neural_network_demo_classification_FocalLoss.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Neural Network From Scratch\n",
    "## classification demo"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn import datasets\n",
    "from sklearn.model_selection import train_test_split\n",
    "import matplotlib.pyplot as plt\n",
    "import time\n",
    "import numpy as np\n",
    "\n",
    "from andreiNet.neural_net import NeuralNetwork\n",
    "from andreiNet.utils import norm_data, one_hot_encode\n",
    "from andreiNet.metrics import Accuracy\n",
    "from andreiNet.losses import Loss, CrossEntropy, FocalLoss"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "$$ Cross Entropy = - \\frac{1}{N} \\sum_{k} \\sum_{i} y_{i, k} \\log(p_{i, k}) $$"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "$$ Focal Loss = - \\frac{1}{N} \\sum_{k} \\sum_{i} \\alpha_t (1 - p^t_{i, k})^{\\gamma} y_{i, k} \\log(p^t_{i, k}) $$\n",
    "$$ p^t_{i, k} = y_{i, k} \\times p_{i, k} $$"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cross entropy: 0.9830564281864164\n",
      "focal loss (gamma=0.0): 0.9830564281864164\n",
      "focal loss (gamma=2.0): 0.5310705044561551\n"
     ]
    }
   ],
   "source": [
    "y_true = np.array([[1, 0, 0], [0, 1, 0]])\n",
    "y_pred = np.array([[0.7, 0.2, 0.1], [0.2, 0.2, 0.6]])\n",
    "\n",
    "ce = CrossEntropy().loss(y_true, y_pred)\n",
    "print(\"cross entropy: {}\".format(ce))\n",
    "\n",
    "gamma = 0.0\n",
    "fl = FocalLoss(gamma=gamma).loss(y_true, y_pred)\n",
    "print(\"focal loss (gamma={}): {}\".format(gamma, fl))\n",
    "\n",
    "gamma = 2.0\n",
    "fl = FocalLoss(gamma=gamma).loss(y_true, y_pred)\n",
    "print(\"focal loss (gamma={}): {}\".format(gamma, fl))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Example 1: Easy Case, Correctly Classified"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cross entropy: 0.10536051565782628\n",
      "focal loss (gamma=2.0): 0.0010536051565782623\n",
      "ce / fl = 100.00000000000006\n"
     ]
    }
   ],
   "source": [
    "y_true = np.array([[1, 0, 0], ])\n",
    "y_pred = np.array([[0.9, 0.05, 0.05], ])\n",
    "\n",
    "ce = CrossEntropy().loss(y_true, y_pred)\n",
    "print(\"cross entropy: {}\".format(ce))\n",
    "\n",
    "fl = FocalLoss(gamma=gamma).loss(y_true, y_pred)\n",
    "print(\"focal loss (gamma={}): {}\".format(gamma, fl))\n",
    "\n",
    "print('ce / fl = {}'.format(ce / fl))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Example 2: Hard Case, Correctly Classified"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cross entropy: 0.916290731874155\n",
      "focal loss (gamma=2.0): 0.3298646634746958\n",
      "ce / fl = 2.7777777777777777\n"
     ]
    }
   ],
   "source": [
    "y_true = np.array([[1, 0, 0], ])\n",
    "y_pred = np.array([[0.4, 0.3, 0.3], ])\n",
    "\n",
    "ce = CrossEntropy().loss(y_true, y_pred)\n",
    "print(\"cross entropy: {}\".format(ce))\n",
    "\n",
    "fl = FocalLoss(gamma=gamma).loss(y_true, y_pred)\n",
    "print(\"focal loss (gamma={}): {}\".format(gamma, fl))\n",
    "\n",
    "print('ce / fl = {}'.format(ce / fl))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Example 3: Incorrectly Classified"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cross entropy: 1.2039728043259361\n",
      "focal loss (gamma=2.0): 0.5899466741197086\n",
      "ce / fl = 2.0408163265306127\n"
     ]
    }
   ],
   "source": [
    "y_true = np.array([[1, 0, 0], ])\n",
    "y_pred = np.array([[0.3, 0.4, 0.4], ])\n",
    "\n",
    "ce = CrossEntropy().loss(y_true, y_pred)\n",
    "print(\"cross entropy: {}\".format(ce))\n",
    "\n",
    "fl = FocalLoss(gamma=gamma).loss(y_true, y_pred)\n",
    "print(\"focal loss (gamma={}): {}\".format(gamma, fl))\n",
    "\n",
    "print('ce / fl = {}'.format(ce / fl))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Example 4: (very) Incorrectly Classified"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cross entropy: 2.995732273553991\n",
      "focal loss (gamma=2.0): 2.7036483768824766\n",
      "ce / fl = 1.10803324099723\n"
     ]
    }
   ],
   "source": [
    "y_true = np.array([[1, 0, 0], ])\n",
    "y_pred = np.array([[0.05, 0.9, 0.05], ])\n",
    "\n",
    "ce = CrossEntropy().loss(y_true, y_pred)\n",
    "print(\"cross entropy: {}\".format(ce))\n",
    "\n",
    "fl = FocalLoss(gamma=gamma).loss(y_true, y_pred)\n",
    "print(\"focal loss (gamma={}): {}\".format(gamma, fl))\n",
    "\n",
    "print('ce / fl = {}'.format(ce / fl))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_trn.shape = (75, 4), X_test.shape = (75, 4)\n"
     ]
    }
   ],
   "source": [
    "# Load Iris Dataset\n",
    "iris = datasets.load_iris()\n",
    "X = iris.data  \n",
    "y = iris.target\n",
    "\n",
    "# We will also split the dataset into training and testing so we can evaluate the kNN classifier\n",
    "X_trn, X_test, y_trn, y_test = train_test_split(X, \n",
    "                                                y, \n",
    "                                                test_size=0.50, \n",
    "                                                random_state=0,\n",
    "                                                stratify=y)\n",
    "\n",
    "print(\"X_trn.shape = {}, X_test.shape = {}\".format(X_trn.shape, X_test.shape))\n",
    "X_trn_norm, (trn_mean, trn_std) = norm_data(X_trn)\n",
    "X_test_norm = (X_test - trn_mean) / trn_std\n",
    "\n",
    "accuracy = Accuracy().eval"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Set parameters\n",
    "activation = 'relu'\n",
    "batch_size = 50\n",
    "random_state = 0\n",
    "lr = 0.001\n",
    "n_epochs = 20000\n",
    "weight_init = 'he_norm'\n",
    "hidden_layers = (20, 20, )\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 20000: final trn loss = 0.018051244244326994 trn metrics {'accuracy': 1.0, 'cross_entropy': 0.018051244244326994}\n",
      "val metrics {'accuracy': 0.92, 'cross_entropy': 0.1855674976623962}\n",
      "setting best model from epoch 8543\n",
      "--- 24.945825338363647 seconds ---\n",
      "trn acc 0.9866666666666667\n",
      "test acc 0.92\n"
     ]
    }
   ],
   "source": [
    "loss = CrossEntropy() #'cross_entropy'\n",
    "\n",
    "\n",
    "# Train model\n",
    "\n",
    "# Initialize model\n",
    "start_time = time.time()\n",
    "nn = NeuralNetwork(hidden=hidden_layers, \n",
    "                   init_weights=weight_init,\n",
    "                   loss=loss,\n",
    "                   activation=activation,\n",
    "                   shuffle=True,\n",
    "                   random_state=random_state,\n",
    "                   metrics=['accuracy'],\n",
    "                   verbose=False\n",
    "                   )\n",
    "\n",
    "nn.train(X_trn_norm, y_trn, \n",
    "         n_epochs=n_epochs,\n",
    "         batch_size=batch_size, \n",
    "         early_stop=None, # ('accuracy', 500),\n",
    "         lr=lr, \n",
    "         val_data=(X_test_norm, y_test),\n",
    "         save_best=True)\n",
    "\n",
    "# Run Inference\n",
    "y_pred_trn = nn.predict(X_trn_norm).argmax(axis=1)\n",
    "y_pred_test = nn.predict(X_test_norm).argmax(axis=1)\n",
    "\n",
    "delta_1 = time.time() - start_time\n",
    "print(\"--- %s seconds ---\" % (delta_1))\n",
    "print('trn acc', accuracy(y_pred_trn, y_trn))\n",
    "print('test acc', accuracy(y_pred_test, y_test))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot neural network history\n",
    "fig_size = (12, 5)\n",
    "x_axis = np.arange(len(nn.trn_metric_hist[str(loss)]))\n",
    "metrics = nn.trn_metric_hist.keys()\n",
    "fig, axs = plt.subplots(1, 2, figsize=fig_size)\n",
    "\n",
    "for i, metric in enumerate(metrics):\n",
    "    axs[i].plot(x_axis, nn.trn_metric_hist[metric], c='r', linewidth=2, label='trn')\n",
    "    axs[i].plot(x_axis, nn.val_metric_hist[metric], c='b', linewidth=2, label='val')\n",
    "    axs[i].set_xlabel('epoch', fontsize=18)\n",
    "    axs[i].set_ylabel(metric, fontsize=18)\n",
    "    if metric == loss:\n",
    "        axs[i].set_ylabel(str(metric)+' (loss)', \n",
    "                          fontsize=18)\n",
    "    axs[i].legend(prop={'size': 14})\n",
    "    \n",
    "plt.suptitle('andreiNet model history', fontsize=22)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 20000: final trn loss = 0.013490211754507623 trn metrics {'accuracy': 1.0, 'focal_loss': 0.0028348231001572455}\n",
      "val metrics {'accuracy': 0.92, 'focal_loss': 0.07749369816437636}\n",
      "setting best model from epoch 5509\n",
      "--- 18.261728048324585 seconds ---\n",
      "trn acc 1.0\n",
      "test acc 0.9333333333333333\n"
     ]
    }
   ],
   "source": [
    "loss = FocalLoss(gamma=1.0) #'cross_entropy'\n",
    "\n",
    "\n",
    "# Train model\n",
    "\n",
    "# Initialize model\n",
    "del nn\n",
    "start_time = time.time()\n",
    "nn = NeuralNetwork(hidden=hidden_layers, \n",
    "                   init_weights=weight_init,\n",
    "                   loss=loss,\n",
    "                   activation=activation,\n",
    "                   shuffle=True,\n",
    "                   random_state=random_state,\n",
    "                   metrics=['accuracy'],\n",
    "                   verbose=False\n",
    "                   )\n",
    "\n",
    "nn.train(X_trn_norm, y_trn, \n",
    "         n_epochs=n_epochs,\n",
    "         batch_size=batch_size, \n",
    "         early_stop=None, # ('accuracy', 500),\n",
    "         lr=10*lr, \n",
    "         val_data=(X_test_norm, y_test),\n",
    "         save_best=True)\n",
    "\n",
    "# Run Inference\n",
    "y_pred_trn = nn.predict(X_trn_norm).argmax(axis=1)\n",
    "y_pred_test = nn.predict(X_test_norm).argmax(axis=1)\n",
    "\n",
    "delta_1 = time.time() - start_time\n",
    "print(\"--- %s seconds ---\" % (delta_1))\n",
    "print('trn acc', accuracy(y_pred_trn, y_trn))\n",
    "print('test acc', accuracy(y_pred_test, y_test))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x360 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot neural network history\n",
    "fig_size = (12, 5)\n",
    "x_axis = np.arange(len(nn.trn_metric_hist[str(loss)]))\n",
    "metrics = nn.trn_metric_hist.keys()\n",
    "fig, axs = plt.subplots(1, 2, figsize=fig_size)\n",
    "\n",
    "for i, metric in enumerate(metrics):\n",
    "    axs[i].plot(x_axis, nn.trn_metric_hist[metric], c='r', linewidth=2, label='trn')\n",
    "    axs[i].plot(x_axis, nn.val_metric_hist[metric], c='b', linewidth=2, label='val')\n",
    "    axs[i].set_xlabel('epoch', fontsize=18)\n",
    "    axs[i].set_ylabel(metric, fontsize=18)\n",
    "    if metric == loss:\n",
    "        axs[i].set_ylabel(str(metric)+' (loss)', \n",
    "                          fontsize=18)\n",
    "    axs[i].legend(prop={'size': 14})\n",
    "    \n",
    "plt.suptitle('andreiNet model history', fontsize=22)\n",
    "plt.show()"
   ]
  }
 ],
 "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": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}