From 6511aef7a970c0106c6255e8e2298f1d9c1f1a86 Mon Sep 17 00:00:00 2001
From: Leon Chen <leon@md.ai>
Date: Tue, 13 Sep 2016 09:51:48 -0400
Subject: [PATCH] implement Embedding layer, with tests

---
 index.html                          |   3 +
 notebooks/embedding/Embedding.ipynb | 248 ++++++++++++++++++++++++++++
 src/layers/embeddings/Embedding.js  |  49 ++++++
 src/layers/embeddings/index.js      |   5 +
 src/layers/index.js                 |   1 +
 test/embeddings/Embedding.js        |  51 ++++++
 test/embeddings/data_Embedding.js   |  58 +++++++
 7 files changed, 415 insertions(+)
 create mode 100644 notebooks/embedding/Embedding.ipynb
 create mode 100644 src/layers/embeddings/Embedding.js
 create mode 100644 src/layers/embeddings/index.js
 create mode 100644 test/embeddings/Embedding.js
 create mode 100644 test/embeddings/data_Embedding.js

diff --git a/index.html b/index.html
index 20b66a3..fb7260b 100644
--- a/index.html
+++ b/index.html
@@ -97,6 +97,9 @@
   <script src="/test/normalization/data_BatchNormalization.js"></script>
   <script src="/test/normalization/BatchNormalization.js"></script>
 
+  <script src="/test/embeddings/data_Embedding.js"></script>
+  <script src="/test/embeddings/Embedding.js"></script>
+
   <script>
     // mocha.checkLeaks();
     mocha.globals(['jQuery']);
diff --git a/notebooks/embedding/Embedding.ipynb b/notebooks/embedding/Embedding.ipynb
new file mode 100644
index 0000000..c06e262
--- /dev/null
+++ b/notebooks/embedding/Embedding.ipynb
@@ -0,0 +1,248 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Using TensorFlow backend.\n"
+     ]
+    }
+   ],
+   "source": [
+    "import numpy as np\n",
+    "from keras.models import Model\n",
+    "from keras.layers import Input\n",
+    "from keras.layers.embeddings import Embedding\n",
+    "from keras import backend as K"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": [
+    "def format_decimal(arr, places=6):\n",
+    "    return [round(x * 10**places) / 10**places for x in arr]"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Embedding"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "**[embeddings.Embedding.0] input_dim 5, output_dim 3, input_length=7, mask_zero=False, dropout=0.**"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 16,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "W shape: (5, 3)\n",
+      "W: [0.123907, 0.924736, 0.347325, 0.711111, -0.843888, -0.295, -0.702738, -0.469642, -0.024244, 0.962892, -0.00486, -0.636172, -0.318551, 0.114577, 0.674536]\n",
+      "\n",
+      "in shape: (7,)\n",
+      "in: [1, 1, 1, 2, 0, 0, 3]\n",
+      "out shape: (7, 3)\n",
+      "out: [0.711111, -0.843888, -0.295, 0.711111, -0.843888, -0.295, 0.711111, -0.843888, -0.295, -0.702738, -0.469642, -0.024244, 0.123907, 0.924736, 0.347325, 0.123907, 0.924736, 0.347325, 0.962892, -0.00486, -0.636172]\n"
+     ]
+    }
+   ],
+   "source": [
+    "input_dim = 5\n",
+    "output_dim = 3\n",
+    "input_length = 7\n",
+    "data_in_shape = (input_length,)\n",
+    "emb = Embedding(input_dim, output_dim, input_length=input_length, mask_zero=False, dropout=0.)\n",
+    "\n",
+    "layer_0 = Input(shape=data_in_shape)\n",
+    "layer_1 = emb(layer_0)\n",
+    "model = Model(input=layer_0, output=layer_1)\n",
+    "\n",
+    "# set weights to random (use seed for reproducibility)\n",
+    "weights = []\n",
+    "for i, w in enumerate(model.get_weights()):\n",
+    "    np.random.seed(1200 + i)\n",
+    "    weights.append(2 * np.random.random(w.shape) - 1)\n",
+    "model.set_weights(weights)\n",
+    "print('W shape:', weights[0].shape)\n",
+    "print('W:', format_decimal(weights[0].ravel().tolist()))\n",
+    "\n",
+    "data_in = np.random.randint(0, input_dim - 1, data_in_shape)\n",
+    "print('')\n",
+    "print('in shape:', data_in_shape)\n",
+    "print('in:', data_in.ravel().tolist())\n",
+    "result = model.predict(np.array([data_in]))\n",
+    "print('out shape:', result[0].shape)\n",
+    "print('out:', format_decimal(result[0].ravel().tolist()))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "**[embeddings.Embedding.1] input_dim 20, output_dim 5, input_length=10, mask_zero=True, dropout=0.**"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 18,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "W shape: (20, 5)\n",
+      "W: [-0.383154, 0.451816, 0.335889, 0.941217, 0.853643, -0.668432, 0.074737, -0.716747, -0.816627, 0.866955, -0.792696, 0.164527, 0.452684, 0.628771, -0.254798, -0.612431, 0.523026, -0.591887, -0.064898, 0.053117, -0.867557, 0.504834, 0.679206, -0.556031, 0.36241, 0.709619, 0.433085, 0.934247, 0.971585, -0.162598, 0.335228, -0.050332, 0.277192, 0.254304, -0.351008, 0.564886, 0.928261, -0.292494, -0.169661, 0.444558, 0.417255, 0.922794, -0.33368, 0.50358, -0.971797, -0.328566, -0.691183, -0.490174, 0.510752, -0.597686, 0.531579, 0.618621, -0.66854, 0.905422, 0.862149, 0.764005, -0.881062, 0.956836, 0.507665, -0.423755, -0.036345, -0.009073, -0.612606, 0.135378, -0.748721, -0.996749, -0.556871, 0.103616, -0.929907, 0.337223, 0.171248, -0.19617, -0.093472, -0.148244, -0.785821, 0.920643, -0.685849, -0.962657, -0.460922, -0.169631, -0.516211, -0.804437, -0.39244, 0.665189, 0.635366, 0.501686, 0.997166, -0.210946, 0.606477, -0.474869, 0.713225, 0.382837, -0.566294, 0.239002, 0.784362, -0.693148, -0.893049, 0.509413, -0.576682, -0.620533]\n",
+      "\n",
+      "in shape: (10,)\n",
+      "in: [8, 6, 14, 8, 11, 17, 10, 18, 0, 13]\n",
+      "out shape: (10, 5)\n",
+      "out: [0.417255, 0.922794, -0.33368, 0.50358, -0.971797, 0.335228, -0.050332, 0.277192, 0.254304, -0.351008, 0.171248, -0.19617, -0.093472, -0.148244, -0.785821, 0.417255, 0.922794, -0.33368, 0.50358, -0.971797, 0.764005, -0.881062, 0.956836, 0.507665, -0.423755, 0.501686, 0.997166, -0.210946, 0.606477, -0.474869, 0.531579, 0.618621, -0.66854, 0.905422, 0.862149, 0.713225, 0.382837, -0.566294, 0.239002, 0.784362, -0.383154, 0.451816, 0.335889, 0.941217, 0.853643, -0.996749, -0.556871, 0.103616, -0.929907, 0.337223]\n"
+     ]
+    }
+   ],
+   "source": [
+    "input_dim = 20\n",
+    "output_dim = 5\n",
+    "input_length = 10\n",
+    "data_in_shape = (input_length,)\n",
+    "emb = Embedding(input_dim, output_dim, input_length=input_length, mask_zero=True, dropout=0.)\n",
+    "\n",
+    "layer_0 = Input(shape=data_in_shape)\n",
+    "layer_1 = emb(layer_0)\n",
+    "model = Model(input=layer_0, output=layer_1)\n",
+    "\n",
+    "# set weights to random (use seed for reproducibility)\n",
+    "weights = []\n",
+    "for i, w in enumerate(model.get_weights()):\n",
+    "    np.random.seed(1210 + i)\n",
+    "    weights.append(2 * np.random.random(w.shape) - 1)\n",
+    "model.set_weights(weights)\n",
+    "print('W shape:', weights[0].shape)\n",
+    "print('W:', format_decimal(weights[0].ravel().tolist()))\n",
+    "\n",
+    "data_in = np.random.randint(0, input_dim - 1, data_in_shape)\n",
+    "print('')\n",
+    "print('in shape:', data_in_shape)\n",
+    "print('in:', data_in.ravel().tolist())\n",
+    "result = model.predict(np.array([data_in]))\n",
+    "print('out shape:', result[0].shape)\n",
+    "print('out:', format_decimal(result[0].ravel().tolist()))"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "**[embeddings.Embedding.2] input_dim 33, output_dim 2, input_length=5, mask_zero=False, dropout=0.5**"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 19,
+   "metadata": {
+    "collapsed": false
+   },
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "W shape: (33, 2)\n",
+      "W: [-0.616375, -0.01389, 0.09463, 0.74616, 0.903976, 0.46756, 0.810735, -0.196574, 0.312078, 0.931824, -0.658021, -0.305244, 0.16033, -0.01051, -0.669953, -0.377897, 0.842174, -0.54385, -0.321611, 0.19085, 0.963801, 0.015549, 0.436929, -0.069782, 0.088399, -0.63811, 0.250636, 0.205862, 0.01425, 0.436179, -0.020659, 0.844792, -0.635104, -0.550835, 0.334066, -0.703519, -0.232756, 0.493263, -0.097717, -0.695123, 0.712031, 0.722442, -0.432591, 0.711508, 0.074558, 0.216026, 0.255217, 0.406698, -0.983917, -0.519696, -0.199758, 0.134159, -0.177873, -0.491386, 0.456317, 0.961498, 0.387117, 0.162469, 0.907715, -0.253443, -0.717262, -0.557943, 0.154291, 0.706519, 0.53983, -0.907997]\n",
+      "\n",
+      "in shape: (5,)\n",
+      "in: [28, 9, 19, 30, 18]\n",
+      "out shape: (5, 2)\n",
+      "out: [0.387117, 0.162469, -0.321611, 0.19085, -0.097717, -0.695123, -0.717262, -0.557943, -0.232756, 0.493263]\n"
+     ]
+    }
+   ],
+   "source": [
+    "input_dim = 33\n",
+    "output_dim = 2\n",
+    "input_length = 5\n",
+    "data_in_shape = (input_length,)\n",
+    "emb = Embedding(input_dim, output_dim, input_length=input_length, mask_zero=False, dropout=0.5)\n",
+    "\n",
+    "layer_0 = Input(shape=data_in_shape)\n",
+    "layer_1 = emb(layer_0)\n",
+    "model = Model(input=layer_0, output=layer_1)\n",
+    "\n",
+    "# set weights to random (use seed for reproducibility)\n",
+    "weights = []\n",
+    "for i, w in enumerate(model.get_weights()):\n",
+    "    np.random.seed(1220 + i)\n",
+    "    weights.append(2 * np.random.random(w.shape) - 1)\n",
+    "model.set_weights(weights)\n",
+    "print('W shape:', weights[0].shape)\n",
+    "print('W:', format_decimal(weights[0].ravel().tolist()))\n",
+    "\n",
+    "data_in = np.random.randint(0, input_dim - 1, data_in_shape)\n",
+    "print('')\n",
+    "print('in shape:', data_in_shape)\n",
+    "print('in:', data_in.ravel().tolist())\n",
+    "result = model.predict(np.array([data_in]))\n",
+    "print('out shape:', result[0].shape)\n",
+    "print('out:', format_decimal(result[0].ravel().tolist()))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "collapsed": true
+   },
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "anaconda-cloud": {},
+  "kernelspec": {
+   "display_name": "Python [default]",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.5.2"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 1
+}
diff --git a/src/layers/embeddings/Embedding.js b/src/layers/embeddings/Embedding.js
new file mode 100644
index 0000000..940fd45
--- /dev/null
+++ b/src/layers/embeddings/Embedding.js
@@ -0,0 +1,49 @@
+import Layer from '../../engine/Layer'
+import Tensor from '../../Tensor'
+import ops from 'ndarray-ops'
+
+/**
+ * Embedding layer class
+ */
+export default class Embedding extends Layer {
+  /**
+   * Creates a Embedding layer
+   */
+  constructor (inputDim, outputDim, attrs = {}) {
+    super(attrs)
+    const {
+      inputLength = 0,
+      maskZero = false,
+      dropout = 0.0
+    } = attrs
+
+    this.inputDim = inputDim
+    this.outputDim = outputDim
+    this.inputLength = inputLength
+
+    // maskZero will be important for subsequence layers
+    this.maskZero = maskZero
+
+    // relevant only during training phase
+    this.dropout = dropout
+
+    // Layer weights specification
+    this.params = ['W']
+  }
+
+  /**
+   * Method for layer computational logic
+   * @param {Tensor} x
+   * @returns {Tensor} x
+   */
+  call (x) {
+    let y = new Tensor([], [x.tensor.shape[0], this.weights.W.tensor.shape[1]])
+
+    for (let i = 0, len = x.tensor.shape[0]; i < len; i++) {
+      ops.assign(y.tensor.pick(i, null), this.weights.W.tensor.pick(x.tensor.get(i), null))
+    }
+
+    x.tensor = y.tensor
+    return x
+  }
+}
diff --git a/src/layers/embeddings/index.js b/src/layers/embeddings/index.js
new file mode 100644
index 0000000..83a5f7f
--- /dev/null
+++ b/src/layers/embeddings/index.js
@@ -0,0 +1,5 @@
+import Embedding from './Embedding'
+
+export {
+  Embedding
+}
diff --git a/src/layers/index.js b/src/layers/index.js
index acbbfc8..4c92d7a 100644
--- a/src/layers/index.js
+++ b/src/layers/index.js
@@ -3,3 +3,4 @@ export * from './core'
 export * from './convolutional'
 export * from './pooling'
 export * from './normalization'
+export * from './embeddings'
diff --git a/test/embeddings/Embedding.js b/test/embeddings/Embedding.js
new file mode 100644
index 0000000..2fbb9ca
--- /dev/null
+++ b/test/embeddings/Embedding.js
@@ -0,0 +1,51 @@
+/* eslint-env browser, mocha */
+
+describe.only('embeddings layer: Embedding', function () {
+  const assert = chai.assert
+  const styles = testGlobals.styles
+  const logTime = testGlobals.logTime
+  const stringifyCondensed = testGlobals.stringifyCondensed
+  const approxEquals = KerasJS.testUtils.approxEquals
+  const layers = KerasJS.layers
+
+  const testParams = [
+    {
+      inputShape: [7],
+      attrs: { inputDim: 5, outputDim: 3, inputLength: 7, maskZero: false, dropout: 0 }
+    },
+    {
+      inputShape: [10],
+      attrs: { inputDim: 20, outputDim: 5, inputLength: 10, maskZero: true, dropout: 0 }
+    },
+    {
+      inputShape: [5],
+      attrs: { inputDim: 33, outputDim: 2, inputLength: 5, maskZero: false, dropout: 0.5 }
+    }
+  ]
+
+  before(function () {
+    console.log('\n%cembeddings layer: Embedding', styles.h1)
+  })
+
+  testParams.forEach(({ attrs }, i) => {
+    const key = `embeddings.Embedding.${i}`
+    const title = `[${key}] test: inputDim='${attrs.inputDim}', outputDim=${attrs.outputDim}, inputLength=${attrs.inputLength}, maskZero=${attrs.maskZero}, dropout=${attrs.dropout}`
+
+    it(title, function () {
+      console.log(`\n%c${title}`, styles.h3)
+      let testLayer = new layers.Embedding(attrs)
+      testLayer.setWeights(TEST_DATA[key].weights.map(w => new KerasJS.Tensor(w.data, w.shape)))
+      let t = new KerasJS.Tensor(TEST_DATA[key].input.data, TEST_DATA[key].input.shape)
+      console.log('%cin', styles.h4, stringifyCondensed(t.tensor))
+      const startTime = performance.now()
+      t = testLayer.call(t)
+      const endTime = performance.now()
+      console.log('%cout', styles.h4, stringifyCondensed(t.tensor))
+      logTime(startTime, endTime)
+      const dataExpected = new Float32Array(TEST_DATA[key].expected.data)
+      const shapeExpected = TEST_DATA[key].expected.shape
+      assert.deepEqual(t.tensor.shape, shapeExpected)
+      assert.isTrue(approxEquals(t.tensor, dataExpected))
+    })
+  })
+})
diff --git a/test/embeddings/data_Embedding.js b/test/embeddings/data_Embedding.js
new file mode 100644
index 0000000..57bb917
--- /dev/null
+++ b/test/embeddings/data_Embedding.js
@@ -0,0 +1,58 @@
+// TEST DATA
+// Keyed by mocha test ID
+// Python code for generating test data can be found in the matching jupyter notebook in folder `notebooks/`.
+
+(function () {
+  var DATA = {
+    'embeddings.Embedding.0': {
+      input: {
+        data: [1, 1, 1, 2, 0, 0, 3],
+        shape: [7]
+      },
+      weights: [
+        {
+          data: [0.123907, 0.924736, 0.347325, 0.711111, -0.843888, -0.295, -0.702738, -0.469642, -0.024244, 0.962892, -0.00486, -0.636172, -0.318551, 0.114577, 0.674536],
+          shape: [5, 3]
+        }
+      ],
+      expected: {
+        data: [0.711111, -0.843888, -0.295, 0.711111, -0.843888, -0.295, 0.711111, -0.843888, -0.295, -0.702738, -0.469642, -0.024244, 0.123907, 0.924736, 0.347325, 0.123907, 0.924736, 0.347325, 0.962892, -0.00486, -0.636172],
+        shape: [7, 3]
+      }
+    },
+    'embeddings.Embedding.1': {
+      input: {
+        data: [8, 6, 14, 8, 11, 17, 10, 18, 0, 13],
+        shape: [10]
+      },
+      weights: [
+        {
+          data: [-0.383154, 0.451816, 0.335889, 0.941217, 0.853643, -0.668432, 0.074737, -0.716747, -0.816627, 0.866955, -0.792696, 0.164527, 0.452684, 0.628771, -0.254798, -0.612431, 0.523026, -0.591887, -0.064898, 0.053117, -0.867557, 0.504834, 0.679206, -0.556031, 0.36241, 0.709619, 0.433085, 0.934247, 0.971585, -0.162598, 0.335228, -0.050332, 0.277192, 0.254304, -0.351008, 0.564886, 0.928261, -0.292494, -0.169661, 0.444558, 0.417255, 0.922794, -0.33368, 0.50358, -0.971797, -0.328566, -0.691183, -0.490174, 0.510752, -0.597686, 0.531579, 0.618621, -0.66854, 0.905422, 0.862149, 0.764005, -0.881062, 0.956836, 0.507665, -0.423755, -0.036345, -0.009073, -0.612606, 0.135378, -0.748721, -0.996749, -0.556871, 0.103616, -0.929907, 0.337223, 0.171248, -0.19617, -0.093472, -0.148244, -0.785821, 0.920643, -0.685849, -0.962657, -0.460922, -0.169631, -0.516211, -0.804437, -0.39244, 0.665189, 0.635366, 0.501686, 0.997166, -0.210946, 0.606477, -0.474869, 0.713225, 0.382837, -0.566294, 0.239002, 0.784362, -0.693148, -0.893049, 0.509413, -0.576682, -0.620533],
+          shape: [20, 5]
+        }
+      ],
+      expected: {
+        data: [0.417255, 0.922794, -0.33368, 0.50358, -0.971797, 0.335228, -0.050332, 0.277192, 0.254304, -0.351008, 0.171248, -0.19617, -0.093472, -0.148244, -0.785821, 0.417255, 0.922794, -0.33368, 0.50358, -0.971797, 0.764005, -0.881062, 0.956836, 0.507665, -0.423755, 0.501686, 0.997166, -0.210946, 0.606477, -0.474869, 0.531579, 0.618621, -0.66854, 0.905422, 0.862149, 0.713225, 0.382837, -0.566294, 0.239002, 0.784362, -0.383154, 0.451816, 0.335889, 0.941217, 0.853643, -0.996749, -0.556871, 0.103616, -0.929907, 0.337223],
+        shape: [10, 5]
+      }
+    },
+    'embeddings.Embedding.2': {
+      input: {
+        data: [28, 9, 19, 30, 18],
+        shape: [5]
+      },
+      weights: [
+        {
+          data: [-0.616375, -0.01389, 0.09463, 0.74616, 0.903976, 0.46756, 0.810735, -0.196574, 0.312078, 0.931824, -0.658021, -0.305244, 0.16033, -0.01051, -0.669953, -0.377897, 0.842174, -0.54385, -0.321611, 0.19085, 0.963801, 0.015549, 0.436929, -0.069782, 0.088399, -0.63811, 0.250636, 0.205862, 0.01425, 0.436179, -0.020659, 0.844792, -0.635104, -0.550835, 0.334066, -0.703519, -0.232756, 0.493263, -0.097717, -0.695123, 0.712031, 0.722442, -0.432591, 0.711508, 0.074558, 0.216026, 0.255217, 0.406698, -0.983917, -0.519696, -0.199758, 0.134159, -0.177873, -0.491386, 0.456317, 0.961498, 0.387117, 0.162469, 0.907715, -0.253443, -0.717262, -0.557943, 0.154291, 0.706519, 0.53983, -0.907997],
+          shape: [33, 2]
+        }
+      ],
+      expected: {
+        data: [0.387117, 0.162469, -0.321611, 0.19085, -0.097717, -0.695123, -0.717262, -0.557943, -0.232756, 0.493263],
+        shape: [5, 2]
+      }
+    }
+  }
+
+  window.TEST_DATA = Object.assign({}, window.TEST_DATA, DATA)
+})()