FcMLP_BP: Forecasting with Multilayer Perceptron Feedforward Network...

In Mthrun/TSAT: Time Series Analysis Tools (TSAT)

Forecasting with Multilayer Perceptron Feedforward Network with Backpropagation

Description

Foracasting by multilayer perceptron feedforward network with resilient backpropagation with weight backtracking.

Usage

FcMLP_BP(Response, SplitAt, Predictor1,

Predictor2, Time, HiddenLayers = 4, Threshold = 0.05, Repetitions = 10,

ErrorFunction = "sse", PlotEvaluation = FALSE, PlotIt = FALSE)

Arguments

Response	[1:n] vector with an value of each time j in [1,n]
SplitAt	Index of row k where the DataVec is divided into test and train data. If not given n is used
Predictor1	[1:n] vector with an value of each time j in [1,n]
Predictor2	[1:n] vector with an value of each time j in [1,n]
Time	[1:n] character vector of Time in the length of data
HiddenLayers	Number of hidden layers, see neuralnet
Threshold	Threshold for the partial derivatives of ErrorFunction , please see neuralnet
Repetitions	Number of repetitions for the neural network's training.
ErrorFunction	Differentiable function that is used for the calculation of the error, string alternatives are 'ce' and 'sse', please see neuralnet
PlotEvaluation	Plot output of training for neuralnet
PlotFuture	A logical determining whether or not to plot the forecast in comparison to the validation set.

Details

This functions trains a MLP/BP network and then forecasts a new sample of data using the trained ANN. Training and Testdata are splitted up from the arguments using ForecastHorizont. Testing data is only used to compare against the forecast.

Seems to be good for longterm forcasting. Short term forcasting does not work well.

Value

list with

Forecast	[k:n], the forecast, of the time interval [k,n] which was not used in the model
TestSet	[k:n,1], the part of Response not used in the model
TrainData	[1:k,1:3], the part of Response not used in the model
TestDataOutput	[output of compute
Accuracy	ME, RMSE, MAE, MPE, MAPE of training and test dataset in a matrix
Model	Output of neuralnet

Note

This version requires currently two predictors.

Author(s)

Michael Thrun

References

Riedmiller M. and Braun H.: A direct adaptive method for faster backpropagation learning: The RPROP algorithm. Proceedings of the IEEE International Conference on Neural Networks (ICNN), pages 586-591. San Francisco, 1993.

See Also

neuralnet, compute

Examples

#no open acces data available with two predcitors

Mthrun/TSAT documentation built on Feb. 5, 2024, 11:15 p.m.