RemixAutoML (version 0.4.2)

OptimizeNNET: OptimizeNNET is a function that takes raw data and returns time series data

Description

OptimizeNNET is a function that takes raw data and returns the necessary time series data and objects for model building. It also fills any time gaps with zeros. Use this before you run any time series model functions.

Usage

OptimizeNNET(
  Output,
  Path = NULL,
  MetricSelection = "MAE",
  DataSetName = NULL,
  train = NULL,
  test = NULL,
  FullData = NULL,
  HoldOutPeriods = NULL,
  MinVal = NULL,
  TargetName = NULL,
  DateName = NULL,
  Lags = NULL,
  SeasonalLags = NULL,
  MaxFourierTerms = NULL,
  TrainValidateShare = NULL,
  MaxRunsWithoutNewWinner = 20,
  MaxNumberModels = NULL,
  MaxRunMinutes = NULL,
  FinalGrid = NULL
)

Arguments

Output

This is passed through as output from TimeSeriesDataPrepare() and passed through ParallelArima()

Path

Path to where you want the model and xregs saved. Leave NULL to not save.

MetricSelection

Select from "MSE", "MAE", or "MAPE"

DataSetName

This is the name of the data set passed through in parallel loop

train

Training data returned from TimeSeriesDataPrepare()

test

Test data returned from TimeSeriesDataPrepare()

FullData

Full series data for scoring and ensemble

HoldOutPeriods

Holdout periods returned from TimeSeriesDataPrepare()

MinVal

Minimum value of target variable returned from TimeSeriesDataPrepare()

TargetName

Target variable name returned from TimeSeriesDataPrepare()

DateName

Date variable name returned from TimeSeriesDataPrepare()

Lags

Max value of lag returned from TimeSeriesDataPrepare()

SeasonalLags

Max value of seasonal lags returned from TimeSeriesDataPrepare()

MaxFourierTerms

Max value of fourier pairs

TrainValidateShare

A two-element numeric vector. The first element is the weight applied to the training performance and the remainder is applied to the validation performance.

MaxRunsWithoutNewWinner

The number of runs without a new winner which if passed tells the function to stop

MaxNumberModels

The number of models you want to test.

MaxRunMinutes

Time

FinalGrid

If NULL, regular train optimization occurs. If the grid is supplied, final builds are conducted.

Value

Time series data sets to pass onto auto modeling functions

See Also

Other Time Series Helper: FinalBuildArfima(), FinalBuildArima(), FinalBuildETS(), FinalBuildNNET(), FinalBuildTBATS(), FinalBuildTSLM(), GenerateParameterGrids(), OptimizeArfima(), OptimizeArima(), OptimizeETS(), OptimizeTBATS(), OptimizeTSLM(), ParallelAutoARIMA(), ParallelAutoArfima(), ParallelAutoETS(), ParallelAutoNNET(), ParallelAutoTBATS(), ParallelAutoTSLM(), PredictArima(), RL_Performance(), Regular_Performance(), StackedTimeSeriesEnsembleForecast(), TimeSeriesDataPrepare(), WideTimeSeriesEnsembleForecast()

Examples

# NOT RUN {
Results <- OptimizeNNET(
  Output,
  Path = NULL,
  MetricSelection = "MAE",
  DataSetName = NULL,
  train = NULL,
  test = NULL,
  FullData = NULL,
  HoldOutPeriods = NULL,
  MinVal = NULL,
  TargetName = NULL,
  DateName = NULL,
  Lags = NULL,
  SeasonalLags = NULL,
  MaxFourierTerms = NULL,
  TrainValidateShare = NULL,
  MaxRunsWithoutNewWinner = 20,
  MaxNumberModels = 5,
  MaxRunMinutes = NULL,
  FinalGrid = NULL)
# }