ganGenerativeData (version 2.1.6)

gdTrain: Train a generative model for a data source

Description

Read a data source from a file, train a generative model that generates normalized generative data for the data source in iterative training steps, write trained generative model and normalized generated data to a file in binary format. When a higher number of iterations is used the distribution of generated data will get closer to that of the data source. When a name of an existing generative model file is passed training will be continued.

Usage

gdTrain(
  generativeModelFileName,
  generativeDataFileName,
  dataSourceFileName,
  columnIndices,
  trainParameters = gdTrainParameters(numberOfTrainingIterations = 10000,
    numberOfInitializationIterations = 1500, numberOfHiddenLayerUnits = 1024,
    learningRate = 7e-05, dropout = 0.05)
)

Value

None

Arguments

generativeModelFileName

Name of generative model file

generativeDataFileName

Name of generative data file. When name is NULL or empty string generated data will not be written to a file.

dataSourceFileName

Name of data source file

columnIndices

Vector of two column indices that are used to plot two-dimensional projections of normalized generated generative data and data source for a training step. Indices refer to indices of active columns of data source. Plotting can be disabled by passing NULL or an empty vector.

trainParameters

Generative model training parameters, see function gdTrainParameters().

Examples

if (FALSE) {
trainParameters <- gdTrainParameters(numberOfTrainingIterations = 10000)
gdTrain("gm.bin", "gd.bin", "ds.bin", c(1, 2), trainParameters)}