Function reference • threedx

Fit Model and Forecast

Core functions to fit a model and then forecast a time series.

learn_weights(): Fit a 3DX model to a time series

predict(<threedx>): Draw forecast sample paths from a fitted 3DX model

Initialize Grid of Parameters

Helper functions to construct a grid of alpha parameters to be evaluated during model fitting, or use your own.

list_sampled_alphas(): Generate a data frame of possible alpha values to evaluate during training

list_edge_alphas(): Generate a data frame of possible alpha values to evaluate during training

Loss Functions

Loss functions to be used during model fitting, or use your own.

loss_mae(): Mean absolute error loss function

loss_mae_ignoring_bias(): Mean absolute error loss function ignoring bias

loss_mae_with_observation_weight(): Observation-weighted mean absolute error loss function

loss_rmse(): Root-mean-square error loss function

loss_rmse_ignoring_bias(): Root-mean-square error loss function ignoring bias

Innovation Methods

Methods of generating innovations to construct sample paths during prediction, or use your own.

draw_bootstrap(): Draw innovations by bootstrapping from unweighted residual errors

draw_bootstrap_weighted(): Draw innovations by bootstrapping from weighted residual errors

draw_bootstrap_zero_mean(): Draw innovations by bootstrapping from unweighted zero-mean residual errors

draw_normal_with_drift(): Draw i.i.d. innovations from a Normal distribution with non-zero mean

draw_normal_with_zero_mean(): Draw i.i.d. innovations from a Normal distribution with zero mean

Plot Model and Forecast

Custom functions to plot the fitted model and the forecast using ggplot2.

autoplot(<threedx>): Autoplot method for threedx objects

autoplot(<threedx_paths>): Autoplot method for threedx_paths objects

Generate Weights

Generate the weights that are the fundamental component of threedx models.

weights_exponential(): Derive exponential weights

weights_seasonal(): Derive seasonal exponential weights

weights_seasonal_decay(): Derive weights with seasonal exponential decay

weights_threedx(): Derive three-dimensional exponential (3DX) weights

Sample Path Postprocessing

Functions to postprocess samples during generation of forecast sample paths.

to_moment_matched_nbinom(): Postprocess to samples from a moment-matched negative-binomial distribution

to_non_negative_with_identical_mean(): Turn values non-negative while preserving their sample mean

Helper Functions

k_largest_weights_sum_to_less_than_p_percent(): Do the k largest weights sum up to less than p% of the total weights?

trade_loss_for_simplicity(): Allow for an increase in loss to find a simpler model

fitted(<threedx>): Extract fitted values from a threedx model

residuals(<threedx>): Extract residuals from a threedx model