08_Weight Initializers.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Weight and bias initialisers \n",
    "\n",
    "In this reading we investigate different ways to initialise weights and biases in the layers of neural networks."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.0.0\n"
     ]
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "import tensorflow as tf\n",
    "import pandas as pd\n",
    "print(tf.__version__)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Default weights and biases\n",
    "\n",
    "In the models we have worked with so far, we have not specified the initial values of the weights and biases in each layer of our neural networks.\n",
    "\n",
    "The default values of the weights and biases in TensorFlow depend on the type of layers we are using. \n",
    "\n",
    "For example, in a `Dense` layer, the biases are set to zero (`zeros`) by default, while the weights are set according to `glorot_uniform`, the Glorot uniform initialiser. \n",
    "\n",
    "The Glorot uniform initialiser draws the weights uniformly at random from the closed interval $[-c,c]$, where $$c = \\sqrt{\\frac{6}{n_{input}+n_{output}}}$$"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "and $n_{input}$ and $n_{output}$ are the number of inputs to, and outputs from the layer respectively."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Initialising your own weights and biases\n",
    "We often would like to initialise our own weights and biases, and TensorFlow makes this process quite straightforward.\n",
    "\n",
    "When we construct a model in TensorFlow, each layer has optional arguments `kernel_initialiser` and `bias_initialiser`, which are used to set the weights and biases respectively.\n",
    "\n",
    "If a layer has no weights or biases (e.g. it is a max pooling layer), then trying to set either `kernel_initialiser` or `bias_initialiser` will throw an error.\n",
    "\n",
    "Let's see an example, which uses some of the different initialisations available in Keras."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.models import Sequential\n",
    "from tensorflow.keras.layers import Flatten, Dense, Conv1D, MaxPooling1D "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Construct a model\n",
    "\n",
    "model = Sequential([\n",
    "    Conv1D(filters=16, kernel_size=3, input_shape=(128, 64), kernel_initializer='random_uniform', bias_initializer=\"zeros\", activation='relu'),\n",
    "    MaxPooling1D(pool_size=4),\n",
    "    Flatten(),\n",
    "    Dense(64, kernel_initializer='he_uniform', bias_initializer='ones', activation='relu'),\n",
    "])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As the following example illustrates, we can also instantiate initialisers in a slightly different manner, allowing us to set optional arguments of the initialisation method."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Add some layers to our model\n",
    "\n",
    "model.add(Dense(64, \n",
    "                kernel_initializer=tf.keras.initializers.RandomNormal(mean=0.0, stddev=0.05), \n",
    "                bias_initializer=tf.keras.initializers.Constant(value=0.4), \n",
    "                activation='relu'),)\n",
    "\n",
    "model.add(Dense(8, \n",
    "                kernel_initializer=tf.keras.initializers.Orthogonal(gain=1.0, seed=None), \n",
    "                bias_initializer=tf.keras.initializers.Constant(value=0.4), \n",
    "                activation='relu'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Custom weight and bias initialisers\n",
    "It is also possible to define your own weight and bias initialisers.\n",
    "Initializers must take in two arguments, the `shape` of the tensor to be initialised, and its `dtype`.\n",
    "\n",
    "Here is a small example, which also shows how you can use your custom initializer in a layer."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow.keras.backend as K"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define a custom initializer\n",
    "\n",
    "def my_init(shape, dtype=None):\n",
    "    return K.random_normal(shape, dtype=dtype)\n",
    "\n",
    "model.add(Dense(64, kernel_initializer=my_init))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's take a look at the summary of our finalised model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv1d (Conv1D)              (None, 126, 16)           3088      \n",
      "_________________________________________________________________\n",
      "max_pooling1d (MaxPooling1D) (None, 31, 16)            0         \n",
      "_________________________________________________________________\n",
      "flatten (Flatten)            (None, 496)               0         \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 64)                31808     \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 64)                4160      \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 8)                 520       \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 64)                576       \n",
      "=================================================================\n",
      "Total params: 40,152\n",
      "Trainable params: 40,152\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# Print the model summary\n",
    "\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Visualising the initialised weights and biases"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Finally, we can see the effect of our initialisers on the weights and biases by plotting histograms of the resulting values. Compare these plots with the selected initialisers for each layer above."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x1152 with 10 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot histograms of weight and bias values\n",
    "\n",
    "fig, axes = plt.subplots(5, 2, figsize=(12,16))\n",
    "fig.subplots_adjust(hspace=0.5, wspace=0.5)\n",
    "\n",
    "# Filter out the pooling and flatten layers, that don't have any weights\n",
    "weight_layers = [layer for layer in model.layers if len(layer.weights) > 0]\n",
    "\n",
    "for i, layer in enumerate(weight_layers):\n",
    "    for j in [0, 1]:\n",
    "        axes[i, j].hist(layer.weights[j].numpy().flatten(), align='left')\n",
    "        axes[i, j].set_title(layer.weights[j].name)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Further reading and resources \n",
    "* https://keras.io/initializers/\n",
    "* https://www.tensorflow.org/versions/r2.0/api_docs/python/tf/keras/initializers"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.7.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}