metatemp_train: Train Metalearning Models Trains a metalearning model by...

In robjhyndman/M4metalearning: Metalearning Tools For Time Series Forecasting

Description

Train Metalearning Models Trains a metalearning model by performing search on the hyperparameters

Usage

metatemp_train(train_dataset, eval_dataset, obj_fun = c("select",
  "combi:smax_abs", "combi:smax_sq", "combi:square"),
  filename = "meta_results.RData", verbose = FALSE, n.cores = 3)

train_selection_ensemble(data, errors, param = NULL)

predict_selection_ensemble(model, newdata)

summary_performance(predictions, dataset, print.summary = TRUE)

Arguments

train_dataset	A list with elements in the metalearning format. E.g. the output of a combination of process_forecast_dataset and generate_THA_feature_dataset x A time series object ts with the historical data. h The number of required forecasts. xx The number of required forecasts. THA_features The number of required forecasts. ff The number of required forecasts. errors The number of required forecasts.
eval_dataset	A list in the format of train_dataset, used for evaluating the error. Can be the same as train_dataset to show training error
obj_fun	The objective loss function that the metalearning minimizes
filename	Name of the file used for saving the metalearning process
verbose	Boolean indicating whether training progress messages may be printed
data	A matrix with the input features data (extracted from the series). One observation (the features from the original series) per row.
errors	A matrix with the errors produced by each of the forecasting methods. Each row is a vector with the errors of the forecasting methods.
model	The xgboost model
newdata	The feature matrix, one row per series
predictions	A NXM matrix with N the number of observations(time series) and M the number of methods.
dataset	The list with the meta information, if given additional details will be provided
print.summary	Boolean indicating wheter to print the information
labels	A numeric vector from 0 to (nclass -1) with the targe labels for classification.
errors	The NXM matrix with the erros per series and per method
labels	Integer vector The true labels of the would be classification problem. Possible values are from 0 to M-1

Value

model

The xgboost model found by the metalearning

eval_log

The log of hyperparametes tested with their produced errors

Functions

• train_selection_ensemble: Train a method-selecting ensemble that minimizes forecasting error

• predict_selection_ensemble: Produces predictions probabilities for the selection ensemble.

• summary_performance: Analysis of the predictions

robjhyndman/M4metalearning documentation built on May 21, 2019, 12:22 p.m.