Fast neural network for everyone

This article will demonstrate the ability to easily write your neural network in Java. In order not to reinvent the wheel, take the already well-developed library of the Fast Artificial Neural Network. Using neural networks in their Java projects is real. You can often hear accusations against Java regarding execution speed. Although the difference is not so great - you can learn more about this in the publication “C ++ vs. Performance. Java vs. PHP vs. Python. Test "in the forehead." We will use a wrapper around the FANN library.

Task

You need to write a system that can make decisions for a character who can meet one or more enemies. The system may be aware of:

• character's health percentage;
• the presence of a pistol;
• number of enemies.

The answer should be in the form of one of the actions:

• Attack;
• to run;
• hide (for a surprise attack);
• nothing to do.

For learning we will make a table of "lessons":

Health	pistol	The enemies	Act
50%	one	one	Attack
90%	one	2	Attack
80%	0	one	Attack
thirty%	one	one	Hide
60%	one	2	Hide
40%	0	one	Hide
90%	one	7	To run
60%	one	four	To run
ten%	0	one	To run
60%	one	0	Nothing
100%	0	0	Nothing

Training

The first thing to do is build and install libfann .
Then download fannj and jna .

Make a file that will contain a set of "lessons":

11 3 4 0.5 1 1 1 0 0 0 0.9 1 2 1 0 0 0 0.8 0 1 1 0 0 0 0.3 1 1 0 1 0 0 0.6 1 2 0 1 0 0 0.4 0 1 0 1 0 0 0.9 1 7 0 0 1 0 0.5 1 4 0 0 1 0 0.1 0 1 0 0 1 0 0.6 1 0 0 0 0 1 1.0 0 0 0 0 0 1 

Now we will train our ANN and save it to a file:
')

  public static void main(String[] args) { //        List<Layer> layerList = new ArrayList<Layer>(); layerList.add(Layer.create(3, ActivationFunction.FANN_SIGMOID_SYMMETRIC, 0.01f)); layerList.add(Layer.create(16, ActivationFunction.FANN_SIGMOID_SYMMETRIC, 0.01f)); layerList.add(Layer.create(4, ActivationFunction.FANN_SIGMOID_SYMMETRIC, 0.01f)); Fann fann = new Fann(layerList); //      Trainer trainer = new Trainer(fann); trainer.setTrainingAlgorithm(TrainingAlgorithm.FANN_TRAIN_RPROP); /*      ,     100000,    100      0.0001 */ trainer.train(new File("train.data").getAbsolutePath(), 100000, 100, 0.0001f); fann.save("ann"); } 

Explanation

Layer

ANN consists of layers of neurons. The first layer is the “receptor” neurons or input neurons. The last layer of neuron output. All the others are hidden layers. In our case, the first layer has 3 neurons:

health level (0.1-1.0);
the presence of weapons (1-there, 0-no);
number of enemies.

Fann

The object of the class Fann is the neural network, which is created on the basis of the layers created earlier.

Trainer

The class object trainer encapsulates the neural network learning algorithms transmitted during the creation of the trainer. After learning, do not forget to save it to a file.

Checking results

To test our training, we use the following code:

  public static void main(String[] args) { Fann fann = new Fann("ann"); float[][] tests = { {1.0f, 0, 1}, {0.9f, 1, 3}, {0.3f, 0, 8}, {1, 1, 8}, {0.1f, 0, 0}, }; for (float[] test:tests){ System.out.println(getAction(fann.run(test))); } } private static String getAction(float[] out){ int i = 0; for (int j = 1; j < 4; j++) { if(out[i]<out[j]){ i = j; } } switch (i){ case 0:return ""; case 1:return ""; case 2:return ""; case 3:return "  "; } return ""; } 

I got the following results:

Health	pistol	The enemies	Act
100%	Not	one	Attack
90%	there is	3	Hide
thirty%	not	eight	To run
100%	there is	eight	To run
ten%	Not	0	Nothing to do

I will be glad to hear constructive criticism.