JRNN

JRNN is an easy to use recurrent neural network (RNN) library implemented in Java.

Example Usage

// Load data
List<float[]> traindata = DataLoader.loadData("C:\\Users\\user\\Documents\\AI Projects\\JRNN\\traindata.txt", Seperator.COMMA);
		
// Create neural network
NeuralNetwork nn = new NeuralNetwork(0.3f, 5E-3f, 3000000, true, ErrorFunction.MSE);
nn.addLayer(new HiddenLayer(2, ActivationFunction.SIGMOID));
nn.addLayer(new OutputLayer(3, ActivationFunction.SIGMOID)); // Neuron count must be same with the input vector length
	
// Train and test
NetworkController nc = new NetworkController(nn);
nc.showIterations(20000);
nc.trainNetwork(traindata);	

// Do regression for 10 steps, starting from the last input of train data
nc.doRegression(10);

// OR

// Predict next 6 output for the given input sequence
List<float[]> inputSequence = new ArrayList<>();
inputSequence.add(new float[]{0, 0, 1});
inputSequence.add(new float[]{0, 1, 0});
inputSequence.add(new float[]{0, 1, 1});
inputSequence.add(new float[]{0, 0, 1});
inputSequence.add(new float[]{0, 1, 0});
inputSequence.add(new float[]{0, 1, 1});

nc.predictNext(inputSequence, 6);

Sample Output

======= Training Starts =======
Max Epochs        : 3000000
Max Error         : 0.005
Learning Rate     : 0.3
===============================

======== Training Ends ========
Epochs : 313
Error  : 0.004987748
===============================

======= Regression Starts =======
0	0	1	
0	1	0	
0	1	1	
0	0	1	
0	1	0	
0	1	1	
0	0	1	
0	1	0	
0	1	1	
0	1	1	
====== Regression Finished ======

Usage

// Load data
List<float[]> traindata = DataLoader.loadData("C:\\Users\\user\\Documents\\AI Projects\\JRNN\\traindata.txt", Seperator.COMMA);
		
// Create neural network
// Set learning rate, max error, input data type (true for binary) anderror function.
NeuralNetwork nn = new NeuralNetwork(0.3f, 5E-3f, 3000000, true, ErrorFunction.MSE); 

// Add layers. Minimum 1 hidden and 1 output layer is required.
nn.addLayer(new HiddenLayer(2, ActivationFunction.SIGMOID));
nn.addLayer(new OutputLayer(3, ActivationFunction.SIGMOID)); // Neuron count must be same with the input vector length
	
// Train and test
NetworkController nc = new NetworkController(nn);
nc.showIterations(20000); // Show iteration stats every 20000 iterations
nc.trainNetwork(traindata);	

// Do regression for 10 steps, starting from the last input of train data
nc.doRegression(10);

// OR

// Predict next 6 output for the given input sequence
List<float[]> inputSequence = new ArrayList<>();
inputSequence.add(new float[]{0, 0, 1});
inputSequence.add(new float[]{0, 1, 0});
inputSequence.add(new float[]{0, 1, 1});
inputSequence.add(new float[]{0, 0, 1});
inputSequence.add(new float[]{0, 1, 0});
inputSequence.add(new float[]{0, 1, 1});

nc.predictNext(inputSequence, 6);

Data

The training data and test data should be in the form of:

1, 0, 1, 1
1, 0, 0, 1
1, 0, 1, 1
0, 0, 0, 1
1, 0, 1, 0
...
...
...

Note: Data columns can be seperated by space or tab too.

Contribution

This library is still under development. You can open issues for the bugs you found. Also you can send pull requests for enhancements/bug fixes.

License

See more at LICENSE page.