time_splits

Generates train/test splits given a vector of dates and other options

A suite of auxiliary functions that enhance time series estimation and forecasting, including a robust anomaly detection routine based on Chen and Liu (1993) <doi:10.2307/2290724> (imported and wrapped from the 'tsoutliers' package), utilities for managing calendar and time conversions, performance metrics to assess both point forecasts and distributional predictions, advanced simulation by allowing the generation of time series components—such as trend, seasonal, ARMA, irregular, and anomalies—in a modular fashion based on the innovations form of the state space model and a number of transformation methods including Box-Cox, Logit, 'Softplus-Logit' and Sigmoid.

Alexios Ghalanos

tsaux

Time Series Forecasting Auxiliary Functions

Alexios Galanos

time_splits function

<dl><dt>x</dt>
<dd>a vector of timestamps (POSIXct) or dates (Date) in the dataset</dd>
<dt>start</dt>
<dd>starting date (first estimation/train date)</dd>
<dt>test_length</dt>
<dd>type of calendar period to split on</dd>
<dt>by</dt>
<dd>every how many periods to split on</dd>
<dt>window_size</dt>
<dd>the size of the training set (for moving window). If NULL
will use an expanding window.</dd>
<dt>calendar_end</dt>
<dd>an optional function to use for the period ending split,
such as <code>calendar_eow</code>, applied to x. This should be greater in
frequency than the underlying frequency of x (i.e. do not use calendar_eow on
monthly indices). This overwrites the use of window_size.</dd>
<dt>complete_index</dt>
<dd>whether to return the full indices for train and test else
just the start and end indices.</dd>
<dt>...</dt>
<dd>any additional parameters passed to the calendar_end function. For
example, the &#8220;day&#8221; argument when using the calendar_eow function.</dd></dl>

Arguments

Author

Generate Train/Test Splits — time_splits

<dl>

<dt>x</dt>
<dd>a vector of timestamps (POSIXct) or dates (Date) in the dataset</dd>


<dt>start</dt>
<dd>starting date (first estimation/train date)</dd>


<dt>test_length</dt>
<dd>type of calendar period to split on</dd>


<dt>by</dt>
<dd>every how many periods to split on</dd>


<dt>window_size</dt>
<dd>the size of the training set (for moving window). If NULL
will use an expanding window.</dd>


<dt>calendar_end</dt>
<dd>an optional function to use for the period ending split,
such as <code>calendar_eow</code>, applied to x. This should be greater in
frequency than the underlying frequency of x (i.e. do not use calendar_eow on
monthly indices). This overwrites the use of window_size.</dd>


<dt>complete_index</dt>
<dd>whether to return the full indices for train and test else
just the start and end indices.</dd>


<dt>...</dt>
<dd>any additional parameters passed to the calendar_end function. For
example, the &#8220;day&#8221; argument when using the calendar_eow function.</dd>

</dl>

time_splits: Generate Train/Test Splits

Description

Usage

Value

Arguments

Author