workflow(ts_lm) ## End(Not run)
This function returns an output list of data and plots that
come from using the K-Means clustering algorithm on a time series data.
ts_feature_cluster( .data, .date_col, .value_col, ..., .features = c("frequency", "entropy", "acf_features"), .scale = TRUE, .prefix = "ts_", .centers = 3 )ts_feature_cluster( .data, .date_col, .value_col, ..., .features = c("frequency", "entropy", "acf_features"), .scale = TRUE, .prefix = "ts_", .centers = 3 )
.data |
The data passed must be a |
.date_col |
The date column. |
.value_col |
The column that holds the value of the time series where you want the features and clustering performed on. |
... |
This is where you can place grouping variables that are passed off
to |
.features |
This is a quoted string vector using c() of features that you
would like to pass. You can pass any feature you make or those from the |
.scale |
If TRUE, time series are scaled to mean 0 and sd 1 before features are computed |
.prefix |
A prefix to prefix the feature columns. Default: "ts_" |
.centers |
An integer of how many different centers you would like to generate. The default is 3. |
This function will return a list object output. The function itself
requires that a time series tibble/data.frame get passed to it, along with
the .date_col, the .value_col and a period of data. It uses the underlying
function timetk::tk_tsfeatures() and takes the output of that and performs
a clustering analysis using the K-Means algorithm.
The function has a parameter of .features which can take any of the features
listed in the tsfeatures package by Rob Hyndman. You can also create custom
functions in the .GlobalEnviron and it will take them as quoted arguments.
So you can make a function as follows
my_mean <- function(x){return(mean(x, na.rm = TRUE))}
You can then call this by using .features = c("my_mean").
The output of this function includes the following:
Data Section
ts_feature_tbl
user_item_matrix_tbl
mapped_tbl
scree_data_tbl
input_data_tbl (the original data)
Plots
static_plot
plotly_plot
A list output
Steven P. Sanderson II, MPH
https://pkg.robjhyndman.com/tsfeatures/index.html
Other Clustering:
ts_feature_cluster_plot()
library(dplyr) data_tbl <- ts_to_tbl(AirPassengers) %>% mutate(group_id = rep(1:12, 12)) ts_feature_cluster( .data = data_tbl, .date_col = date_col, .value_col = value, group_id, .features = c("acf_features","entropy"), .scale = TRUE, .prefix = "ts_", .centers = 3 )library(dplyr) data_tbl <- ts_to_tbl(AirPassengers) %>% mutate(group_id = rep(1:12, 12)) ts_feature_cluster( .data = data_tbl, .date_col = date_col, .value_col = value, group_id, .features = c("acf_features","entropy"), .scale = TRUE, .prefix = "ts_", .centers = 3 )
This function returns an output list of data and plots that
come from using the K-Means clustering algorithm on a time series data.
ts_feature_cluster_plot( .data, .date_col, .value_col, ..., .center = 3, .facet_ncol = 3, .smooth = FALSE )ts_feature_cluster_plot( .data, .date_col, .value_col, ..., .center = 3, .facet_ncol = 3, .smooth = FALSE )
.data |
The data passed must be the output of the |
.date_col |
The date column. |
.value_col |
The column that holds the value of the time series that the featurs were built from. |
... |
This is where you can place grouping variables that are passed off
to |
.center |
An integer of the chosen amount of centers from the |
.facet_ncol |
This is passed to the |
.smooth |
This is passed to the |
This function will return a list object output. The function itself
requires that the ts_feature_cluster() be passed to it as it will look for
a specific attribute internally.
The output of this function includes the following:
Data Section
original_data
kmm_data_tbl
user_item_tbl
cluster_tbl
Plots
static_plot
plotly_plot
K-Means Object
k-means object
A list output
Steven P. Sanderson II, MPH
Other Clustering:
ts_feature_cluster()
library(dplyr) data_tbl <- ts_to_tbl(AirPassengers) %>% mutate(group_id = rep(1:12, 12)) output <- ts_feature_cluster( .data = data_tbl, .date_col = date_col, .value_col = value, group_id, .features = c("acf_features","entropy"), .scale = TRUE, .prefix = "ts_", .centers = 3 ) ts_feature_cluster_plot( .data = output, .date_col = date_col, .value_col = value, .center = 2, group_id )library(dplyr) data_tbl <- ts_to_tbl(AirPassengers) %>% mutate(group_id = rep(1:12, 12)) output <- ts_feature_cluster( .data = data_tbl, .date_col = date_col, .value_col = value, group_id, .features = c("acf_features","entropy"), .scale = TRUE, .prefix = "ts_", .centers = 3 ) ts_feature_cluster_plot( .data = output, .date_col = date_col, .value_col = value, .center = 2, group_id )
Creating different forecast paths for forecast objects (when applicable),
by utilizing the underlying model distribution with the simulate function.
ts_forecast_simulator( .model, .data, .ext_reg = NULL, .frequency = NULL, .bootstrap = TRUE, .horizon = 4, .iterations = 25, .sim_color = "steelblue", .alpha = 0.05 )ts_forecast_simulator( .model, .data, .ext_reg = NULL, .frequency = NULL, .bootstrap = TRUE, .horizon = 4, .iterations = 25, .sim_color = "steelblue", .alpha = 0.05 )
.model |
A forecasting model of one of the following from the
|
.data |
The data that is used for the |
.ext_reg |
A |
.frequency |
This is for the conversion of an internal table and should match the time frequency of the data. |
.bootstrap |
A boolean value of TRUE/FALSE. From |
.horizon |
An integer defining the forecast horizon. |
.iterations |
An integer, set the number of iterations of the simulation. |
.sim_color |
Set the color of the simulation paths lines. |
.alpha |
Set the opacity level of the simulation path lines. |
This function expects to take in a model of either Arima,
auto.arima, ets or nnetar from the forecast package. You can supply a
forecasting horizon, iterations and a few other items. You may also specify
an Arima() model using xregs.
The original time series, the simulated values and a some plots
Steven P. Sanderson II, MPH
Other Simulator:
ts_arima_simulator()
suppressPackageStartupMessages(library(forecast)) suppressPackageStartupMessages(library(dplyr)) # Create a model fit <- auto.arima(AirPassengers) data_tbl <- ts_to_tbl(AirPassengers) # Simulate 50 possible forecast paths, with .horizon of 12 months output <- ts_forecast_simulator( .model = fit , .horizon = 12 , .iterations = 50 , .data = data_tbl ) output$ggplotsuppressPackageStartupMessages(library(forecast)) suppressPackageStartupMessages(library(dplyr)) # Create a model fit <- auto.arima(AirPassengers) data_tbl <- ts_to_tbl(AirPassengers) # Simulate 50 possible forecast paths, with .horizon of 12 months output <- ts_forecast_simulator( .model = fit , .horizon = 12 , .iterations = 50 , .data = data_tbl ) output$ggplot
Create a Geometric Brownian Motion.
ts_geometric_brownian_motion( .num_sims = 100, .time = 25, .mean = 0, .sigma = 0.1, .initial_value = 100, .delta_time = 1/365, .return_tibble = TRUE )ts_geometric_brownian_motion( .num_sims = 100, .time = 25, .mean = 0, .sigma = 0.1, .initial_value = 100, .delta_time = 1/365, .return_tibble = TRUE )
.num_sims |
Total number of simulations. |
.time |
Total time of the simulation. |
.mean |
Expected return |
.sigma |
Volatility |
.initial_value |
Integer representing the initial value. |
.delta_time |
Time step size. |
.return_tibble |
The default is TRUE. If set to FALSE then an object of class matrix will be returned. |
Geometric Brownian Motion (GBM) is a statistical method for modeling the evolution of a given financial asset over time. It is a type of stochastic process, which means that it is a system that undergoes random changes over time.
GBM is widely used in the field of finance to model the behavior of stock prices, foreign exchange rates, and other financial assets. It is based on the assumption that the asset's price follows a random walk, meaning that it is influenced by a number of unpredictable factors such as market trends, news events, and investor sentiment.
The equation for GBM is:
dS/S = mdt + sdW
where S is the price of the asset, t is time, m is the expected return on the asset, s is the volatility of the asset, and dW is a small random change in the asset's price.
GBM can be used to estimate the likelihood of different outcomes for a given asset, and it is often used in conjunction with other statistical methods to make more accurate predictions about the future performance of an asset.
This function provides the ability of simulating and estimating the parameters of a GBM process. It can be used to analyze the behavior of financial assets and to make informed investment decisions.
A tibble/matrix
Steven P. Sanderson II, MPH
Other Data Generator:
tidy_fft(),
ts_brownian_motion(),
ts_brownian_motion_augment(),
ts_geometric_brownian_motion_augment(),
ts_random_walk()
ts_geometric_brownian_motion()ts_geometric_brownian_motion()
Create a Geometric Brownian Motion.
ts_geometric_brownian_motion_augment( .data, .date_col, .value_col, .num_sims = 10, .time = 25, .mean = 0, .sigma = 0.1, .delta_time = 1/365 )ts_geometric_brownian_motion_augment( .data, .date_col, .value_col, .num_sims = 10, .time = 25, .mean = 0, .sigma = 0.1, .delta_time = 1/365 )
.data |
The data you are going to pass to the function to augment. |
.date_col |
The column that holds the date |
.value_col |
The column that holds the value |
.num_sims |
Total number of simulations. |
.time |
Total time of the simulation. |
.mean |
Expected return |
.sigma |
Volatility |
.delta_time |
Time step size. |
Geometric Brownian Motion (GBM) is a statistical method for modeling the evolution of a given financial asset over time. It is a type of stochastic process, which means that it is a system that undergoes random changes over time.
GBM is widely used in the field of finance to model the behavior of stock prices, foreign exchange rates, and other financial assets. It is based on the assumption that the asset's price follows a random walk, meaning that it is influenced by a number of unpredictable factors such as market trends, news events, and investor sentiment.
The equation for GBM is:
dS/S = mdt + sdW
where S is the price of the asset, t is time, m is the expected return on the asset, s is the volatility of the asset, and dW is a small random change in the asset's price.
GBM can be used to estimate the likelihood of different outcomes for a given asset, and it is often used in conjunction with other statistical methods to make more accurate predictions about the future performance of an asset.
This function provides the ability of simulating and estimating the parameters of a GBM process. It can be used to analyze the behavior of financial assets and to make informed investment decisions.
A tibble/matrix
Steven P. Sanderson II, MPH
Other Data Generator:
tidy_fft(),
ts_brownian_motion(),
ts_brownian_motion_augment(),
ts_geometric_brownian_motion(),
ts_random_walk()
rn <- rnorm(31) df <- data.frame( date_col = seq.Date(from = as.Date("2022-01-01"), to = as.Date("2022-01-31"), by = "day"), value = rn ) ts_geometric_brownian_motion_augment( .data = df, .date_col = date_col, .value_col = value )rn <- rnorm(31) df <- data.frame( date_col = seq.Date(from = as.Date("2022-01-01"), to = as.Date("2022-01-31"), by = "day"), value = rn ) ts_geometric_brownian_motion_augment( .data = df, .date_col = date_col, .value_col = value )
Get date or datetime variables (column names)
ts_get_date_columns(.data)ts_get_date_columns(.data)
.data |
An object of class |
ts_get_date_columns returns the column names of date or datetime variables
in a data frame.
A vector containing the column names that are of date/date-like classes.
Steven P. Sanderson II, MPH
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
ts_to_tbl(AirPassengers) %>% ts_get_date_columns()ts_to_tbl(AirPassengers) %>% ts_get_date_columns()
This function is used to augment a data frame or tibble with time series
growth rates of selected columns. You can provide a data frame or tibble as
the first argument, the column(s) for which you want to calculate the growth
rates using the .value parameter, and optionally specify custom names for
the new columns using the .names parameter.
ts_growth_rate_augment(.data, .value, .names = "auto")ts_growth_rate_augment(.data, .value, .names = "auto")
.data |
A data frame or tibble containing the data to be augmented. |
.value |
A quosure specifying the column(s) for which you want to calculate growth rates. |
.names |
Optional. A character vector specifying the names of the new columns to be created. Use "auto" for automatic naming. |
A tibble that includes the original data and additional columns representing
the growth rates of the selected columns. The column names are either
automatically generated or as specified in the .names parameter.
Steven P. Sanderson II, MPH
Other Augment Function:
ts_acceleration_augment(),
ts_velocity_augment()
data <- data.frame( Year = 1:5, Income = c(100, 120, 150, 180, 200), Expenses = c(50, 60, 75, 90, 100) ) ts_growth_rate_augment(data, .value = c(Income, Expenses))data <- data.frame( Year = 1:5, Income = c(100, 120, 150, 180, 200), Expenses = c(50, 60, 75, 90, 100) ) ts_growth_rate_augment(data, .value = c(Income, Expenses))
This function computes the growth rate of a numeric vector, typically representing a time series, with optional transformations like scaling, power, and lag differences.
ts_growth_rate_vec(.x, .scale = 100, .power = 1, .log_diff = FALSE, .lags = 1)ts_growth_rate_vec(.x, .scale = 100, .power = 1, .log_diff = FALSE, .lags = 1)
.x |
A numeric vector |
.scale |
A numeric value that is used to scale the output |
.power |
A numeric value that is used to raise the output to a power |
.log_diff |
A logical value that determines whether the output is a log difference |
.lags |
An integer that determines the number of lags to use |
The function calculates growth rates for a time series, allowing for scaling, exponentiation, and lag differences. It can be useful for financial data analysis, among other applications.
The growth rate is computed as follows:
If lags is positive and log_diff is FALSE: growth_rate = (((x / lag(x, lags))^power) - 1) * scale
If lags is positive and log_diff is TRUE: growth_rate = log(x / lag(x, lags)) * scale
If lags is negative and log_diff is FALSE: growth_rate = (((x / lead(x, -lags))^power) - 1) * scale
If lags is negative and log_diff is TRUE: growth_rate = log(x / lead(x, -lags)) * scale
A list object of workflows.
Steven P. Sanderson II, MPH
Other Vector Function:
ts_acceleration_vec(),
ts_velocity_vec()
# Calculate the growth rate of a time series without any transformations. ts_growth_rate_vec(c(100, 110, 120, 130)) # Calculate the growth rate with scaling and a power transformation. ts_growth_rate_vec(c(100, 110, 120, 130), .scale = 10, .power = 2) # Calculate the log differences of a time series with lags. ts_growth_rate_vec(c(100, 110, 120, 130), .log_diff = TRUE, .lags = -1) # Plot plot.ts(AirPassengers) plot.ts(ts_growth_rate_vec(AirPassengers))# Calculate the growth rate of a time series without any transformations. ts_growth_rate_vec(c(100, 110, 120, 130)) # Calculate the growth rate with scaling and a power transformation. ts_growth_rate_vec(c(100, 110, 120, 130), .scale = 10, .power = 2) # Calculate the log differences of a time series with lags. ts_growth_rate_vec(c(100, 110, 120, 130), .log_diff = TRUE, .lags = -1) # Plot plot.ts(AirPassengers) plot.ts(ts_growth_rate_vec(AirPassengers))
This function will take in a data set and return to you a tibble of useful information.
ts_info_tbl(.data, .date_col)ts_info_tbl(.data, .date_col)
.data |
The data you are passing to the function |
.date_col |
This is only needed if you are passing a tibble. |
This function can accept objects of the following classes:
ts
xts
mts
zoo
tibble/data.frame
The function will return the following pieces of information in a tibble:
name
class
frequency
start
end
var
length
A tibble
Steven P. Sanderson II, MPH
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
ts_info_tbl(AirPassengers) ts_info_tbl(BJsales)ts_info_tbl(AirPassengers) ts_info_tbl(BJsales)
Check if an object is a date class
ts_is_date_class(.x)ts_is_date_class(.x)
.x |
A vector to check |
Logical (TRUE/FALSE)
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
seq.Date(from = as.Date("2022-01-01"), by = "day", length.out = 10) %>% ts_is_date_class() letters %>% ts_is_date_class()seq.Date(from = as.Date("2022-01-01"), by = "day", length.out = 10) %>% ts_is_date_class() letters %>% ts_is_date_class()
This function outputs a list object of both data and plots.
The data output are the following:
lag_list
lag_tbl
correlation_lag_matrix
correlation_lag_tbl
The plots output are the following:
lag_plot
plotly_lag_plot
correlation_heatmap
plotly_heatmap
ts_lag_correlation( .data, .date_col, .value_col, .lags = 1, .heatmap_color_low = "white", .heatmap_color_hi = "steelblue" )ts_lag_correlation( .data, .date_col, .value_col, .lags = 1, .heatmap_color_low = "white", .heatmap_color_hi = "steelblue" )
.data |
A tibble of time series data |
.date_col |
A date column |
.value_col |
The value column being analyzed |
.lags |
This is a vector of integer lags, ie 1 or c(1,6,12) |
.heatmap_color_low |
What color should the low values of the heatmap of the correlation matrix be, the default is 'white' |
.heatmap_color_hi |
What color should the low values of the heatmap of the correlation matrix be, the default is 'steelblue' |
This function takes in a time series data in the form of a tibble and outputs a list object of data and plots. This function will take in an argument of '.lags' and get those lags in your data, outputting a correlation matrix, heatmap and lag plot among other things of the input data.
A list object
Steven P. Sanderson II, MPH
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
library(dplyr) df <- ts_to_tbl(AirPassengers) %>% select(-index) lags <- c(1,3,6,12) output <- ts_lag_correlation( .data = df, .date_col = date_col, .value_col = value, .lags = lags ) output$data$correlation_lag_matrix output$plots$lag_plotlibrary(dplyr) df <- ts_to_tbl(AirPassengers) %>% select(-index) lags <- c(1,3,6,12) output <- ts_lag_correlation( .data = df, .date_col = date_col, .value_col = value, .lags = lags ) output$data$correlation_lag_matrix output$plots$lag_plot
This function will produce two plots. Both of these are moving average plots.
One of the plots is from xts::plot.xts() and the other a ggplot2 plot. This
is done so that the user can choose which type is best for them. The plots are
stacked so each graph is on top of the other.
ts_ma_plot( .data, .date_col, .value_col, .ts_frequency = "monthly", .main_title = NULL, .secondary_title = NULL, .tertiary_title = NULL )ts_ma_plot( .data, .date_col, .value_col, .ts_frequency = "monthly", .main_title = NULL, .secondary_title = NULL, .tertiary_title = NULL )
.data |
The data you want to visualize. This should be pre-processed and
the aggregation should match the |
.date_col |
The data column from the |
.value_col |
The value column from the |
.ts_frequency |
The frequency of the aggregation, quoted, ie. "monthly", anything else will default to weekly, so it is very important that the data passed to this function be in either a weekly or monthly aggregation. |
.main_title |
The title of the main plot. |
.secondary_title |
The title of the second plot. |
.tertiary_title |
The title of the third plot. |
This function expects to take in a data.frame/tibble. It will return a list object so it is a good idea to save the output to a variable and extract from there.
A few time series data sets and two plots.
Steven P. Sanderson II, MPH
suppressPackageStartupMessages(library(dplyr)) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) output <- ts_ma_plot( .data = data_tbl, .date_col = date_col, .value_col = value ) output$pgrid output$xts_plt output$data_summary_tbl %>% head() output <- ts_ma_plot( .data = data_tbl, .date_col = date_col, .value_col = value, .ts_frequency = "week" ) output$pgrid output$xts_plt output$data_summary_tbl %>% head()suppressPackageStartupMessages(library(dplyr)) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) output <- ts_ma_plot( .data = data_tbl, .date_col = date_col, .value_col = value ) output$pgrid output$xts_plt output$data_summary_tbl %>% head() output <- ts_ma_plot( .data = data_tbl, .date_col = date_col, .value_col = value, .ts_frequency = "week" ) output$pgrid output$xts_plt output$data_summary_tbl %>% head()
This function will create a tuned model. It uses the ts_model_spec_tune_template()
under the hood to get the generic template that is used in the grid search.
ts_model_auto_tune( .modeltime_model_id, .calibration_tbl, .splits_obj, .drop_training_na = TRUE, .date_col, .value_col, .tscv_assess = "12 months", .tscv_skip = "6 months", .slice_limit = 6, .facet_ncol = 2, .grid_size = 30, .num_cores = 1, .best_metric = "rmse" )ts_model_auto_tune( .modeltime_model_id, .calibration_tbl, .splits_obj, .drop_training_na = TRUE, .date_col, .value_col, .tscv_assess = "12 months", .tscv_skip = "6 months", .slice_limit = 6, .facet_ncol = 2, .grid_size = 30, .num_cores = 1, .best_metric = "rmse" )
.modeltime_model_id |
The .model_id from a calibrated modeltime table. |
.calibration_tbl |
A calibrated modeltime table. |
.splits_obj |
The time_series_split object. |
.drop_training_na |
A boolean that will drop NA values from the training(splits) data |
.date_col |
The column that holds the date values. |
.value_col |
The column that holds the time series values. |
.tscv_assess |
A character expression like "12 months". This gets passed to
|
.tscv_skip |
A character expression like "6 months". This gets passed to
|
.slice_limit |
An integer that gets passed to |
.facet_ncol |
The number of faceted columns to be passed to plot_time_series_cv_plan |
.grid_size |
An integer that gets passed to the |
.num_cores |
The default is 1, you can set this to any integer value as long as it is equal to or less than the available cores on your machine. |
.best_metric |
The default is "rmse" and this can be set to any default dials metric. This must be passed as a character. |
This function can work with the following parsnip/modeltime engines:
"auto_arima"
"auto_arima_xgboost"
"ets"
"croston"
"theta"
"stlm_ets"
"tbats"
"stlm_arima"
"nnetar"
"prophet"
"prophet_xgboost"
"lm"
"glmnet"
"stan"
"spark"
"keras"
"earth"
"xgboost"
"kernlab"
This function returns a list object with several items inside of it. There are three categories of items that are inside of the list.
data
model_info
plots
The data section has the following items:
calibration_tbl This is the calibration data passed into the function.
calibration_tuned_tbl This is a calibration tibble that has used the
tuned workflow.
tscv_data_tbl This is the tibble of the time series cross validation.
tuned_results This is a tuning results tibble with all slices from the
time series cross validation.
best_tuned_results_tbl This is a tibble of the parameters for the best
test set with the chosen metric.
tscv_obj This is the actual time series cross validation object returned
from timetk::time_series_cv()
The model_info section has the following items:
model_spec This is the original modeltime/parsnip model specification.
model_spec_engine This is the engine used for the model specification.
model_spec_tuner This is the tuning model template returned from ts_model_spec_tune_template()
plucked_model This is the model that we have plucked from the calibration tibble
for tuning.
wflw_tune_spec This is a new workflow with the model_spec_tuner attached.
grid_spec This is the grid search specification for the tuning process.
tuned_tscv_wflw_spec This is the final tuned model where the workflow and
model have been finalized. This would be the model that you would want to
pull out if you are going to work with it further.
The plots section has the following items:
tune_results_plt This is a static ggplot of the grid search.
tscv_pl This is the time series cross validation plan plot.
A list object with multiple items.
Steven P. Sanderson II, MPH
Other Model Tuning:
ts_model_spec_tune_template()
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
## Not run: suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) data <- ts_to_tbl(AirPassengers) %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = data , .date_col = date_col , .pred_col = value ) wfsets <- ts_wfs_mars( .model_type = "earth" , .recipe_list = rec_objs ) wf_fits <- wfsets %>% modeltime_fit_workflowset( data = training(splits) , control = control_fit_workflowset( allow_par = TRUE , verbose = TRUE ) ) models_tbl <- wf_fits %>% filter(.model != "NULL") calibration_tbl <- models_tbl %>% modeltime_calibrate(new_data = testing(splits)) output <- ts_model_auto_tune( .modeltime_model_id = 1, .calibration_tbl = calibration_tbl, .splits_obj = splits, .drop_training_na = TRUE, .date_col = date_col, .value_col = value, .tscv_assess = "12 months", .tscv_skip = "3 months", .num_cores = parallel::detectCores() - 1 ) ## End(Not run)## Not run: suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) data <- ts_to_tbl(AirPassengers) %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = data , .date_col = date_col , .pred_col = value ) wfsets <- ts_wfs_mars( .model_type = "earth" , .recipe_list = rec_objs ) wf_fits <- wfsets %>% modeltime_fit_workflowset( data = training(splits) , control = control_fit_workflowset( allow_par = TRUE , verbose = TRUE ) ) models_tbl <- wf_fits %>% filter(.model != "NULL") calibration_tbl <- models_tbl %>% modeltime_calibrate(new_data = testing(splits)) output <- ts_model_auto_tune( .modeltime_model_id = 1, .calibration_tbl = calibration_tbl, .splits_obj = splits, .drop_training_na = TRUE, .date_col = date_col, .value_col = value, .tscv_assess = "12 months", .tscv_skip = "3 months", .num_cores = parallel::detectCores() - 1 ) ## End(Not run)
This function will expect to take in two models that will be used for comparison.
It is useful to use this after appropriately following the modeltime workflow and
getting two models to compare. This is an extension of the calibrate and plot, but
it only takes two models and is most likely better suited to be used after running
a model through the ts_model_auto_tune() function to see the difference in performance
after a base model has been tuned.
ts_model_compare( .model_1, .model_2, .type = "testing", .splits_obj, .data, .print_info = TRUE, .metric = "rmse" )ts_model_compare( .model_1, .model_2, .type = "testing", .splits_obj, .data, .print_info = TRUE, .metric = "rmse" )
.model_1 |
The model being compared to the base, this can also be a hyperparameter tuned model. |
.model_2 |
The base model. |
.type |
The default is the testing tibble, can be set to training as well. |
.splits_obj |
The splits object |
.data |
The original data that was passed to splits |
.print_info |
This is a boolean, the default is TRUE |
.metric |
This should be one of the following character strings:
|
This function expects to take two models. You must tell it if it will
be assessing the training or testing data, where the testing data is the default.
You must therefore supply the splits object to this function along with the origianl
dataset. You must also tell it which default modeltime accuracy metric should
be printed on the graph itself. You can also tell this function to print
information to the console or not. A static ggplot2 polot and an interactive
plotly plot will be returned inside of the output list.
The function outputs a list invisibly.
Steven P. Sanderson II, MPH
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
## Not run: suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(rsample)) suppressPackageStartupMessages(library(dplyr)) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) splits <- time_series_split( data = data_tbl, date_var = date_col, assess = "12 months", cumulative = TRUE ) rec_obj <- ts_auto_recipe( .data = data_tbl, .date_col = date_col, .pred_col = value ) wfs_mars <- ts_wfs_mars(.recipe_list = rec_obj) wf_fits <- wfs_mars %>% modeltime_fit_workflowset( data = training(splits) , control = control_fit_workflowset( allow_par = FALSE , verbose = TRUE ) ) calibration_tbl <- wf_fits %>% modeltime_calibrate(new_data = testing(splits)) base_mars <- calibration_tbl %>% pluck_modeltime_model(1) date_mars <- calibration_tbl %>% pluck_modeltime_model(2) ts_model_compare( .model_1 = base_mars, .model_2 = date_mars, .type = "testing", .splits_obj = splits, .data = data_tbl, .print_info = TRUE, .metric = "rmse" )$plots$static_plot ## End(Not run)## Not run: suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(rsample)) suppressPackageStartupMessages(library(dplyr)) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) splits <- time_series_split( data = data_tbl, date_var = date_col, assess = "12 months", cumulative = TRUE ) rec_obj <- ts_auto_recipe( .data = data_tbl, .date_col = date_col, .pred_col = value ) wfs_mars <- ts_wfs_mars(.recipe_list = rec_obj) wf_fits <- wfs_mars %>% modeltime_fit_workflowset( data = training(splits) , control = control_fit_workflowset( allow_par = FALSE , verbose = TRUE ) ) calibration_tbl <- wf_fits %>% modeltime_calibrate(new_data = testing(splits)) base_mars <- calibration_tbl %>% pluck_modeltime_model(1) date_mars <- calibration_tbl %>% pluck_modeltime_model(2) ts_model_compare( .model_1 = base_mars, .model_2 = date_mars, .type = "testing", .splits_obj = splits, .data = data_tbl, .print_info = TRUE, .metric = "rmse" )$plots$static_plot ## End(Not run)
This takes in a calibration tibble and computes the ranks of the models inside of it.
ts_model_rank_tbl(.calibration_tbl)ts_model_rank_tbl(.calibration_tbl)
.calibration_tbl |
A calibrated modeltime table. |
This takes in a calibration tibble and computes the ranks of the models inside
of it. It computes for now only the default yardstick metrics from modeltime
These are the following using the dplyr min_rank() function with desc use
on rsq:
"rmse"
"mae"
"mape"
"smape"
"rsq"
A tibble with models ranked by metric performance order
Steven P. Sanderson II, MPH
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
# NOT RUN ## Not run: suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(rsample)) suppressPackageStartupMessages(library(workflows)) suppressPackageStartupMessages(library(parsnip)) suppressPackageStartupMessages(library(recipes)) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) splits <- time_series_split( data_tbl, date_var = date_col, assess = "12 months", cumulative = TRUE ) rec_obj <- recipe(value ~ ., training(splits)) model_spec_arima <- arima_reg() %>% set_engine(engine = "auto_arima") model_spec_mars <- mars(mode = "regression") %>% set_engine("earth") wflw_fit_arima <- workflow() %>% add_recipe(rec_obj) %>% add_model(model_spec_arima) %>% fit(training(splits)) wflw_fit_mars <- workflow() %>% add_recipe(rec_obj) %>% add_model(model_spec_mars) %>% fit(training(splits)) model_tbl <- modeltime_table(wflw_fit_arima, wflw_fit_mars) calibration_tbl <- model_tbl %>% modeltime_calibrate(new_data = testing(splits)) ts_model_rank_tbl(calibration_tbl) ## End(Not run)# NOT RUN ## Not run: suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(rsample)) suppressPackageStartupMessages(library(workflows)) suppressPackageStartupMessages(library(parsnip)) suppressPackageStartupMessages(library(recipes)) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) splits <- time_series_split( data_tbl, date_var = date_col, assess = "12 months", cumulative = TRUE ) rec_obj <- recipe(value ~ ., training(splits)) model_spec_arima <- arima_reg() %>% set_engine(engine = "auto_arima") model_spec_mars <- mars(mode = "regression") %>% set_engine("earth") wflw_fit_arima <- workflow() %>% add_recipe(rec_obj) %>% add_model(model_spec_arima) %>% fit(training(splits)) wflw_fit_mars <- workflow() %>% add_recipe(rec_obj) %>% add_model(model_spec_mars) %>% fit(training(splits)) model_tbl <- modeltime_table(wflw_fit_arima, wflw_fit_mars) calibration_tbl <- model_tbl %>% modeltime_calibrate(new_data = testing(splits)) ts_model_rank_tbl(calibration_tbl) ## End(Not run)
This function will create a generic tuneable model specification, this function
can be used by itself and is called internally by ts_model_auto_tune().
ts_model_spec_tune_template(.parsnip_engine = NULL, .model_spec_class = NULL)ts_model_spec_tune_template(.parsnip_engine = NULL, .model_spec_class = NULL)
.parsnip_engine |
The model engine that is used by |
.model_spec_class |
The model spec class that is use by |
This function takes in a single parameter and uses that to output a generic tuneable model specification. This function can work with the following parsnip/modeltime engines:
"auto_arima"
"auto_arima_xgboost"
"ets"
"croston"
"theta"
"smooth_es"
"stlm_ets"
"tbats"
"stlm_arima"
"nnetar"
"prophet"
"prophet_xgboost"
"lm"
"glmnet"
"stan"
"spark"
"keras"
"earth"
"xgboost"
"kernlab"
A tuneable parsnip model specification.
Steven P. Sanderson II, MPH
Other Model Tuning:
ts_model_auto_tune()
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
ts_model_spec_tune_template("ets") ts_model_spec_tune_template("prophet")ts_model_spec_tune_template("ets") ts_model_spec_tune_template("prophet")
A control chart is a specific type of graph that shows data points between upper and lower limits over a period of time. You can use it to understand if the process is in control or not. These charts commonly have three types of lines such as upper and lower specification limits, upper and lower limits and planned value. By the help of these lines, Control Charts show the process behavior over time.
ts_qc_run_chart( .data, .date_col, .value_col, .interactive = FALSE, .median = TRUE, .cl = TRUE, .mcl = TRUE, .ucl = TRUE, .lc = FALSE, .lmcl = FALSE, .llcl = FALSE )ts_qc_run_chart( .data, .date_col, .value_col, .interactive = FALSE, .median = TRUE, .cl = TRUE, .mcl = TRUE, .ucl = TRUE, .lc = FALSE, .lmcl = FALSE, .llcl = FALSE )
.data |
The data.frame/tibble to be passed. |
.date_col |
The column holding the timestamp. |
.value_col |
The column with the values to be analyzed. |
.interactive |
Default is FALSE, TRUE for an interactive plotly plot. |
.median |
Default is TRUE. This will show the median line of the data. |
.cl |
This is the first upper control line |
.mcl |
This is the second sigma control line positive |
.ucl |
This is the third sigma control line positive |
.lc |
This is the first negative control line |
.lmcl |
This is the second sigma negative control line |
.llcl |
This si the thrid sigma negative control line |
Expects a time-series tibble/data.frame
Expects a date column and a value column
A static ggplot2 graph or if .interactive is set to TRUE a plotly plot
Steven P. Sanderson II, MPH
library(dplyr) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) data_tbl %>% ts_qc_run_chart( .date_col = date_col , .value_col = value , .llcl = TRUE )library(dplyr) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) data_tbl %>% ts_qc_run_chart( .date_col = date_col , .value_col = value , .llcl = TRUE )
This takes in a calibration tibble and will produce a QQ plot.
ts_qq_plot(.calibration_tbl, .model_id = NULL, .interactive = FALSE)ts_qq_plot(.calibration_tbl, .model_id = NULL, .interactive = FALSE)
.calibration_tbl |
A calibrated modeltime table. |
.model_id |
The id of a particular model from a calibration tibble. If
there are multiple models in the tibble and this remains NULL then the
plot will be returned using |
.interactive |
A boolean with a default value of FALSE. TRUE will produce
an interactive |
This takes in a calibration tibble and will create a QQ plot. You can also
pass in a model_id and a boolean for interactive which will return a
plotly::ggplotly interactive plot.
A QQ plot.
Steven P. Sanderson II, MPH
https://en.wikipedia.org/wiki/Q%E2%80%93Q_plot
Other Plot:
ts_brownian_motion_plot(),
ts_event_analysis_plot(),
ts_scedacity_scatter_plot()
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
# NOT RUN ## Not run: suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(rsample)) suppressPackageStartupMessages(library(workflows)) suppressPackageStartupMessages(library(parsnip)) suppressPackageStartupMessages(library(recipes)) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) splits <- time_series_split( data_tbl, date_var = date_col, assess = "12 months", cumulative = TRUE ) rec_obj <- recipe(value ~ ., training(splits)) model_spec_arima <- arima_reg() %>% set_engine(engine = "auto_arima") model_spec_mars <- mars(mode = "regression") %>% set_engine("earth") wflw_fit_arima <- workflow() %>% add_recipe(rec_obj) %>% add_model(model_spec_arima) %>% fit(training(splits)) wflw_fit_mars <- workflow() %>% add_recipe(rec_obj) %>% add_model(model_spec_mars) %>% fit(training(splits)) model_tbl <- modeltime_table(wflw_fit_arima, wflw_fit_mars) calibration_tbl <- model_tbl %>% modeltime_calibrate(new_data = testing(splits)) ts_qq_plot(calibration_tbl) ## End(Not run)# NOT RUN ## Not run: suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(rsample)) suppressPackageStartupMessages(library(workflows)) suppressPackageStartupMessages(library(parsnip)) suppressPackageStartupMessages(library(recipes)) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) splits <- time_series_split( data_tbl, date_var = date_col, assess = "12 months", cumulative = TRUE ) rec_obj <- recipe(value ~ ., training(splits)) model_spec_arima <- arima_reg() %>% set_engine(engine = "auto_arima") model_spec_mars <- mars(mode = "regression") %>% set_engine("earth") wflw_fit_arima <- workflow() %>% add_recipe(rec_obj) %>% add_model(model_spec_arima) %>% fit(training(splits)) wflw_fit_mars <- workflow() %>% add_recipe(rec_obj) %>% add_model(model_spec_mars) %>% fit(training(splits)) model_tbl <- modeltime_table(wflw_fit_arima, wflw_fit_mars) calibration_tbl <- model_tbl %>% modeltime_calibrate(new_data = testing(splits)) ts_qq_plot(calibration_tbl) ## End(Not run)
This function takes in four arguments and returns a tibble of random walks.
ts_random_walk( .mean = 0, .sd = 0.1, .num_walks = 100, .periods = 100, .initial_value = 1000 )ts_random_walk( .mean = 0, .sd = 0.1, .num_walks = 100, .periods = 100, .initial_value = 1000 )
.mean |
The desired mean of the random walks |
.sd |
The standard deviation of the random walks |
.num_walks |
The number of random walks you want generated |
.periods |
The length of the random walk(s) you want generated |
.initial_value |
The initial value where the random walks should start |
Monte Carlo simulations were first formally designed in the 1940’s while developing nuclear weapons, and since have been heavily used in various fields to use randomness solve problems that are potentially deterministic in nature. In finance, Monte Carlo simulations can be a useful tool to give a sense of how assets with certain characteristics might behave in the future. While there are more complex and sophisticated financial forecasting methods such as ARIMA (Auto-Regressive Integrated Moving Average) and GARCH (Generalized Auto-Regressive Conditional Heteroskedasticity) which attempt to model not only the randomness but underlying macro factors such as seasonality and volatility clustering, Monte Carlo random walks work surprisingly well in illustrating market volatility as long as the results are not taken too seriously.
A tibble
Other Data Generator:
tidy_fft(),
ts_brownian_motion(),
ts_brownian_motion_augment(),
ts_geometric_brownian_motion(),
ts_geometric_brownian_motion_augment()
ts_random_walk( .mean = 0, .sd = 1, .num_walks = 25, .periods = 180, .initial_value = 6 )ts_random_walk( .mean = 0, .sd = 1, .num_walks = 25, .periods = 180, .initial_value = 6 )
ggplot2 layersGet layers to add to a ggplot graph from the ts_random_walk() function.
ts_random_walk_ggplot_layers(.data)ts_random_walk_ggplot_layers(.data)
.data |
The data passed to the function. |
Set the intercept of the initial value from the random walk
Set the max and min of the cumulative sum of the random walks
A ggplot2 layers object
Steven P. Sanderson II, MPH
library(ggplot2) df <- ts_random_walk() df %>% ggplot( mapping = aes( x = x , y = cum_y , color = factor(run) , group = factor(run) ) ) + geom_line(alpha = 0.8) + ts_random_walk_ggplot_layers(df)library(ggplot2) df <- ts_random_walk() df %>% ggplot( mapping = aes( x = x , y = cum_y , color = factor(run) , group = factor(run) ) ) + geom_line(alpha = 0.8) + ts_random_walk_ggplot_layers(df)
8 Hex RGB color definitions suitable for charts for colorblind people.
ts_scale_color_colorblind(..., theme = "ts")ts_scale_color_colorblind(..., theme = "ts")
... |
Data passed in from a |
theme |
Right now this is |
This function is used in others in order to help render plots for those that are color blind.
A gggplot layer
Steven P. Sanderson II, MPH
8 Hex RGB color definitions suitable for charts for colorblind people.
ts_scale_fill_colorblind(..., theme = "ts")ts_scale_fill_colorblind(..., theme = "ts")
... |
Data passed in from a |
theme |
Right now this is |
This function is used in others in order to help render plots for those that are color blind.
A gggplot layer
Steven P. Sanderson II, MPH
This takes in a calibration tibble and will produce a scedacity plot.
ts_scedacity_scatter_plot( .calibration_tbl, .model_id = NULL, .interactive = FALSE )ts_scedacity_scatter_plot( .calibration_tbl, .model_id = NULL, .interactive = FALSE )
.calibration_tbl |
A calibrated modeltime table. |
.model_id |
The id of a particular model from a calibration tibble. If
there are multiple models in the tibble and this remains NULL then the
plot will be returned using |
.interactive |
A boolean with a default value of FALSE. TRUE will produce
an interactive |
This takes in a calibration tibble and will create a scedacity plot. You can also
pass in a model_id and a boolean for interactive which will return a
plotly::ggplotly interactive plot.
A Scedacity plot.
Steven P. Sanderson II, MPH
https://en.wikipedia.org/wiki/Homoscedasticity
Other Plot:
ts_brownian_motion_plot(),
ts_event_analysis_plot(),
ts_qq_plot()
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
# NOT RUN ## Not run: suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(rsample)) suppressPackageStartupMessages(library(workflows)) suppressPackageStartupMessages(library(parsnip)) suppressPackageStartupMessages(library(recipes)) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) splits <- time_series_split( data_tbl, date_var = date_col, assess = "12 months", cumulative = TRUE ) rec_obj <- recipe(value ~ ., training(splits)) model_spec_arima <- arima_reg() %>% set_engine(engine = "auto_arima") model_spec_mars <- mars(mode = "regression") %>% set_engine("earth") wflw_fit_arima <- workflow() %>% add_recipe(rec_obj) %>% add_model(model_spec_arima) %>% fit(training(splits)) wflw_fit_mars <- workflow() %>% add_recipe(rec_obj) %>% add_model(model_spec_mars) %>% fit(training(splits)) model_tbl <- modeltime_table(wflw_fit_arima, wflw_fit_mars) calibration_tbl <- model_tbl %>% modeltime_calibrate(new_data = testing(splits)) ts_scedacity_scatter_plot(calibration_tbl) ## End(Not run)# NOT RUN ## Not run: suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(rsample)) suppressPackageStartupMessages(library(workflows)) suppressPackageStartupMessages(library(parsnip)) suppressPackageStartupMessages(library(recipes)) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) splits <- time_series_split( data_tbl, date_var = date_col, assess = "12 months", cumulative = TRUE ) rec_obj <- recipe(value ~ ., training(splits)) model_spec_arima <- arima_reg() %>% set_engine(engine = "auto_arima") model_spec_mars <- mars(mode = "regression") %>% set_engine("earth") wflw_fit_arima <- workflow() %>% add_recipe(rec_obj) %>% add_model(model_spec_arima) %>% fit(training(splits)) wflw_fit_mars <- workflow() %>% add_recipe(rec_obj) %>% add_model(model_spec_mars) %>% fit(training(splits)) model_tbl <- modeltime_table(wflw_fit_arima, wflw_fit_mars) calibration_tbl <- model_tbl %>% modeltime_calibrate(new_data = testing(splits)) ts_scedacity_scatter_plot(calibration_tbl) ## End(Not run)
This function will take in a value column and return any number n moving averages.
ts_sma_plot( .data, .date_col, .value_col, .sma_order = 2, .func = mean, .align = "center", .partial = FALSE )ts_sma_plot( .data, .date_col, .value_col, .sma_order = 2, .func = mean, .align = "center", .partial = FALSE )
.data |
The data that you are passing, must be a data.frame/tibble. |
.date_col |
The column that holds the date. |
.value_col |
The column that holds the value. |
.sma_order |
This will default to 1. This can be a vector like c(2,4,6,12) |
.func |
The unquoted function you want to pass, mean, median, etc |
.align |
This can be either "left", "center", "right" |
.partial |
This is a bool value of TRUE/FALSE, the default is TRUE |
This function will accept a time series object or a tibble/data.frame. This is a
simple wrapper around timetk::slidify_vec(). It uses that function to do the underlying
moving average work.
It can only handle a single moving average at a time and therefore if multiple are called for, it will loop through and append data to a tibble object.
Will return a list object.
Steven P. Sanderson II, MPH
df <- ts_to_tbl(AirPassengers) out <- ts_sma_plot(df, date_col, value, .sma_order = c(3,6)) out$data out$plots$static_plotdf <- ts_to_tbl(AirPassengers) out <- ts_sma_plot(df, date_col, value, .sma_order = c(3,6)) out$data out$plots$static_plot
Sometimes we want to see the training and testing data in a plot. This is a
simple wrapper around a couple of functions from the timetk package.
ts_splits_plot(.splits_obj, .date_col, .value_col)ts_splits_plot(.splits_obj, .date_col, .value_col)
.splits_obj |
The predefined splits object. |
.date_col |
The date column for the time series. |
.value_col |
The value column of the time series. |
You should already have a splits object defined. This function takes in three parameters, the splits object, a date column and the value column.
A time series cv plan plot
Steven P. Sanderson II, MPH
https://business-science.github.io/timetk/reference/plot_time_series_cv_plan.html(tk_time_sers_cv_plan)
https://business-science.github.io/timetk/reference/plot_time_series_cv_plan.html(plot_time_series_cv_plan)
suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) data <- ts_to_tbl(AirPassengers) %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_splits_plot( .splits_obj = splits, .date_col = date_col, .value_col = value )suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) data <- ts_to_tbl(AirPassengers) %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) ts_splits_plot( .splits_obj = splits, .date_col = date_col, .value_col = value )
Given a tibble/data.frame, you can get information on what happens before, after,
or in both directions of some given event, where the event is defined by some
percentage increase/decrease in values from time t to t+1
ts_time_event_analysis_tbl( .data, .date_col, .value_col, .percent_change = 0.05, .horizon = 12, .precision = 2, .direction = "forward", .filter_non_event_groups = TRUE )ts_time_event_analysis_tbl( .data, .date_col, .value_col, .percent_change = 0.05, .horizon = 12, .precision = 2, .direction = "forward", .filter_non_event_groups = TRUE )
.data |
The date.frame/tibble that holds the data. |
.date_col |
The column with the date value. |
.value_col |
The column with the value you are measuring. |
.percent_change |
This defaults to 0.05 which is a 5% increase in the
|
.horizon |
How far do you want to look back or ahead. |
.precision |
The default is 2 which means it rounds the lagged 1 value percent change to 2 decimal points. You may want more for more finely tuned results, this will result in fewer groupings. |
.direction |
The default is |
.filter_non_event_groups |
The default is TRUE, this drops groupings with no events on the rare occasion it does occur. |
This takes in a data.frame/tibble of a time series. It requires a date column,
and a value column. You can convert a ts/xts/zoo/mts object into a tibble by
using the ts_to_tbl() function.
You will provide the function with a percentage change in the form of -1 to 1
inclusive. You then provide a time horizon in which you want to see. For example
you may want to see what happens to AirPassengers after a 0.1 percent increase
in volume.
The next most important thing to supply is the direction. Do you want to see what typically happens after such an event, what leads up to such an event, or both.
A tibble.
Steven P. Sanderson II, MPH
Other Time_Filtering:
ts_compare_data()
suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(ggplot2)) df_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) tst <- ts_time_event_analysis_tbl(df_tbl, date_col, value, .direction = "both", .horizon = 6) glimpse(tst) tst %>% ggplot(aes(x = x, y = mean_event_change)) + geom_line() + geom_line(aes(y = event_change_ci_high), color = "blue", linetype = "dashed") + geom_line(aes(y = event_change_ci_low), color = "blue", linetype = "dashed") + geom_vline(xintercept = 7, color = "red", linetype = "dashed") + theme_minimal() + labs( title = "'AirPassengers' Event Analysis at 5% Increase", subtitle = "Vertical Red line is normalized event epoch - Direction: Both", x = "", y = "Mean Event Change" )suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(ggplot2)) df_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) tst <- ts_time_event_analysis_tbl(df_tbl, date_col, value, .direction = "both", .horizon = 6) glimpse(tst) tst %>% ggplot(aes(x = x, y = mean_event_change)) + geom_line() + geom_line(aes(y = event_change_ci_high), color = "blue", linetype = "dashed") + geom_line(aes(y = event_change_ci_low), color = "blue", linetype = "dashed") + geom_vline(xintercept = 7, color = "red", linetype = "dashed") + theme_minimal() + labs( title = "'AirPassengers' Event Analysis at 5% Increase", subtitle = "Vertical Red line is normalized event epoch - Direction: Both", x = "", y = "Mean Event Change" )
This function takes in a time-series object and returns it in a
tibble format.
ts_to_tbl(.data)ts_to_tbl(.data)
.data |
The time-series object you want transformed into a |
This function makes use of timetk::tk_tbl() under the hood to obtain
the initial tibble object. After the inital object is obtained a new column
called date_col is constructed from the index column using lubridate if
an index column is returned.
A tibble
Steven P. Sanderson II, MPH
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
ts_to_tbl(BJsales) ts_to_tbl(AirPassengers)ts_to_tbl(BJsales) ts_to_tbl(AirPassengers)
Takes a numeric vector and will return the velocity of that vector.
ts_velocity_augment(.data, .value, .names = "auto")ts_velocity_augment(.data, .value, .names = "auto")
.data |
The data being passed that will be augmented by the function. |
.value |
This is passed |
.names |
The default is "auto" |
Takes a numeric vector and will return the velocity of that vector. The velocity of a time series is computed by taking the first difference, so
This function is intended to be used on its own in order to add columns to a tibble.
A augmented
Steven P. Sanderson II, MPH
Other Augment Function:
ts_acceleration_augment(),
ts_growth_rate_augment()
suppressPackageStartupMessages(library(dplyr)) len_out = 10 by_unit = "month" start_date = as.Date("2021-01-01") data_tbl <- tibble( date_col = seq.Date(from = start_date, length.out = len_out, by = by_unit), a = rnorm(len_out), b = runif(len_out) ) ts_velocity_augment(data_tbl, b)suppressPackageStartupMessages(library(dplyr)) len_out = 10 by_unit = "month" start_date = as.Date("2021-01-01") data_tbl <- tibble( date_col = seq.Date(from = start_date, length.out = len_out, by = by_unit), a = rnorm(len_out), b = runif(len_out) ) ts_velocity_augment(data_tbl, b)
Takes a numeric vector and will return the velocity of that vector.
ts_velocity_vec(.x)ts_velocity_vec(.x)
.x |
A numeric vector |
Takes a numeric vector and will return the velocity of that vector. The velocity of a time series is computed by taking the first difference, so
This function can be used on it's own. It is also the basis for the function
ts_velocity_augment().
A numeric vector
Steven P. Sanderson II, MPH
Other Vector Function:
ts_acceleration_vec(),
ts_growth_rate_vec()
suppressPackageStartupMessages(library(dplyr)) len_out = 25 by_unit = "month" start_date = as.Date("2021-01-01") data_tbl <- tibble( date_col = seq.Date(from = start_date, length.out = len_out, by = by_unit), a = rnorm(len_out), b = runif(len_out) ) vec_1 <- ts_velocity_vec(data_tbl$b) plot(data_tbl$b) lines(data_tbl$b) lines(vec_1, col = "blue")suppressPackageStartupMessages(library(dplyr)) len_out = 25 by_unit = "month" start_date = as.Date("2021-01-01") data_tbl <- tibble( date_col = seq.Date(from = start_date, length.out = len_out, by = by_unit), a = rnorm(len_out), b = runif(len_out) ) vec_1 <- ts_velocity_vec(data_tbl$b) plot(data_tbl$b) lines(data_tbl$b) lines(vec_1, col = "blue")
This function will produce three plots faceted on a single graph. The three graphs are the following:
Value Plot (Actual values)
Value Velocity Plot
Value Acceleration Plot
ts_vva_plot(.data, .date_col, .value_col)ts_vva_plot(.data, .date_col, .value_col)
.data |
The data you want to visualize. This should be pre-processed and
the aggregation should match the |
.date_col |
The data column from the |
.value_col |
The value column from the |
This function expects to take in a data.frame/tibble. It will return
a list object that contains the augmented data along with a static plot and
an interactive plotly plot. It is important that the data be prepared and have
at minimum a date column and the value column as they need to be supplied to
the function. If your data is a ts, xts, zoo or mts then use ts_to_tbl() to
convert it to a tibble.
The original time series augmented with the differenced data, a static plot and a plotly plot of the ggplot object. The output is a list that gets returned invisibly.
Steven P. Sanderson II, MPH
suppressPackageStartupMessages(library(dplyr)) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) ts_vva_plot(data_tbl, date_col, value)$plots$static_plotsuppressPackageStartupMessages(library(dplyr)) data_tbl <- ts_to_tbl(AirPassengers) %>% select(-index) ts_vva_plot(data_tbl, date_col, value)$plots$static_plot
This function is used to quickly create a workflowsets object.
ts_wfs_arima_boost( .model_type = "all_engines", .recipe_list, .trees = 10, .min_node = 2, .tree_depth = 6, .learn_rate = 0.015, .stop_iter = NULL, .seasonal_period = 0, .non_seasonal_ar = 0, .non_seasonal_differences = 0, .non_seasonal_ma = 0, .seasonal_ar = 0, .seasonal_differences = 0, .seasonal_ma = 0 )ts_wfs_arima_boost( .model_type = "all_engines", .recipe_list, .trees = 10, .min_node = 2, .tree_depth = 6, .learn_rate = 0.015, .stop_iter = NULL, .seasonal_period = 0, .non_seasonal_ar = 0, .non_seasonal_differences = 0, .non_seasonal_ma = 0, .seasonal_ar = 0, .seasonal_differences = 0, .seasonal_ma = 0 )
.model_type |
This is where you will set your engine. It uses
|
.recipe_list |
You must supply a list of recipes. list(rec_1, rec_2, ...) |
.trees |
An integer for the number of trees contained in the ensemble. |
.min_node |
An integer for the minimum number of data points in a node that is required for the node to be split further. |
.tree_depth |
An integer for the maximum depth of the tree (i.e. number of splits) (specific engines only). |
.learn_rate |
A number for the rate at which the boosting algorithm adapts from iteration-to-iteration (specific engines only). |
.stop_iter |
The number of iterations without improvement before stopping (xgboost only). |
.seasonal_period |
Set to 0, |
.non_seasonal_ar |
Set to 0, |
.non_seasonal_differences |
Set to 0, |
.non_seasonal_ma |
Set to 0, |
.seasonal_ar |
Set to 0, |
.seasonal_differences |
Set to 0, |
.seasonal_ma |
Set to 0, |
This function expects to take in the recipes that you want to use in the modeling process. This is an automated workflow process. There are sensible defaults set for the model specification, but if you choose you can set them yourself if you have a good understanding of what they should be. The mode is set to "regression".
This uses the option set_engine("auto_arima_xgboost") or set_engine("arima_xgboost")
modeltime::arima_boost() arima_boost() is a way to generate a specification
of a time series model that uses boosting to improve modeling errors
(residuals) on Exogenous Regressors. It works with both "automated" ARIMA
(auto.arima) and standard ARIMA (arima). The main algorithms are:
Auto ARIMA + XGBoost Errors (engine = auto_arima_xgboost, default)
ARIMA + XGBoost Errors (engine = arima_xgboost)
Returns a workflowsets object.
Steven P. Sanderson II, MPH
https://workflowsets.tidymodels.org/
https://business-science.github.io/modeltime/reference/arima_boost.html
Other Auto Workflowsets:
ts_wfs_auto_arima(),
ts_wfs_ets_reg(),
ts_wfs_lin_reg(),
ts_wfs_mars(),
ts_wfs_nnetar_reg(),
ts_wfs_prophet_reg(),
ts_wfs_svm_poly(),
ts_wfs_svm_rbf(),
ts_wfs_xgboost()
suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_arima_boost("all_engines", rec_objs) wf_setssuppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_arima_boost("all_engines", rec_objs) wf_sets
This function is used to quickly create a workflowsets object.
ts_wfs_auto_arima(.model_type = "auto_arima", .recipe_list)ts_wfs_auto_arima(.model_type = "auto_arima", .recipe_list)
.model_type |
This is where you will set your engine. It uses
|
.recipe_list |
You must supply a list of recipes. list(rec_1, rec_2, ...) |
This function expects to take in the recipes that you want to use in the modeling process. This is an automated workflow process. There are sensible defaults set for the model specification, but if you choose you can set them yourself if you have a good understanding of what they should be. The mode is set to "regression".
This only uses the option set_engine("auto_arima") and therefore the .model_type
is not needed. The parameter is kept because it is possible in the future that
this could change, and it keeps with the framework of how other functions
are written.
modeltime::arima_reg() arima_reg() is a way to generate a specification of
an ARIMA model before fitting and allows the model to be created using
different packages. Currently the only package is forecast.
Returns a workflowsets object.
Steven P. Sanderson II, MPH
https://workflowsets.tidymodels.org/
https://business-science.github.io/modeltime/reference/arima_reg.html
Other Auto Workflowsets:
ts_wfs_arima_boost(),
ts_wfs_ets_reg(),
ts_wfs_lin_reg(),
ts_wfs_mars(),
ts_wfs_nnetar_reg(),
ts_wfs_prophet_reg(),
ts_wfs_svm_poly(),
ts_wfs_svm_rbf(),
ts_wfs_xgboost()
suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_auto_arima("auto_arima", rec_objs) wf_setssuppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_auto_arima("auto_arima", rec_objs) wf_sets
This function is used to quickly create a workflowsets object.
ts_wfs_ets_reg( .model_type = "all_engines", .recipe_list, .seasonal_period = "auto", .error = "auto", .trend = "auto", .season = "auto", .damping = "auto", .smooth_level = 0.1, .smooth_trend = 0.1, .smooth_seasonal = 0.1 )ts_wfs_ets_reg( .model_type = "all_engines", .recipe_list, .seasonal_period = "auto", .error = "auto", .trend = "auto", .season = "auto", .damping = "auto", .smooth_level = 0.1, .smooth_trend = 0.1, .smooth_seasonal = 0.1 )
.model_type |
This is where you will set your engine. It uses
|
.recipe_list |
You must supply a list of recipes. list(rec_1, rec_2, ...) |
.seasonal_period |
A seasonal frequency. Uses "auto" by default. A character phrase of "auto" or time-based phrase of "2 weeks" can be used if a date or date-time variable is provided. See Fit Details below. |
.error |
The form of the error term: "auto", "additive", or "multiplicative". If the error is multiplicative, the data must be non-negative. |
.trend |
The form of the trend term: "auto", "additive", "multiplicative" or0 "none". |
.season |
The form of the seasonal term: "auto", "additive", "multiplicative" or "none". |
.damping |
Apply damping to a trend: "auto", "damped", or "none". |
.smooth_level |
This is often called the "alpha" parameter used as the base level smoothing factor for exponential smoothing models. |
.smooth_trend |
This is often called the "beta" parameter used as the trend smoothing factor for exponential smoothing models. |
.smooth_seasonal |
This is often called the "gamma" parameter used as the seasonal smoothing factor for exponential smoothing models. |
This function expects to take in the recipes that you want to use in the modeling process. This is an automated workflow process. There are sensible defaults set for the model specification, but if you choose you can set them yourself if you have a good understanding of what they should be. The mode is set to "regression".
This uses the following engines:
modeltime::exp_smoothing() exp_smoothing() is a way to generate a specification
of an Exponential Smoothing model before fitting and allows the model to be
created using different packages. Currently the only package is forecast.
Several algorithms are implemented:
"ets"
"croston"
"theta"
"smooth_es
Returns a workflowsets object.
Steven P. Sanderson II, MPH
https://workflowsets.tidymodels.org/
https://business-science.github.io/modeltime/reference/exp_smoothing.html
Other Auto Workflowsets:
ts_wfs_arima_boost(),
ts_wfs_auto_arima(),
ts_wfs_lin_reg(),
ts_wfs_mars(),
ts_wfs_nnetar_reg(),
ts_wfs_prophet_reg(),
ts_wfs_svm_poly(),
ts_wfs_svm_rbf(),
ts_wfs_xgboost()
suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_ets_reg("all_engines", rec_objs) wf_setssuppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_ets_reg("all_engines", rec_objs) wf_sets
This function is used to quickly create a workflowsets object.
ts_wfs_lin_reg(.model_type, .recipe_list, .penalty = 1, .mixture = 0.5)ts_wfs_lin_reg(.model_type, .recipe_list, .penalty = 1, .mixture = 0.5)
.model_type |
This is where you will set your engine. It uses
Not yet implemented are:
|
.recipe_list |
You must supply a list of recipes. list(rec_1, rec_2, ...) |
.penalty |
The penalty parameter of the glmnet. The default is 1 |
.mixture |
The mixture parameter of the glmnet. The default is 0.5 |
This function expects to take in the recipes that you want to use in
the modeling process. This is an automated workflow process. There are sensible
defaults set for the glmnet model specification, but if you choose you can
set them yourself if you have a good understanding of what they should be.
Returns a workflowsets object.
Steven P. Sanderson II, MPH
https://workflowsets.tidymodels.org/(workflowsets)
Other Auto Workflowsets:
ts_wfs_arima_boost(),
ts_wfs_auto_arima(),
ts_wfs_ets_reg(),
ts_wfs_mars(),
ts_wfs_nnetar_reg(),
ts_wfs_prophet_reg(),
ts_wfs_svm_poly(),
ts_wfs_svm_rbf(),
ts_wfs_xgboost()
suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_lin_reg("all_engines", rec_objs) wf_setssuppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_lin_reg("all_engines", rec_objs) wf_sets
This function is used to quickly create a workflowsets object.
ts_wfs_mars( .model_type = "earth", .recipe_list, .num_terms = 200, .prod_degree = 1, .prune_method = "backward" )ts_wfs_mars( .model_type = "earth", .recipe_list, .num_terms = 200, .prod_degree = 1, .prune_method = "backward" )
.model_type |
This is where you will set your engine. It uses
|
.recipe_list |
You must supply a list of recipes. list(rec_1, rec_2, ...) |
.num_terms |
The number of features that will be retained in the final model, including the intercept. |
.prod_degree |
The highest possible interaction degree. |
.prune_method |
The pruning method. This is a character, the default is "backward". You can choose from one of the following:
|
This function expects to take in the recipes that you want to use in the modeling process. This is an automated workflow process. There are sensible defaults set for the model specification, but if you choose you can set them yourself if you have a good understanding of what they should be. The mode is set to "regression".
This only uses the option set_engine("earth") and therefore the .model_type
is not needed. The parameter is kept because it is possible in the future that
this could change, and it keeps with the framework of how other functions
are written.
Returns a workflowsets object.
Steven P. Sanderson II, MPH
https://workflowsets.tidymodels.org/
https://parsnip.tidymodels.org/reference/mars.html
Other Auto Workflowsets:
ts_wfs_arima_boost(),
ts_wfs_auto_arima(),
ts_wfs_ets_reg(),
ts_wfs_lin_reg(),
ts_wfs_nnetar_reg(),
ts_wfs_prophet_reg(),
ts_wfs_svm_poly(),
ts_wfs_svm_rbf(),
ts_wfs_xgboost()
suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_mars("earth", rec_objs) wf_setssuppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_mars("earth", rec_objs) wf_sets
This function is used to quickly create a workflowsets object.
ts_wfs_nnetar_reg( .model_type = "nnetar", .recipe_list, .non_seasonal_ar = 0, .seasonal_ar = 0, .hidden_units = 5, .num_networks = 10, .penalty = 0.1, .epochs = 10 )ts_wfs_nnetar_reg( .model_type = "nnetar", .recipe_list, .non_seasonal_ar = 0, .seasonal_ar = 0, .hidden_units = 5, .num_networks = 10, .penalty = 0.1, .epochs = 10 )
.model_type |
This is where you will set your engine. It uses
|
.recipe_list |
You must supply a list of recipes. list(rec_1, rec_2, ...) |
.non_seasonal_ar |
The order of the non-seasonal auto-regressive (AR) terms. Often denoted "p" in pdq-notation. |
.seasonal_ar |
The order of the seasonal auto-regressive (SAR) terms. Often denoted "P" in PDQ-notation. |
|
An integer for the number of units in the hidden model. |
|
.num_networks |
Number of networks to fit with different random starting weights. These are then averaged when producing forecasts. |
.penalty |
A non-negative numeric value for the amount of weight decay. |
.epochs |
An integer for the number of training iterations. |
This function expects to take in the recipes that you want to use in the modeling process. This is an automated workflow process. There are sensible defaults set for the model specification, but if you choose you can set them yourself if you have a good understanding of what they should be. The mode is set to "regression".
This uses the following engines:
modeltime::nnetar_reg() nnetar_reg() is a way to generate a specification
of an NNETAR model before fitting and allows the model to be created using
different packages. Currently the only package is forecast.
"nnetar"
Returns a workflowsets object.
Steven P. Sanderson II, MPH
https://workflowsets.tidymodels.org/
https://business-science.github.io/modeltime/reference/nnetar_reg.html
Other Auto Workflowsets:
ts_wfs_arima_boost(),
ts_wfs_auto_arima(),
ts_wfs_ets_reg(),
ts_wfs_lin_reg(),
ts_wfs_mars(),
ts_wfs_prophet_reg(),
ts_wfs_svm_poly(),
ts_wfs_svm_rbf(),
ts_wfs_xgboost()
suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_nnetar_reg("nnetar", rec_objs) wf_setssuppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_nnetar_reg("nnetar", rec_objs) wf_sets
This function is used to quickly create a workflowsets object.
ts_wfs_prophet_reg( .model_type = "all_engines", .recipe_list, .growth = NULL, .changepoint_num = 25, .changepoint_range = 0.8, .seasonality_yearly = "auto", .seasonality_weekly = "auto", .seasonality_daily = "auto", .season = "additive", .prior_scale_changepoints = 25, .prior_scale_seasonality = 1, .prior_scale_holidays = 1, .logistic_cap = NULL, .logistic_floor = NULL, .trees = 50, .min_n = 10, .tree_depth = 5, .learn_rate = 0.01, .loss_reduction = NULL, .stop_iter = NULL )ts_wfs_prophet_reg( .model_type = "all_engines", .recipe_list, .growth = NULL, .changepoint_num = 25, .changepoint_range = 0.8, .seasonality_yearly = "auto", .seasonality_weekly = "auto", .seasonality_daily = "auto", .season = "additive", .prior_scale_changepoints = 25, .prior_scale_seasonality = 1, .prior_scale_holidays = 1, .logistic_cap = NULL, .logistic_floor = NULL, .trees = 50, .min_n = 10, .tree_depth = 5, .learn_rate = 0.01, .loss_reduction = NULL, .stop_iter = NULL )
.model_type |
This is where you will set your engine. It uses
|
.recipe_list |
You must supply a list of recipes. list(rec_1, rec_2, ...) |
.growth |
String 'linear' or 'logistic' to specify a linear or logistic trend. |
.changepoint_num |
Number of potential changepoints to include for modeling trend. |
.changepoint_range |
Adjusts the flexibility of the trend component by limiting to a percentage of data before the end of the time series. 0.80 means that a changepoint cannot exist after the first 80% of the data. |
.seasonality_yearly |
One of "auto", TRUE or FALSE. Set to FALSE for |
.seasonality_weekly |
One of "auto", TRUE or FALSE. Toggles on/off a
seasonal component that models week-over-week seasonality. Set to FALSE for |
.seasonality_daily |
One of "auto", TRUE or FALSE. Toggles on/off a
seasonal componet that models day-over-day seasonality. Set to FALSE for |
.season |
'additive' (default) or 'multiplicative'. |
.prior_scale_changepoints |
Parameter modulating the flexibility of the automatic changepoint selection. Large values will allow many changepoints, small values will allow few changepoints. |
.prior_scale_seasonality |
Parameter modulating the strength of the seasonality model. Larger values allow the model to fit larger seasonal fluctuations, smaller values dampen the seasonality. |
.prior_scale_holidays |
Parameter modulating the strength of the holiday components model, unless overridden in the holidays input. |
.logistic_cap |
When growth is logistic, the upper-bound for "saturation". |
.logistic_floor |
When growth is logistic, the lower-bound for "saturation" |
.trees |
An integer for the number of trees contained in the ensemble. |
.min_n |
An integer for the minimum number of data points in a node that is required for the node to be split further. |
.tree_depth |
An integer for the maximum depth of the tree (i.e. number of splits) (specific engines only). |
.learn_rate |
A number for the rate at which the boosting algorithm adapts from iteration-to-iteration (specific engines only). |
.loss_reduction |
A number for the reduction in the loss function required to split further (specific engines only). |
.stop_iter |
The number of iterations without improvement before stopping (xgboost only). |
This function expects to take in the recipes that you want to use in
the modeling process. This is an automated workflow process. There are sensible
defaults set for the prophet and prophet_xgboost model specification,
but if you choose you can set them yourself if you have a good understanding
of what they should be.
Returns a workflowsets object.
Steven P. Sanderson II, MPH
https://workflowsets.tidymodels.org/(workflowsets)
https://business-science.github.io/modeltime/reference/prophet_reg.html
https://business-science.github.io/modeltime/reference/prophet_boost.html
Other Auto Workflowsets:
ts_wfs_arima_boost(),
ts_wfs_auto_arima(),
ts_wfs_ets_reg(),
ts_wfs_lin_reg(),
ts_wfs_mars(),
ts_wfs_nnetar_reg(),
ts_wfs_svm_poly(),
ts_wfs_svm_rbf(),
ts_wfs_xgboost()
suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_prophet_reg("all_engines", rec_objs) wf_setssuppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_prophet_reg("all_engines", rec_objs) wf_sets
This function is used to quickly create a workflowsets object.
ts_wfs_svm_poly( .model_type = "kernlab", .recipe_list, .cost = 1, .degree = 1, .scale_factor = 1, .margin = 0.1 )ts_wfs_svm_poly( .model_type = "kernlab", .recipe_list, .cost = 1, .degree = 1, .scale_factor = 1, .margin = 0.1 )
.model_type |
This is where you will set your engine. It uses
|
.recipe_list |
You must supply a list of recipes. list(rec_1, rec_2, ...) |
.cost |
A positive number for the cose of predicting a sample within or on the wrong side of the margin. |
.degree |
A positive number for polynomial degree. |
.scale_factor |
A positive number for the polynomial scaling factor. |
.margin |
A positive number for the epsilon in the SVM insensitive loss function (regression only.) |
This function expects to take in the recipes that you want to use in the modeling process. This is an automated workflow process. There are sensible defaults set for the model specification, but if you choose you can set them yourself if you have a good understanding of what they should be. The mode is set to "regression".
This only uses the option set_engine("kernlab") and therefore the .model_type
is not needed. The parameter is kept because it is possible in the future that
this could change, and it keeps with the framework of how other functions
are written.
parsnip::svm_poly() svm_poly() defines a support vector machine model.
For classification, the model tries to maximize the width of the margin
between classes. For regression, the model optimizes a robust loss function
that is only affected by very large model residuals.
This SVM model uses a nonlinear function, specifically a polynomial function, to create the decision boundary or regression line.
Returns a workflowsets object.
Steven P. Sanderson II, MPH
https://workflowsets.tidymodels.org/
https://parsnip.tidymodels.org/reference/svm_poly.html
Other Auto Workflowsets:
ts_wfs_arima_boost(),
ts_wfs_auto_arima(),
ts_wfs_ets_reg(),
ts_wfs_lin_reg(),
ts_wfs_mars(),
ts_wfs_nnetar_reg(),
ts_wfs_prophet_reg(),
ts_wfs_svm_rbf(),
ts_wfs_xgboost()
suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_svm_poly("kernlab", rec_objs) wf_setssuppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_svm_poly("kernlab", rec_objs) wf_sets
This function is used to quickly create a workflowsets object.
ts_wfs_svm_rbf( .model_type = "kernlab", .recipe_list, .cost = 1, .rbf_sigma = 0.01, .margin = 0.1 )ts_wfs_svm_rbf( .model_type = "kernlab", .recipe_list, .cost = 1, .rbf_sigma = 0.01, .margin = 0.1 )
.model_type |
This is where you will set your engine. It uses
|
.recipe_list |
You must supply a list of recipes. list(rec_1, rec_2, ...) |
.cost |
A positive number for the cost of predicting a sample within or on the wrong side of the margin. |
.rbf_sigma |
A positive number for the radial basis function. |
.margin |
A positive number for the epsilon in the SVM insensitive loss function (regression only). |
This function expects to take in the recipes that you want to use in the modeling process. This is an automated workflow process. There are sensible defaults set for the model specification, but if you choose you can set them yourself if you have a good understanding of what they should be. The mode is set to "regression".
This only uses the option set_engine("kernlab") and therefore the .model_type
is not needed. The parameter is kept because it is possible in the future that
this could change, and it keeps with the framework of how other functions
are written.
parsnip::svm_rbf() svm_rbf() defines a support vector machine model.
For classification, the model tries to maximize the width of the margin
between classes. For regression, the model optimizes a robust loss function
that is only affected by very large model residuals.
This SVM model uses a nonlinear function, specifically a polynomial function, to create the decision boundary or regression line.
Returns a workflowsets object.
Steven P. Sanderson II, MPH
https://workflowsets.tidymodels.org/
https://parsnip.tidymodels.org/reference/svm_rbf.html
Other Auto Workflowsets:
ts_wfs_arima_boost(),
ts_wfs_auto_arima(),
ts_wfs_ets_reg(),
ts_wfs_lin_reg(),
ts_wfs_mars(),
ts_wfs_nnetar_reg(),
ts_wfs_prophet_reg(),
ts_wfs_svm_poly(),
ts_wfs_xgboost()
suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_svm_rbf("kernlab", rec_objs) wf_setssuppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_svm_rbf("kernlab", rec_objs) wf_sets
This function is used to quickly create a workflowsets object.
ts_wfs_xgboost( .model_type = "xgboost", .recipe_list, .trees = 15L, .min_n = 1L, .tree_depth = 6L, .learn_rate = 0.3, .loss_reduction = 0, .sample_size = 1, .stop_iter = Inf )ts_wfs_xgboost( .model_type = "xgboost", .recipe_list, .trees = 15L, .min_n = 1L, .tree_depth = 6L, .learn_rate = 0.3, .loss_reduction = 0, .sample_size = 1, .stop_iter = Inf )
.model_type |
This is where you will set your engine. It uses parsnip::boost_tree under the hood and can take one of the following:
|
.recipe_list |
You must supply a list of recipes. list(rec_1, rec_2, ...) |
.trees |
The number of trees (type: integer, default: 15L) |
.min_n |
Minimal Node Size (type: integer, default: 1L) |
.tree_depth |
Tree Depth (type: integer, default: 6L) |
.learn_rate |
Learning Rate (type: double, default: 0.3) |
.loss_reduction |
Minimum Loss Reduction (type: double, default: 0.0) |
.sample_size |
Proportion Observations Sampled (type: double, default: 1.0) |
.stop_iter |
The number of ierations Before Stopping (type: integer, default: Inf) |
This function expects to take in the recipes that you want to use in the modeling process. This is an automated workflow process. There are sensible defaults set for the model specification, but if you choose you can set them yourself if you have a good understanding of what they should be. The mode is set to "regression".
This only uses the option set_engine("xgboost") and therefore the .model_type
is not needed. The parameter is kept because it is possible in the future that
this could change, and it keeps with the framework of how other functions
are written.
parsnip::boost_tree() xgboost::xgb.train() creates a series of decision trees
forming an ensemble. Each tree depends on the results of previous trees.
All trees in the ensemble are combined to produce a final prediction.
Returns a workflowsets object.
Steven P. Sanderson II, MPH
https://workflowsets.tidymodels.org/
https://parsnip.tidymodels.org/reference/details_boost_tree_xgboost.html
https://arxiv.org/abs/1603.02754
Other Auto Workflowsets:
ts_wfs_arima_boost(),
ts_wfs_auto_arima(),
ts_wfs_ets_reg(),
ts_wfs_lin_reg(),
ts_wfs_mars(),
ts_wfs_nnetar_reg(),
ts_wfs_prophet_reg(),
ts_wfs_svm_poly(),
ts_wfs_svm_rbf()
suppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_xgboost("xgboost", rec_objs) wf_setssuppressPackageStartupMessages(library(modeltime)) suppressPackageStartupMessages(library(timetk)) suppressPackageStartupMessages(library(dplyr)) suppressPackageStartupMessages(library(rsample)) data <- AirPassengers %>% ts_to_tbl() %>% select(-index) splits <- time_series_split( data , date_col , assess = 12 , skip = 3 , cumulative = TRUE ) rec_objs <- ts_auto_recipe( .data = training(splits) , .date_col = date_col , .pred_col = value ) wf_sets <- ts_wfs_xgboost("xgboost", rec_objs) wf_sets
This function attempts to make a non-stationary time series stationary by applying differencing with a logarithmic transformation. It iteratively increases the differencing order until stationarity is achieved or informs the user if the transformation is not possible.
util_difflog_ts(.time_series)util_difflog_ts(.time_series)
.time_series |
A time series object to be made stationary. |
The function calculates the frequency of the input time series using the stats::frequency function
and checks if the minimum value of the time series is greater than 0. It then applies differencing
with a logarithmic transformation incrementally until the Augmented Dickey-Fuller test indicates
stationarity (p-value < 0.05) or until the differencing order reaches the frequency of the data.
If differencing with a logarithmic transformation successfully makes the time series stationary, it returns the stationary time series and related information as a list with the following elements:
stationary_ts: The stationary time series after the transformation.
ndiffs: The order of differencing applied to make it stationary.
adf_stats: Augmented Dickey-Fuller test statistics on the stationary time series.
trans_type: Transformation type, which is "diff_log" in this case.
ret: TRUE to indicate a successful transformation.
If the data either had a minimum value less than or equal to 0 or requires more differencing than its frequency allows, it informs the user and suggests trying double differencing with a logarithmic transformation.
If the time series is already stationary or the differencing with a logarithmic transformation is successful,
Steven P. Sanderson II, MPH
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
# Example 1: Using a time series dataset util_difflog_ts(AirPassengers) # Example 2: Using a different time series dataset util_difflog_ts(BJsales)$ret# Example 1: Using a time series dataset util_difflog_ts(AirPassengers) # Example 2: Using a different time series dataset util_difflog_ts(BJsales)$ret
This function attempts to make a non-stationary time series stationary by applying double differencing. It iteratively increases the differencing order until stationarity is achieved.
util_doublediff_ts(.time_series)util_doublediff_ts(.time_series)
.time_series |
A time series object to be made stationary. |
The function calculates the frequency of the input time series using the stats::frequency function.
It then applies double differencing incrementally until the Augmented Dickey-Fuller test indicates
stationarity (p-value < 0.05) or until the differencing order reaches the frequency of the data.
If double differencing successfully makes the time series stationary, it returns the stationary time series and related information as a list with the following elements:
stationary_ts: The stationary time series after double differencing.
ndiffs: The order of differencing applied to make it stationary.
adf_stats: Augmented Dickey-Fuller test statistics on the stationary time series.
trans_type: Transformation type, which is "double_diff" in this case.
ret: TRUE to indicate a successful transformation.
If the data requires more double differencing than its frequency allows, it informs the user and suggests trying differencing with the natural logarithm instead.
If the time series is already stationary or the double differencing is successful, it returns a list as described in the details section. If additional differencing is required, it informs the user and returns a list with ret set to FALSE, suggesting trying differencing with the natural logarithm.
Steven P. Sanderson II, MPH
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doubledifflog_ts(),
util_log_ts(),
util_singlediff_ts()
# Example 1: Using a time series dataset util_doublediff_ts(AirPassengers) # Example 2: Using a different time series dataset util_doublediff_ts(BJsales)$ret# Example 1: Using a time series dataset util_doublediff_ts(AirPassengers) # Example 2: Using a different time series dataset util_doublediff_ts(BJsales)$ret
This function attempts to make a non-stationary time series stationary by applying double differencing with a logarithmic transformation. It iteratively increases the differencing order until stationarity is achieved or informs the user if the transformation is not possible.
util_doubledifflog_ts(.time_series)util_doubledifflog_ts(.time_series)
.time_series |
A time series object to be made stationary. |
The function calculates the frequency of the input time series using the stats::frequency function
and checks if the minimum value of the time series is greater than 0. It then applies double differencing
with a logarithmic transformation incrementally until the Augmented Dickey-Fuller test indicates
stationarity (p-value < 0.05) or until the differencing order reaches the frequency of the data.
If double differencing with a logarithmic transformation successfully makes the time series stationary, it returns the stationary time series and related information as a list with the following elements:
stationary_ts: The stationary time series after the transformation.
ndiffs: The order of differencing applied to make it stationary.
adf_stats: Augmented Dickey-Fuller test statistics on the stationary time series.
trans_type: Transformation type, which is "double_diff_log" in this case.
ret: TRUE to indicate a successful transformation.
If the data either had a minimum value less than or equal to 0 or requires more differencing than its frequency allows, it informs the user that the data could not be stationarized.
If the time series is already stationary or the double differencing with a logarithmic transformation is successful, it returns a list as described in the details section. If the transformation is not possible, it informs the user and returns a list with ret set to FALSE, indicating that the data could not be stationarized.
Steven P. Sanderson II, MPH
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_log_ts(),
util_singlediff_ts()
# Example 1: Using a time series dataset util_doubledifflog_ts(AirPassengers) # Example 2: Using a different time series dataset util_doubledifflog_ts(BJsales)$ret# Example 1: Using a time series dataset util_doubledifflog_ts(AirPassengers) # Example 2: Using a different time series dataset util_doubledifflog_ts(BJsales)$ret
This function attempts to make a non-stationary time series stationary by applying a logarithmic transformation. If successful, it returns the stationary time series. If the transformation fails, it informs the user.
util_log_ts(.time_series)util_log_ts(.time_series)
.time_series |
A time series object to be made stationary. |
This function checks if the minimum value of the input time series is greater than or equal to zero. If yes, it performs the Augmented Dickey-Fuller test on the logarithm of the time series. If the p-value of the test is less than 0.05, it concludes that the logarithmic transformation made the time series stationary and returns the result as a list with the following elements:
stationary_ts: The stationary time series after the logarithmic transformation.
ndiffs: Not applicable in this case, marked as NA.
adf_stats: Augmented Dickey-Fuller test statistics on the stationary time series.
trans_type: Transformation type, which is "log" in this case.
ret: TRUE to indicate a successful transformation.
If the minimum value of the time series is less than or equal to 0 or if the logarithmic transformation doesn't make the time series stationary, it informs the user and returns a list with ret set to FALSE.
If the time series is already stationary or the logarithmic transformation is successful, it returns a list as described in the details section. If the transformation fails, it returns a list with ret set to FALSE.
Steven P. Sanderson II, MPH
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_singlediff_ts()
# Example 1: Using a time series dataset util_log_ts(AirPassengers) # Example 2: Using a different time series dataset util_log_ts(BJsales.lead)$ret# Example 1: Using a time series dataset util_log_ts(AirPassengers) # Example 2: Using a different time series dataset util_log_ts(BJsales.lead)$ret
This function attempts to make a non-stationary time series stationary by applying single differencing. It iteratively increases the differencing order until stationarity is achieved.
util_singlediff_ts(.time_series)util_singlediff_ts(.time_series)
.time_series |
A time series object to be made stationary. |
The function calculates the frequency of the input time series using the stats::frequency function.
It then applies single differencing incrementally until the Augmented Dickey-Fuller test indicates
stationarity (p-value < 0.05) or until the differencing order reaches the frequency of the data.
If single differencing successfully makes the time series stationary, it returns the stationary time series and related information as a list with the following elements:
stationary_ts: The stationary time series after differencing.
ndiffs: The order of differencing applied to make it stationary.
adf_stats: Augmented Dickey-Fuller test statistics on the stationary time series.
trans_type: Transformation type, which is "diff" in this case.
ret: TRUE to indicate a successful transformation.
If the data requires more single differencing than its frequency allows, it informs the user and returns a list with ret set to FALSE, indicating that double differencing may be needed.
If the time series is already stationary or the single differencing is successful, it returns a list as described in the details section. If additional differencing is required, it informs the user and returns a list with ret set to FALSE.
Steven P. Sanderson II, MPH
Other Utility:
auto_stationarize(),
calibrate_and_plot(),
internal_ts_backward_event_tbl(),
internal_ts_both_event_tbl(),
internal_ts_forward_event_tbl(),
model_extraction_helper(),
ts_get_date_columns(),
ts_info_tbl(),
ts_is_date_class(),
ts_lag_correlation(),
ts_model_auto_tune(),
ts_model_compare(),
ts_model_rank_tbl(),
ts_model_spec_tune_template(),
ts_qq_plot(),
ts_scedacity_scatter_plot(),
ts_to_tbl(),
util_difflog_ts(),
util_doublediff_ts(),
util_doubledifflog_ts(),
util_log_ts()
# Example 1: Using a time series dataset util_singlediff_ts(AirPassengers) # Example 2: Using a different time series dataset util_singlediff_ts(BJsales)$ret# Example 1: Using a time series dataset util_singlediff_ts(AirPassengers) # Example 2: Using a different time series dataset util_singlediff_ts(BJsales)$ret