deepNeuralNetwork.training: Train the Deep Neural Network to obtain a regression model.

In OscarGVelasco/DeepNeuralNetworks4R: Implementation of Deep Neural Networks in R language

Description

'deepNeuralNetwork.training()' will train the Deep Neural Network built by [build.dnn] function. A dataset must be passed in order to train the DNN, x= columns of the data , y = columns of known results that can be predicted using x

Usage

deepNeuralNetwork.training(x, y, model = NULL, traindata = data,
  testdata = NULL, iterations = 2000, minError = 0.01,
  maxError = 0.05, lr = 0.01, reg = 0.001, display = 100,
  random.seed = 1, standarization = NULL, savePlotIteration = FALSE)

Arguments

x	Numeric Vector. Specifies the columns of the data that will be used as the input variables to predict y value.
y	NUmeric, or Vector. Specifies the column (or columns) of the observed variable, the output that can be predicted using x.
model	An object of class DeepNNModel containing an initialized model.
traindata	Matrix or Data Frame. The actual data with explanatory variables (x) and observed results (y) that will be used for training.
testdata	Matrix or Data Frame (optional). If provided, the algorithm will check for overfitting using the testdata as a input.
iterations	Numeric. Number of training iterations.
minError	Numeric.
maxError	Numeric. Maximun Error permited on training data to chose best model.
lr	Numeric. Initial Learning rate. During training it is automatically adjusted using adagrad.
reg	Numeric. Regularization rate.
display	Numeric. Show training results each [N] iterations.
random.seed	Numeric. deprecated.
standarization	Character or list. Standarization method to be use. If a list of features (corresponding to rows in the input matrix) is supplied, a standarization feature-z-score is done. "r"=robust median z-score. "s"=standar mean z-score.
savePlotIteration	Boolean. If TRUE saves the plot that is shown every [N] iterations specified by [display].

Details

This function trains a deep neural network previously created and initialized by [deepNeuralNetwork.build].

Value

A DeepNNModel object with the trained regression model with the parameters as specified by the user.

Examples

dnn.model <- deepNeuralNetwork.build(x=c(1,2,4,5),y=3, outputNeurons = 1,
                                     HidenLayerNeurons = c(30,10,3),traindata=data,
                                     random.seed = 1, drawDNN = 0)

OscarGVelasco/DeepNeuralNetworks4R documentation built on Jan. 24, 2021, 12:42 a.m.