Taso

Java implementation of Multilayer Perceptron

Implementation

Modules

It basically has three main modules MLP, Gradient Function and Activation Function.

MLP: It is the main driver class that helps create the model to fit and predict.

Activation Function: This java implementation currently supports these following Activation Functions

• Relu
• LeakyRelu
• Tanh
• Sigmoid

Gradient Function: This java implementation currently supports these following Gradient Functions

• AdaGrad
• Adam
• Momentum
• RMSProp
• SGD

Environment Required

• Windows, Linux or MacOS
• Java 1.8

USING THIS PROJECT

After all the pre-requisites mentioned above is installed on the machine then follow the steps provided below.

1. Clone this repository https://github.com/sreetamparida/Taso.git and open MLP.java.
2. Provide the path to Training and Tesing dataset as shown below.

File inputFile = new File("cancer.csv");
File targetFile = new File("cancer_target.csv");

3. Create the model with your specified Gradient Function, Activation Function and Hidden Layers.
4. Specify the Gradient Function while creating the model object by adding the Gradient Function object as parameter.
5. For adding Hidden Layers use addHiddenLayer() function.
6. Specify the Activation Function for the layer by adding the Activation Function object as parameter.
7. Then perform the fit() operation by providing number of epochs.
8. Use the predict() function to specify the target data.

An implementation of a model with

• Gradient Function - AdaGrad
• Activation Function - TanH in the hidden layers and Relu at the output layer
• Number of Hidden Layers - 4

        MLP model = new MLP(new AdaGrad());
        model.addInputLayer(noAttr);
        model.addHiddenLayer(noAttr,new Tanh());
        model.addHiddenLayer(noAttr+1,new Tanh());
        model.addHiddenLayer(noAttr+1,new Tanh());
        model.addHiddenLayer(noAttr,new Tanh());
        model.addOutputLayer(1,new Relu());
        model.generateModel();
        model.fit(input,target,5000);

9. Execute the comand javac MLP.java to get your desired predictions.