feasts: Feature Extraction And Statistics for Time Series

feasts (Feature Extraction And Statistics for Time Series) is a collection of tools for the analysis of time series data. It works with tidy temporal data provided by the tsibble package to produce:

• Time series features
• Decompositions
• Statistical summaries
• Convenient visualisations

It is designed to integrate closely with the tidy forecasting workflow used in the fable package.

You can extract features and statistics across a large collection of time series using the features() function. This is useful for identifying unusual/extreme time series or finding clusters of similar behavior.

Example: Extracting STL features from retail turnover data:

as_retail %>%
  features(Turnover, feat_stl)

Extracted features can then be used for visualization (e.g., plotting feature strengths) or filtering to find extreme cases (e.g., the most or least seasonal series).

aus_retail %>%
  features(Turnover, feat_stl)

You can install the stable version of feasts from CRAN or the development version from GitHub.

Stable version (CRAN):

install.packages("feasts")

Development version (GitHub):

# install.packages("remotes")
remotes::install_github("tidyverts/feasts")

The package uses ggplot2 to provide customisable graphics for understanding time series patterns. Common tasks include:

• Seasonal plots: Use gg_season() to visualize seasonality.
• Subseries plots: Use gg_subseries() to visualize subseries patterns.
• Lag plots: Use gg_lag() to visualize lags.
• Autocorrelation plots: Use ACF() combined with autoplot() to visualize autocorrelation.

Example usage:

# Seasonal plot
aus_production %>% gg_season(Beer)

# Subseries plot
aus_production %>% gg_subseries(Beer)

# Lag plot
aus_production %>% filter(year(Quarter) > 1991) %>% gg_lag(Beer)

# Autocorrelation plot
aus_production %>% ACF(Beer) %>% autoplot()

feasts supports decomposing a time series into simpler components (such as trend, season, and remainder) using methods like Classical decomposition and STL decomposition.

To perform decomposition, use the model() function with a decomposition method (e.g., STL()), then extract the components using components().

# Perform STL decomposition
dcmp <- aus_production %>%
  model(STL(Beer ~ season(window = Inf)))

# Extract components
components(dcmp)

# Plot components
components(dcmp) %>% autoplot()