Merge pull request #193 from BrainJS/192-time-series-baseline

Add Time Step Recurrent Neural Network

README.md

@@ -21,6 +21,7 @@
 - [Training](#training)
     + [Data format](#data-format)
       + [For training with NeuralNetwork](#for-training-with-neuralnetwork)
+      + [For training with `RNNTimeStep`, `LSTMTimeStep` and `GRUTimeStep`](#for-training-with-rnntimestep-lstmtimestep-and-gputimestep)
       + [For training with `RNN`, `LSTM` and `GRU`](#for-training-with-rnn-lstm-and-gpu)
     + [Training Options](#training-options)
     + [Async Training](#async-training)
@@ -128,6 +129,38 @@ net.train([{input: { r: 0.03, g: 0.7 }, output: { black: 1 }},
 
 var output = net.run({ r: 1, g: 0.4, b: 0 });  // { white: 0.81, black: 0.18 }
 ```
+
+#### For training with `RNNTimeStep`, `LSTMTimeStep` and `GRUTimeStep`
+Eeach training pattern can either:
+* Be an array of numbers
+* Be an array of arrays of numbers
+
+Example using an array of numbers:
+```javascript
+var net = new brain.recurrent.LSTMTimeStep();
+
+net.train([
+  1,
+  2,
+  3,
+]);
+
+var output = net.run([1, 2]);  // 3
+```
+
+Example using an array of arrays of numbers:
+```javascript
+var net = new brain.recurrent.LSTMTimeStep();
+
+net.train([
+  [1, 3],
+  [2, 2],
+  [3, 1],
+]);
+
+var output = net.run([[1, 3], [2, 2]]);  // [3, 1]
+```
+
 #### For training with `RNN`, `LSTM` and `GRU`
 Each training pattern can either:
 * Be an array of values
@@ -321,13 +354,17 @@ Likely example see: [simple letter detection](./examples/which-letter-simple.js)
 
 # Neural Network Types
 * [`brain.NeuralNetwork`](src/neural-network.js) - [Feedforward Neural Network](https://en.wikipedia.org/wiki/Feedforward_neural_network) with backpropagation
+* [`brain.recurrent.RNNTimeStep`](src/recurrent/rnn-time-step.js) - [Time Step Recurrent Neural Network or "RNN"](https://en.wikipedia.org/wiki/Recurrent_neural_network)
+* [`brain.recurrent.LSTMTimeStep`](src/recurrent/lstm-time-step.js) - [Time Step Long Short Term Memory Neural Network or "LSTM"](https://en.wikipedia.org/wiki/Long_short-term_memory)
+* [`brain.recurrent.GRUTimeStep`](src/recurrent/gru-time-step.js) - [Time Step Gated Recurrent Unit or "GRU"](https://en.wikipedia.org/wiki/Gated_recurrent_unit)
 * [`brain.recurrent.RNN`](src/recurrent/rnn.js) - [Recurrent Neural Network or "RNN"](https://en.wikipedia.org/wiki/Recurrent_neural_network)
 * [`brain.recurrent.LSTM`](src/recurrent/lstm.js) - [Long Short Term Memory Neural Network or "LSTM"](https://en.wikipedia.org/wiki/Long_short-term_memory)
 * [`brain.recurrent.GRU`](src/recurrent/gru.js) - [Gated Recurrent Unit or "GRU"](https://en.wikipedia.org/wiki/Gated_recurrent_unit)
 
 ### Why different Neural Network Types?
 Different neural nets do different things well. For example:
 * A Feedforward Neural Network can classify simple things very well, but it has no memory of previous actions and has infinite variation of results.
+* A Time Step Recurrent Neural Network _remembers_, and can predict future values.
 * A Recurrent Neural Network _remembers_, and has a finite set of results.
 
 # Get Involved!

browser.js

@@ -6,7 +6,7 @@
  *   license: MIT (http://opensource.org/licenses/MIT)
  *   author: Heather Arthur <[email protected]>
  *   homepage: https://github.com/brainjs/brain.js#readme
- *   version: 1.1.3
+ *   version: 1.2.0
  *
  * acorn:
  *   license: MIT (http://opensource.org/licenses/MIT)
@@ -2463,6 +2463,7 @@ var Equation = function () {
     _classCallCheck(this, Equation);
 
     this.inputRow = 0;
+    this.inputValue = null;
     this.states = [];
   }
 
@@ -2611,6 +2612,25 @@ var Equation = function () {
       return product;
     }
 
+    /**
+     * copy a matrix
+     * @param {Matrix} input
+     * @returns {Matrix}
+     */
+
+  }, {
+    key: 'input',
+    value: function input(_input) {
+      var self = this;
+      this.states.push({
+        product: _input,
+        forwardFn: function forwardFn() {
+          _input.weights = self.inputValue;
+        }
+      });
+      return _input;
+    }
+
     /**
      * connects a matrix via a row
      * @param {Matrix} m
@@ -2722,6 +2742,27 @@ var Equation = function () {
      * @output {Matrix}
      */
 
+  }, {
+    key: 'runInput',
+    value: function runInput(inputValue) {
+      this.inputValue = inputValue;
+      var state = void 0;
+      for (var i = 0, max = this.states.length; i < max; i++) {
+        state = this.states[i];
+        if (!state.hasOwnProperty('forwardFn')) {
+          continue;
+        }
+        state.forwardFn(state.product, state.left, state.right);
+      }
+
+      return state.product;
+    }
+
+    /**
+     * @patam {Number} [rowIndex]
+     * @output {Matrix}
+     */
+
   }, {
     key: 'runBackpropagate',
     value: function runBackpropagate() {
@@ -3476,7 +3517,7 @@ var RNN = function () {
 
     _classCallCheck(this, RNN);
 
-    var defaults = RNN.defaults;
+    var defaults = this.constructor.defaults;
 
     for (var p in defaults) {
       if (!defaults.hasOwnProperty(p)) continue;
@@ -3868,7 +3909,7 @@ var RNN = function () {
     value: function train(data) {
       var options = arguments.length > 1 && arguments[1] !== undefined ? arguments[1] : {};
 
-      options = Object.assign({}, RNN.trainDefaults, options);
+      options = Object.assign({}, this.constructor.trainDefaults, options);
       var iterations = options.iterations;
       var errorThresh = options.errorThresh;
       var log = options.log === true ? console.log : options.log;
@@ -3935,7 +3976,7 @@ var RNN = function () {
   }, {
     key: 'toJSON',
     value: function toJSON() {
-      var defaults = RNN.defaults;
+      var defaults = this.constructor.defaults;
       var model = this.model;
       var options = {};
       for (var p in defaults) {
@@ -3966,7 +4007,7 @@ var RNN = function () {
     key: 'fromJSON',
     value: function fromJSON(json) {
       this.json = json;
-      var defaults = RNN.defaults;
+      var defaults = this.constructor.defaults;
       var model = this.model;
       var options = json.options;
       var allMatrices = model.allMatrices;