Write create_trends_ensemble function

[lshandross]

Write function that generates predictions for the trends ensemble

create_trends_ensemble(component_variations, target_ts, reference_date, horizons, quantile_levels, n_samples, return.base_forecasts = FALSE)

Wrapper function that fits all the user-specified variations of the component baseline model (for all locations provided in target_ts) for the given reference date and ensembles them. Returns the trends ensemble forecasts in a hubverse format, with the option to specify whether to also return a model_out_tbl of all baseline forecasts.

Inputs:

• component_variations a data frame where each row specifies a set of hyperparameters to use for a single baseline model fit, with the following columns and possible values for each:
  • transformation (character): "none" or "sqrt"
  • symmetric (boolean)
  • window_size (integer)
  • temporal_resolution (character): "daily" or "weekly"
• target_ts a data frame of target data in a time series format (contains columns time_index, location, and observation)
• reference_date date giving the reference date for the forecasts
• horizons a numeric vector of prediction horizons relative to the reference date, e.g. 0:3 or 1:4
• quantile_levels numeric vector of quantile levels to output. Set to NA if quantile outputs not requested
• n_samples integer of samples to output (and to predict). Set to NA if sample outputs not requested, in which case 100,000 are generated by default
• return.base_forecasts boolean specifying whether or not to return the component baseline models' forecasts in addition to the trends ensemble forecasts. If TRUE, a two-item list will be returned containing a labeled model_out_tbl of each