Leaky relu alpha option

src/neural-network.js

@@ -32,6 +32,7 @@ export default class NeuralNetwork {
 
   static get defaults() {
     return {
+      leakyReluAlpha: 0.01,
       binaryThresh: 0.5,
       hiddenLayers: [3],     // array of ints for the sizes of the hidden layers in the network
       activation: 'sigmoid'  // Supported activation types ['sigmoid', 'relu', 'leaky-relu', 'tanh']
@@ -249,7 +250,7 @@ export default class NeuralNetwork {
 
   _runInputLeakyRelu(input) {
     this.outputs[0] = input;  // set output state of input layer
-
+    let alpha = this.leakyReluAlpha;
     let output = null;
     for (let layer = 1; layer <= this.outputLayer; layer++) {
       for (let node = 0; node < this.sizes[layer]; node++) {
@@ -260,7 +261,7 @@ export default class NeuralNetwork {
           sum += weights[k] * input[k];
         }
         //leaky relu
-        this.outputs[layer][node] = (sum < 0 ? 0 : 0.01 * sum);
+        this.outputs[layer][node] = (sum < 0 ? 0 : alpha * sum);
       }
       output = input = this.outputs[layer];
     }
@@ -557,6 +558,7 @@ export default class NeuralNetwork {
    * @param target
    */
   _calculateDeltasLeakyRelu(target) {
+    let alpha = this.leakyReluAlpha;
     for (let layer = this.outputLayer; layer >= 0; layer--) {
       for (let node = 0; node < this.sizes[layer]; node++) {
         let output = this.outputs[layer][node];
@@ -572,7 +574,7 @@ export default class NeuralNetwork {
           }
         }
         this.errors[layer][node] = error;
-        this.deltas[layer][node] = output > 0 ? error : 0.01 * error;
+        this.deltas[layer][node] = output > 0 ? error : alpha * error;
       }
     }
   }
@@ -933,6 +935,7 @@ export default class NeuralNetwork {
    */
   toFunction() {
     const activation = this.activation;
+    const leakyReluAlpha = this.leakyReluAlpha;
     let needsVar = false;
     function nodeHandle(layers, layerNumber, nodeKey) {
       if (layerNumber === 0) {
@@ -962,7 +965,7 @@ export default class NeuralNetwork {
         }
         case 'leaky-relu': {
           needsVar = true;
-          return `((v=${result.join('')})<0?0:0.01*v)`;
+          return `((v=${result.join('')})<0?0:${leakyReluAlpha}*v)`;
         }
         case 'tanh':
           return `Math.tanh(${result.join('')})`;
@@ -988,4 +991,4 @@ export default class NeuralNetwork {
 
     return new Function('input', `${ needsVar ? 'var v;' : '' }return ${result};`);
   }
-}
+}

test/base/options.js

@@ -108,4 +108,10 @@ describe ('neural network constructor values', () => {
     var net = new brain.NeuralNetwork(opts);
     assert.equal(opts.activation, net.activation, `activation => ${net.activation} but should be ${opts.activation}`);
   })
+
+  it('leakyReluAlpha should be settable in the constructor', () => {
+    let opts = { leakyReluAlpha: 0.1337 };
+    var net = new brain.NeuralNetwork(opts);
+    assert.equal(opts.leakyReluAlpha, net.leakyReluAlpha, `leakyReluAlpha => ${net.leakyReluAlpha} but should be ${opts.leakyReluAlpha}`);
+  })
 });