new dispose, cleanup and comments

examples/p5js/NeuralNetwork/NeuroEvolution/bird.js

@@ -27,10 +27,6 @@ class Bird {
  }
  }
 
- dispose() {
- this.brain.neuralNetwork.model.dispose();
- }
-
  show() {
  stroke(255);
  fill(255, 100);

examples/p5js/NeuralNetwork/NeuroEvolution/ga.js

@@ -4,11 +4,15 @@
 function nextGeneration() {
  console.log('next generation');
  calculateFitness();
+
+ // Create a new population
  for (let i = 0; i < TOTAL; i++) {
  birds[i] = reproduce();
  }
+
+ // Release all the memory
  for (let i = 0; i < TOTAL; i++) {
- savedBirds[i].dispose();
+ savedBirds[i].brain.dispose();
  }
  savedBirds = [];
 }

examples/p5js/NeuralNetwork/NeuroEvolution_testing/sketch.js

@@ -7,10 +7,12 @@ function setup() {
  }
  const nnA = ml5.neuralNetwork(options);
  const results1 = nnA.classifySync([0.2, 0.1, 0.5]);
+ console.log('----- Classification A ------');
  console.log(results1[0]);
-
+ 
  nnA.mutate(0.1);
  const results2 = nnA.classifySync([0.2, 0.1, 0.5]);
+ console.log('----- Default Mutated Classification A ------');
  console.log(results2[0]);
 
  // Mutating a neural network
@@ -19,21 +21,27 @@ function setup() {
  }
  nnA.mutate(0.1, customMutate);
  const results3 = nnA.classifySync([0.2, 0.1, 0.5]);
+ console.log('----- Custom Mutated Classification A ------');
  console.log(results3[0]);
 
- // const nnCopy = nnA.copy();
- // const results4 = nnCopy.classifySync([0.2, 0.1, 0.5]);
- // console.log(results4[0]);
+ const nnCopy = nnA.copy();
+ const results4 = nnCopy.classifySync([0.2, 0.1, 0.5]);
+ console.log('----- Identical Copy Classification A ------');
+ console.log(results4[0]);
 
- // const nnB = ml5.neuralNetwork(options);
- // const results5 = nnB.classifySync([0.2, 0.1, 0.5]);
- // console.log(results5[0]);
+ const nnB = ml5.neuralNetwork(options);
+ const results5 = nnB.classifySync([0.2, 0.1, 0.5]);
+ console.log('----- Classification B ------');
+ console.log(results5[0]);
 
- // const child = nnA.crossover(nnB);
- // const resultsA = nnA.classifySync([0.2, 0.1, 0.5]);
- // const resultsB = nnB.classifySync([0.2, 0.1, 0.5]);
- // const childResults = child.classifySync([0.2, 0.1, 0.5]);
- // console.log(resultsA[0]);
- // console.log(resultsB[0]);
- // console.log(childResults[0]);
+ const child = nnA.crossover(nnB);
+ const resultsA = nnA.classifySync([0.2, 0.1, 0.5]);
+ const resultsB = nnB.classifySync([0.2, 0.1, 0.5]);
+ const childResults = child.classifySync([0.2, 0.1, 0.5]);
+ console.log('----- Classification A ------');
+ console.log(resultsA[0]);
+ console.log('----- Classification B ------');
+ console.log(resultsB[0]);
+ console.log('----- Crossover AB classification ------');
+ console.log(childResults[0]);
 }

src/NeuralNetwork/NeuralNetwork.js

@@ -272,8 +272,20 @@ class NeuralNetwork {
  return this.model;
  }
 
+ /**
+ * dispose and release the memory for the model
+ */
+ dispose() {
+ this.model.dispose();
+ }
+
  // NeuroEvolution Functions
 
+ /**
+ * mutate the weights of a model
+ * @param {*} rate
+ * @param {*} mutateFunction
+ */ 
  mutate(rate, mutateFunction) {
  tf.tidy(() => {
  const weights = this.model.getWeights();
@@ -284,7 +296,7 @@ class NeuralNetwork {
  // TODO: Evaluate if this should be sync or not
  const values = tensor.dataSync().slice();
  for (let j = 0; j < values.length; j+=1) {
- if (Math.random() < rate) {
+ if (Math.random() < rate || 0.1) {
  if (mutateFunction) {
  values[j] = mutateFunction(values[j]);
  } else {
@@ -299,6 +311,10 @@ class NeuralNetwork {
  });
  }
 
+ /**
+ * create a new neural network with crossover
+ * @param {*} other
+ */
  crossover(other) {
  return tf.tidy(() => {
  const weightsA = this.model.getWeights();

src/NeuralNetwork/index.js

@@ -77,6 +77,9 @@ class DiyNeuralNetwork {
  this.save = this.save.bind(this);
  this.load = this.load.bind(this);
 
+ // release model
+ this.dispose = this.dispose.bind(this);
+
  // neuroevolution
  this.mutate = this.mutate.bind(this);
 
@@ -110,6 +113,9 @@ class DiyNeuralNetwork {
  }
  }
 
+ /**
+ * createLayersNoTraining
+ */
  createLayersNoTraining() {
  // Makes some sample data
  // TODO: Account for regression
@@ -123,6 +129,9 @@ class DiyNeuralNetwork {
  this.addDefaultLayers(this.options.task, this.neuralNetworkData.meta);
  }
 
+ /**
+ * copy
+ */
  copy() {
  const nnCopy = new DiyNeuralNetwork(this.options);
  return tf.tidy(() => {
@@ -778,7 +787,7 @@ class DiyNeuralNetwork {
  */
 
  /**
- * predict
+ * synchronous predict
  * @param {*} _input
  */
  predictSync(_input) {
@@ -804,7 +813,7 @@ class DiyNeuralNetwork {
  }
 
  /**
- * predict
+ * synchronous classify
  * @param {*} _input
  */
  classifySync(_input) {
@@ -830,7 +839,7 @@ class DiyNeuralNetwork {
  }
 
  /**
- * predict
+ * synchronous predict internal
  * @param {*} _input
  * @param {*} _cb
  */
@@ -950,7 +959,7 @@ class DiyNeuralNetwork {
  }
 
  /**
- * classify
+ * synchronous classify internal
  * @param {*} _input
  * @param {*} _cb
  */
@@ -1156,11 +1165,32 @@ class DiyNeuralNetwork {
  });
  }
 
- // NeuroEvolution functions
+ /**
+ * dispose and release memory for a model
+ */
+ dispose() {
+ this.neuralNetwork.dispose();
+ }
+
+ /**
+ * ////////////////////////////////////////////////////////////
+ * New methods for Neuro Evolution
+ * ////////////////////////////////////////////////////////////
+ */
+
+ /**
+ * mutate the weights of a model
+ * @param {*} rate
+ * @param {*} mutateFunction
+ */ 
  mutate(rate, mutateFunction) {
  this.neuralNetwork.mutate(rate, mutateFunction);
  }
 
+ /**
+ * create a new neural network with crossover
+ * @param {*} other
+ */ 
  crossover(other) {
  const nnCopy = this.copy();
  nnCopy.neuralNetwork.crossover(other.neuralNetwork);