In krzjoa/torchts: Time series Models with torch

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

Meaning of formulae in torchts package slightly differs from the typical way we can understand this object in the most of regression models in R. It should not be a surprise - some formulae syntax innovations can be found, for instance, in other time series related packages such as fable and modeltime.

This is how we read a formula in the torchts:

1. Simple autoregressive model

$value \sim date$ Future values of value column are base on the previous values of this variable. For now, lags are typically specified in the model object, not in the formula itself (it may change in the future).

You should always specify a time-related variable as date, integer index etc. You are not obliged to specify value explicitly on the right side of the formula.

2. Explanatory and explained variables

$tmax_daily \sim tmin_daily + date$ Future tmax_daily values are based on the previous tmin_daily values. tmax_daily values are not taken into account as explanatory variable in this case.

3. Explanatory and explained variables(with autoregression)

$tmax_daily \sim tmax_daily + tmin_daily + date$ Future tmax_daily values are based on both the previous tmax_daily and tmin_daily values. It's an example of autoregression with external regressors.

4. Model with values from the future

$value \sim date + value + price + lead(price)$ Future values of the value variable are based on the previous values from this column as well as previous values from the price column and the future value from the price column.

In some cases, especially in the demand forecasting, there exist some variable that are known in advance (as future price or holidays). Sometimes we can treat some other forecasts (e.g. weather forecast) as a known values to feed the model with.

krzjoa/torchts documentation built on June 24, 2022, 5:30 a.m.