Compute features on an input time series dataset

Usage

calculate_features(
  data,
  feature_set = c("catch22", "feasts", "tsfeatures", "kats", "tsfresh", "tsfel", "hctsa",
    "quantiles", "moments", "fftquantiles", "fft"),
  features = NULL,
  catch24 = FALSE,
  tsfresh_cleanup = FALSE,
  use_compengine = FALSE,
  seed = 123,
  z_score = FALSE,
  squared,
  n_jobs = 0,
  warn = TRUE,
  fft_phase = FALSE
)

Arguments

data: tbl_ts containing the time series data
feature_set: character or vector of character denoting the set of time-series features to calculate. Can be one of "catch22", "feasts", "tsfeatures", "tsfresh", "tsfel", "kats", "hctsa", "quantiles", "moments", "fftquantiles", and or "fft"
features: named list containing a set of user-supplied functions to calculate on data. Each function should take a single argument which is the time series. Defaults to NULL for no manually-specified features. Each list entry must have a name as calculate_features looks for these to name the features. If you don't want to use the existing feature sets and only compute those passed to features, set feature_set = NULL
catch24: Boolean specifying whether to compute catch24 in addition to catch22 if catch22 is one of the feature sets selected. Defaults to FALSE
tsfresh_cleanup: Boolean specifying whether to use the in-built tsfresh relevant feature filter or not. Defaults to FALSE. The filter is supervised, so it requires a class label as the second key variable of data. As it uses the class labels of every time series, evaluating the selected features with a classifier on the same data (e.g., with theftdlc::classify) gives optimistically biased results
use_compengine: Boolean specifying whether to use the "compengine" features in tsfeatures. Defaults to FALSE to provide immense computational efficiency benefits
seed: integer denoting a fixed number for R's random number generator to ensure reproducibility. Defaults to 123. The caller's random number generator state is restored on exit
z_score: Boolean specifying whether to z-score the time-series before computing features. Defaults to FALSE
squared: Deprecated and ignored. The "fft" and "fftquantiles" feature sets are now relative spectral power in fixed frequency bands (see fft_features), so there is no magnitude/squared-magnitude choice
n_jobs: integer denoting the number of parallel processes to use. Defaults to 0 for no parallelisation
warn: Boolean specifying whether to produce warnings from feature set packages. Defaults to TRUE
fft_phase: Boolean specifying whether the "fft" and "fftquantiles" feature sets also include each frequency band's phase, as its cosine and sine (see fft_features). Only meaningful when the series share a time origin, e.g., event-locked trials. Defaults to FALSE

Value

object of class feature_calculations that contains the summary statistics for each feature

Author

Trent Henderson

Examples

features <- calculate_features(data = simData,
  feature_set = "catch22")
#> Running computations for catch22...