Multilayer Perceptron

This topic contains the following sections:

Implementation
Code sample
See Also

This is a classical feed-forward multi-layer neural network. This network utilizes fully connected layers.

Comparing to the classical multi-layer perceptron this class allows construct model with more than one hidden layer. In the case of more than one hidden layer, the model training success probability is fairly low.

This model initializes layer weights with random uniform data and next utilizes the iRPROP+ for training.

The advantages and disadvantages of this network type are widely described in the literature. So, this class can be used mostly for other models benchmarking. See for example DeepNeuralNetwork class as an more sophisticated neural network type.

Implementation

Next most important methods and properties are featured in the class:

Method	Description
Train(Matrix, IntegerArray, Int32)	Train classifier with the specified topology.

Code sample

Copy

1var modelMlp = new MLP();
2var watch = new Stopwatch();
3watch.Reset();
4watch.Start();
5modelMlp.Train(trainObs, trainCl);
6watch.Stop();
7Estimate("MLP", watch.Elapsed, modelMlp, testObs, testCl, gridData);

Other Resources