Targeted-Dropout.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "import random as rn\n",
    "import os\n",
    "os.environ['PYTHONHASHSEED'] = '42'\n",
    "np.random.seed(42)\n",
    "rn.seed(42)\n",
    "tf.enable_eager_execution()\n",
    "session_conf = tf.ConfigProto(intra_op_parallelism_threads=1, inter_op_parallelism_threads=1)\n",
    "tf.set_random_seed(42)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "# Targeted Dropout"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Paper: https://openreview.net/pdf?id=HkghWScuoQ"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "**Note** Accuracies are low as I am using very less data for the sake of analysis."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "### Dependencies"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tensorflow:  1.12.0\n",
      "Matplotlib:  3.0.2\n",
      "Seaborn:  0.9.0\n",
      "Numpy:  1.15.4\n",
      "Pandas:  0.18.1\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf; print(\"Tensorflow: \", tf.__version__)\n",
    "import matplotlib.pyplot as plt; import matplotlib; print(\"Matplotlib: \", matplotlib.__version__)\n",
    "import seaborn as sns; print(\"Seaborn: \", sns.__version__)\n",
    "import numpy as np; print(\"Numpy: \", np.__version__)\n",
    "import pandas as pd; print(\"Pandas: \", pd.__version__)\n",
    "from pprint import pprint\n",
    "mnist = tf.keras.datasets.mnist\n",
    "import dropout_functions"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## Using a small subset of MNIST data for fast analysis"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "(x_train, y_train),(x_test, y_test) = mnist.load_data()\n",
    "x_train, x_test = x_train / 255.0 - 0.5, x_test / 255.0 - 0.5\n",
    "y_train = y_train.astype('int64')\n",
    "y_test = y_test.astype('int64')\n",
    "# idx = np.random.choice(range(x_train.shape[0]),size=1000)\n",
    "# x_train, y_train = x_train[idx], y_train[idx]\n",
    "# idx = np.random.choice(range(x_train.shape[0]),size=100)\n",
    "# x_test, y_test = x_test[idx], y_test[idx]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## Vanilla Unregularized DNN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "def get_model_unregularized():\n",
    "    return tf.keras.models.Sequential([\n",
    "        tf.keras.layers.Flatten(),\n",
    "        tf.keras.layers.Dense(1000, activation=tf.nn.relu, use_bias=False, kernel_initializer=tf.keras.initializers.RandomNormal(seed=42)),\n",
    "        tf.keras.layers.Dense(1000, activation=tf.nn.relu, use_bias=False, kernel_initializer=tf.keras.initializers.RandomNormal(seed=42)),\n",
    "        tf.keras.layers.Dense(500, activation=tf.nn.relu, use_bias=False, kernel_initializer=tf.keras.initializers.RandomNormal(seed=42)),\n",
    "        tf.keras.layers.Dense(200, activation=tf.nn.relu, use_bias=False, kernel_initializer=tf.keras.initializers.RandomNormal(seed=42)),\n",
    "        tf.keras.layers.Dense(10, activation=tf.nn.softmax, use_bias=False, kernel_initializer=tf.keras.initializers.RandomNormal(seed=42))\n",
    "    ])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/2\n",
      " - 23s - loss: 0.2888 - acc: 0.9106\n",
      "Epoch 2/2\n",
      " - 23s - loss: 0.1601 - acc: 0.9527\n",
      "['loss', 'acc'] [0.23888619943112135, 0.9347]\n"
     ]
    }
   ],
   "source": [
    "model = get_model_unregularized()\n",
    "model.compile(optimizer=tf.train.AdamOptimizer(),\n",
    "              loss='sparse_categorical_crossentropy',\n",
    "              metrics=['accuracy'])\n",
    "model.fit(x_train, y_train, epochs=2,verbose=2, use_multiprocessing=True)\n",
    "print(model.metrics_names ,model.evaluate(x_test, y_test, verbose=2))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAABJkAAAFACAYAAAAfw61rAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nyQQAAAAAAIDz9v8B3s/Is8BgcGwAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 1440x360 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "df = pd.DataFrame(model.history.history)\n",
    "plt.figure(1, figsize=(20,5))\n",
    "plt.subplot(121)\n",
    "plt.plot(df[\"loss\"])\n",
    "plt.xticks(df.index)\n",
    "plt.xlabel('Epoch')\n",
    "plt.ylabel('Train Loss')\n",
    "plt.subplot(122)\n",
    "plt.plot(df[\"acc\"])\n",
    "plt.xticks(df.index)\n",
    "plt.xlabel('Epoch')\n",
    "plt.ylabel('Train Accuracy');\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "# Pruning Methods"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "def mask_for_big_values(weights, pruning_threshold):\n",
    "    small_weights = np.abs(weights) < pruning_threshold\n",
    "    return np.logical_not(small_weights)\n",
    "\n",
    "def prune_weights(model, prune_rate=10):\n",
    "    weights = model.get_weights()\n",
    "    weight_matrices_tf = model.weights\n",
    "    for (w,\n",
    "         tf_weight_matrix,\n",
    "        ) in zip(\n",
    "         weights[:-1],\n",
    "         weight_matrices_tf[:-1],\n",
    "        ):\n",
    "        w_shape = w.shape\n",
    "        w = tf.reshape(w, [-1, w_shape[-1]])\n",
    "        norm = tf.abs(w)\n",
    "        idx = tf.to_int32(prune_rate/100 * tf.to_float(tf.shape(w)[0]))\n",
    "        threshold = tf.contrib.framework.sort(norm, axis=0)[idx]\n",
    "        mask = norm < threshold[None, :]\n",
    "        mask = tf.where(\n",
    "          mask,\n",
    "          tf.zeros_like(w, dtype=tf.float32), tf.ones_like(w, dtype=tf.float32))\n",
    "        w = mask * w\n",
    "        w = tf.reshape(w, w_shape)\n",
    "        tf_weight_matrix.assign(w)\n",
    "\n",
    "def prune_units(model, pruning_perc=10):\n",
    "    pruned_model = tf.keras.models.Sequential()\n",
    "    prev_mask = []\n",
    "    pruned_model.add(tf.keras.layers.Flatten())\n",
    "\n",
    "    for w in model.trainable_weights[:-1]:\n",
    "        w_shape = w.shape\n",
    "        w = w.numpy()\n",
    "        norm = tf.norm(w, axis=0)\n",
    "        idx = int(pruning_perc /100 * int(w.shape[1]))\n",
    "        sorted_norms = tf.contrib.framework.sort(norm)\n",
    "\n",
    "        threshold = sorted_norms[idx]\n",
    "        mask = (norm >= threshold)\n",
    "#         print(w.shape, mask.shape)\n",
    "        w = w[:, tf.transpose(mask)]\n",
    "        if type(prev_mask)==np.bool_ :\n",
    "            w = w[prev_mask,:]\n",
    "        elif len(prev_mask)!=0:\n",
    "            w = w[prev_mask,:]\n",
    "        prev_mask=mask.numpy()\n",
    "        layer_new = tf.keras.layers.Dense(w.shape[1], activation=tf.nn.relu, use_bias=False, kernel_initializer=tf.keras.initializers.Constant(w))       \n",
    "        pruned_model.add(layer_new)\n",
    "#         print(len(layer_new.weights), len(pruned_model.layers[1].weights))\n",
    "    weights = model.trainable_weights[-1]\n",
    "    weights = weights.numpy()\n",
    "    if type(prev_mask)==np.bool_ :\n",
    "        weights = weights[prev_mask,:]\n",
    "        layer_new = tf.keras.layers.Dense(weights.shape[1], activation=tf.nn.softmax, use_bias=False, kernel_initializer=tf.keras.initializers.Constant(weights))\n",
    "        pruned_model.add(layer_new)\n",
    "#         print(len(layer_new.weights), len(pruned_model.layers[1].weights))\n",
    "    elif len(prev_mask)!=0:\n",
    "        weights = weights[prev_mask,:]\n",
    "        layer_new = tf.keras.layers.Dense(weights.shape[1], activation=tf.nn.softmax, use_bias=False, kernel_initializer=tf.keras.initializers.Constant(weights))\n",
    "        pruned_model.add(layer_new)\n",
    "#         print(len(layer_new.weights), len(pruned_model.layers[1].weights))\n",
    "    return pruned_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.09593052361039445, 0.97]"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model = get_model_unregularized()\n",
    "model.compile(optimizer=tf.train.AdamOptimizer(),\n",
    "          loss='sparse_categorical_crossentropy',\n",
    "          metrics=['accuracy'])\n",
    "model.fit(x_train, y_train, batch_size=128,  epochs=5,verbose=0)\n",
    "model.evaluate(x_test, y_test, verbose=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "### Distribution before pruning"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/f20160295/.local/lib/python3.5/site-packages/scipy/stats/stats.py:1713: FutureWarning: Using a non-tuple sequence for multidimensional indexing is deprecated; use `arr[tuple(seq)]` instead of `arr[seq]`. In the future this will be interpreted as an array index, `arr[np.array(seq)]`, which will result either in an error or a different result.\n",
      "  return np.add.reduce(sorted[indexer] * weights, axis=axis) / sumval\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAW4AAAD8CAYAAABXe05zAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.distplot(model.layers[1].get_weights()[0].flatten())\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true,
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.0956727026754059, 0.9704]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[1000, 1000, 500, 200, 10]"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "prune_weights(model, 10)\n",
    "model.compile(optimizer=tf.train.AdamOptimizer(),\n",
    "          loss='sparse_categorical_crossentropy',\n",
    "          metrics=['accuracy'])\n",
    "print(model.evaluate(x_test, y_test, verbose=0))\n",
    "[layer.units for layer in model.layers[1:]]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "### Distribution after pruning"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/f20160295/.local/lib/python3.5/site-packages/scipy/stats/stats.py:1713: FutureWarning: Using a non-tuple sequence for multidimensional indexing is deprecated; use `arr[tuple(seq)]` instead of `arr[seq]`. In the future this will be interpreted as an array index, `arr[np.array(seq)]`, which will result either in an error or a different result.\n",
      "  return np.add.reduce(sorted[indexer] * weights, axis=axis) / sumval\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAD8CAYAAABn919SAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nrrwlEAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.distplot(model.layers[1].get_weights()[0].flatten())\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "# Weight Targeted Dropout with Pruning"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAY4AAAEWCAYAAABxMXBSAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>unreg</th>\n",
       "      <th>(0.25, 0)</th>\n",
       "      <th>(0.33, 0.75)</th>\n",
       "      <th>(0.5, 0.75)</th>\n",
       "      <th>(0.66, 0.75)</th>\n",
       "      <th>(0.5, 0.5)</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Pruning %</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.9673</td>\n",
       "      <td>0.9714</td>\n",
       "      <td>0.9729</td>\n",
       "      <td>0.9684</td>\n",
       "      <td>0.9707</td>\n",
       "      <td>0.9729</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>0.9669</td>\n",
       "      <td>0.9711</td>\n",
       "      <td>0.973</td>\n",
       "      <td>0.9683</td>\n",
       "      <td>0.9707</td>\n",
       "      <td>0.9727</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20</th>\n",
       "      <td>0.9673</td>\n",
       "      <td>0.9705</td>\n",
       "      <td>0.9735</td>\n",
       "      <td>0.9689</td>\n",
       "      <td>0.9701</td>\n",
       "      <td>0.9731</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>30</th>\n",
       "      <td>0.9674</td>\n",
       "      <td>0.9714</td>\n",
       "      <td>0.9737</td>\n",
       "      <td>0.9681</td>\n",
       "      <td>0.9718</td>\n",
       "      <td>0.9742</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>40</th>\n",
       "      <td>0.9677</td>\n",
       "      <td>0.9702</td>\n",
       "      <td>0.9744</td>\n",
       "      <td>0.969</td>\n",
       "      <td>0.973</td>\n",
       "      <td>0.9748</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50</th>\n",
       "      <td>0.968</td>\n",
       "      <td>0.9703</td>\n",
       "      <td>0.9732</td>\n",
       "      <td>0.9696</td>\n",
       "      <td>0.9738</td>\n",
       "      <td>0.9765</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>60</th>\n",
       "      <td>0.9669</td>\n",
       "      <td>0.9686</td>\n",
       "      <td>0.9746</td>\n",
       "      <td>0.9697</td>\n",
       "      <td>0.9733</td>\n",
       "      <td>0.9701</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>70</th>\n",
       "      <td>0.9655</td>\n",
       "      <td>0.9643</td>\n",
       "      <td>0.9735</td>\n",
       "      <td>0.9709</td>\n",
       "      <td>0.973</td>\n",
       "      <td>0.9678</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>80</th>\n",
       "      <td>0.9513</td>\n",
       "      <td>0.947</td>\n",
       "      <td>0.9675</td>\n",
       "      <td>0.9551</td>\n",
       "      <td>0.9478</td>\n",
       "      <td>0.9495</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>90</th>\n",
       "      <td>0.888</td>\n",
       "      <td>0.8804</td>\n",
       "      <td>0.9065</td>\n",
       "      <td>0.8756</td>\n",
       "      <td>0.857</td>\n",
       "      <td>0.8964</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            unreg (0.25, 0) (0.33, 0.75) (0.5, 0.75) (0.66, 0.75) (0.5, 0.5)\n",
       "Pruning %                                                                   \n",
       "0          0.9673    0.9714       0.9729      0.9684       0.9707     0.9729\n",
       "10         0.9669    0.9711        0.973      0.9683       0.9707     0.9727\n",
       "20         0.9673    0.9705       0.9735      0.9689       0.9701     0.9731\n",
       "30         0.9674    0.9714       0.9737      0.9681       0.9718     0.9742\n",
       "40         0.9677    0.9702       0.9744       0.969        0.973     0.9748\n",
       "50          0.968    0.9703       0.9732      0.9696       0.9738     0.9765\n",
       "60         0.9669    0.9686       0.9746      0.9697       0.9733     0.9701\n",
       "70         0.9655    0.9643       0.9735      0.9709        0.973     0.9678\n",
       "80         0.9513     0.947       0.9675      0.9551       0.9478     0.9495\n",
       "90          0.888    0.8804       0.9065      0.8756        0.857     0.8964"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "confs = [\"unreg\", (0.25, 0), (0.33, 0.75), (0.5,0.75),(0.66,0.75),(0.5, 0.5)]\n",
    "dfw = pd.DataFrame(columns=confs, index=range(0,100,10))\n",
    "for conf in confs:\n",
    "    if conf==\"unreg\":\n",
    "        model = get_model_unregularized()\n",
    "    else:\n",
    "        model = dropout_functions.get_targeted_dropout_model(\"weight\", conf)\n",
    "    \n",
    "    model.compile(optimizer=tf.train.AdamOptimizer(),\n",
    "          loss='sparse_categorical_crossentropy',\n",
    "          metrics=['accuracy'])\n",
    "    model.fit(x_train, y_train, batch_size=128, epochs=5,verbose=0, use_multiprocessing=True)\n",
    "    for pruning in range(0,100,10):\n",
    "        prune_weights(model, pruning)\n",
    "        dfw[conf][pruning] = model.evaluate(x_test, y_test, verbose=0)[1]\n",
    "dfw.index.name = \"Pruning %\"\n",
    "dfw.plot();\n",
    "plt.xlabel(\"Pruning %\")\n",
    "plt.ylabel(\"Accuracy\")\n",
    "plt.title(\"(Dropout Rate, Target Rate)\")\n",
    "plt.show();\n",
    "dfw"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "# Unit Targeted Dropout with Pruning"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYgAAAEWCAYAAAB8LwAVAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n9P8BNYpM3blFW9QAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>unreg</th>\n",
       "      <th>(0.25, 0)</th>\n",
       "      <th>(0.33, 0.75)</th>\n",
       "      <th>(0.5, 0.75)</th>\n",
       "      <th>(0.66, 0.75)</th>\n",
       "      <th>(0.5, 0.5)</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Pruning %</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.9699</td>\n",
       "      <td>0.974</td>\n",
       "      <td>0.9708</td>\n",
       "      <td>0.9639</td>\n",
       "      <td>0.9522</td>\n",
       "      <td>0.9713</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>0.9696</td>\n",
       "      <td>0.9737</td>\n",
       "      <td>0.9716</td>\n",
       "      <td>0.9644</td>\n",
       "      <td>0.9544</td>\n",
       "      <td>0.9714</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20</th>\n",
       "      <td>0.9693</td>\n",
       "      <td>0.9729</td>\n",
       "      <td>0.9719</td>\n",
       "      <td>0.9667</td>\n",
       "      <td>0.958</td>\n",
       "      <td>0.9709</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>30</th>\n",
       "      <td>0.9693</td>\n",
       "      <td>0.9699</td>\n",
       "      <td>0.9721</td>\n",
       "      <td>0.968</td>\n",
       "      <td>0.9596</td>\n",
       "      <td>0.9701</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>40</th>\n",
       "      <td>0.9661</td>\n",
       "      <td>0.9672</td>\n",
       "      <td>0.9714</td>\n",
       "      <td>0.9629</td>\n",
       "      <td>0.9672</td>\n",
       "      <td>0.9671</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50</th>\n",
       "      <td>0.9576</td>\n",
       "      <td>0.9636</td>\n",
       "      <td>0.9706</td>\n",
       "      <td>0.9625</td>\n",
       "      <td>0.9704</td>\n",
       "      <td>0.9639</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>60</th>\n",
       "      <td>0.9496</td>\n",
       "      <td>0.9428</td>\n",
       "      <td>0.9652</td>\n",
       "      <td>0.9554</td>\n",
       "      <td>0.9719</td>\n",
       "      <td>0.9627</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>70</th>\n",
       "      <td>0.8888</td>\n",
       "      <td>0.7967</td>\n",
       "      <td>0.9317</td>\n",
       "      <td>0.9311</td>\n",
       "      <td>0.9673</td>\n",
       "      <td>0.9443</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>80</th>\n",
       "      <td>0.7156</td>\n",
       "      <td>0.4261</td>\n",
       "      <td>0.8137</td>\n",
       "      <td>0.8375</td>\n",
       "      <td>0.9444</td>\n",
       "      <td>0.7889</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>90</th>\n",
       "      <td>0.1772</td>\n",
       "      <td>0.2335</td>\n",
       "      <td>0.2875</td>\n",
       "      <td>0.2021</td>\n",
       "      <td>0.4704</td>\n",
       "      <td>0.423</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            unreg (0.25, 0) (0.33, 0.75) (0.5, 0.75) (0.66, 0.75) (0.5, 0.5)\n",
       "Pruning %                                                                   \n",
       "0          0.9699     0.974       0.9708      0.9639       0.9522     0.9713\n",
       "10         0.9696    0.9737       0.9716      0.9644       0.9544     0.9714\n",
       "20         0.9693    0.9729       0.9719      0.9667        0.958     0.9709\n",
       "30         0.9693    0.9699       0.9721       0.968       0.9596     0.9701\n",
       "40         0.9661    0.9672       0.9714      0.9629       0.9672     0.9671\n",
       "50         0.9576    0.9636       0.9706      0.9625       0.9704     0.9639\n",
       "60         0.9496    0.9428       0.9652      0.9554       0.9719     0.9627\n",
       "70         0.8888    0.7967       0.9317      0.9311       0.9673     0.9443\n",
       "80         0.7156    0.4261       0.8137      0.8375       0.9444     0.7889\n",
       "90         0.1772    0.2335       0.2875      0.2021       0.4704      0.423"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "confs = [\"unreg\", (0.25, 0), (0.33, 0.75), (0.5,0.75),(0.66,0.75),(0.5, 0.5)]\n",
    "\n",
    "dfu = pd.DataFrame(columns=confs, index=range(0,100,10))\n",
    "for conf in confs:\n",
    "    if conf==\"unreg\":\n",
    "        model = get_model_unregularized()\n",
    "    else:\n",
    "        model = dropout_functions.get_targeted_dropout_model(\"unit\", conf)\n",
    "    \n",
    "    model.compile(optimizer=tf.train.AdamOptimizer(),\n",
    "          loss='sparse_categorical_crossentropy',\n",
    "          metrics=['accuracy'])\n",
    "    model.fit(x_train, y_train, batch_size=128, epochs=5,verbose=0, use_multiprocessing=True)\n",
    "    for pruning in range(0,100,10):\n",
    "        pruned_model = prune_units(model, pruning)\n",
    "        pruned_model.compile(optimizer=tf.train.AdamOptimizer(),\n",
    "          loss='sparse_categorical_crossentropy',\n",
    "          metrics=['accuracy'])\n",
    "        dfu[conf][pruning] = pruned_model.evaluate(x_test, y_test, verbose=0)[1]\n",
    "dfu.index.name = \"Pruning %\"\n",
    "dfu.plot();\n",
    "plt.xlabel(\"Pruning %\")\n",
    "plt.ylabel(\"Accuracy\")\n",
    "plt.title(\"(Dropout Rate, Target Rate)\")\n",
    "plt.show()\n",
    "dfu"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "# Analysis\n",
    "Weights Pruning simplified, is setting a set of cells of less consequence in a matrix to 0 after training. And Weight Dropout uses similar idea while training. We are deleting connections.\n",
    "\n",
    "Unit Pruning and Dropout similarly extends the idea to an entire neuron. Here we delete rows and corresponing columns(in the next layer) in the weight matrices which are of less consequence.\n",
    "\n",
    "We can see that weight targeted dropout gives comparable accuracies until ~70% pruning where as unit targeted pruning loses accuracies after ~40% pruning. \n",
    "\n",
    "Different combination of targeted rate and dropout rate are good for unit dropout and weight dropout. Unit dropout does best with `(alpha= 0.33, gamma= 0.75)` configuration whereas weight dropout does well with `(alpha=0.5, gamma=0.5)` configuration. \n",
    "\n",
    "This might be a model dependent phenomenon.\n",
    "\n",
    "### General Advantages of Pruning:\n",
    "- Reduce Model Size\n",
    "- Decrease inference/prediction time and computation\n",
    "- Introduces some amount of regularization.\n",
    "\n",
    "### General Advantages of Dropout:\n",
    "- Regularization\n",
    "- Reduces/prevents dominance of few connections or neurons.\n",
    "\n",
    "### Contributions of Targeted Dropout:\n",
    "- Ignoring connections or neurons with less magnitude. Stochasticity gives the connections and neurons multiple chances to contribute to the decision procedure.\n",
    "- The goal is opposite to vanilla dropout's goal, to ignore less efficient connections/neurons.\n",
    "- Model building simplified, i.e, we could start with a very general model and let the dropout+pruning decide the best model size.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "# Further Work \n",
    "\n",
    "## Layer wise Pruning and Targeted Dropout\n",
    "\n",
    "Extrapolating this idea, we could do the same with entire layers. We could essentially fuse layers. This might prove very useful in deep networks.\n",
    "\n",
    "We can look for nonlinearities present between layers which are not contributing to the performance. By removing such non-linearities, we are left with pairs of weight matrices, which can be \"fused\" by multiplying them.\n",
    "\n",
    "Some metrics to find such non-linearities:\n",
    "- L2 or L1 distance between `layer2(layer1(input))` and `fusedLayer(input)`.\n",
    "- L2 or L1 norm of gradients of pairs of layers. To weigh their contribution to the final performance.\n",
    "\n",
    "## Combination of Vanilla Dropout and Targeted Dropout\n",
    "Why? \n",
    "I believe Vanilla Dropout and Targeted Dropout are trying to achieve opposite goals.\n",
    "\n",
    "Former is trying to make weaker connections/neruons stronger and also avoiding dependency on small set of connections or neurons. Latter is trying to ignore weakly performing connections/neruons while giving them a fair chance to recover.\n",
    "\n",
    "A combination of both might help us acheive robustness as well as better model efficiency at the same time.\n"
   ]
  }
 ],
 "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": "3.5.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}