005-Neural-Network-Multi-layer-perceptron.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Neural Network - Multi-layer Perceptron (MLP)\n",
    "\n",
    "useful link: http://scikit-learn.org/stable/modules/generated/sklearn.neural_network.MLPClassifier.html#sklearn.neural_network.MLPClassifier  \n",
    "http://scikit-learn.org/stable/modules/neural_networks_supervised.html"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('Size of the feature vector', (978, 785))\n",
      "[[ 6.  1.  1. ...,  1.  1.  1.]\n",
      " [ 6.  1.  1. ...,  1.  1.  1.]\n",
      " [ 6.  1.  1. ...,  1.  1.  1.]\n",
      " ..., \n",
      " [ 6.  1.  1. ...,  1.  1.  1.]\n",
      " [ 6.  1.  1. ...,  1.  1.  1.]\n",
      " [ 6.  1.  1. ...,  1.  1.  1.]]\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "\n",
    "# global variable\n",
    "global img_number\n",
    "img_number = 550\n",
    "\n",
    "output_filename = 'digit_fv.train'\n",
    "digit_data = np.loadtxt(output_filename, dtype='float64')\n",
    "\n",
    "print(\"Size of the feature vector\", digit_data.shape)\n",
    "print digit_data[0:10,0:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import math\n",
    "from skimage import io\n",
    "\n",
    "# show image\n",
    "def vec2img(vec):\n",
    "    \n",
    "    img_row = int(math.sqrt(vec.shape[0]))\n",
    "    img_col = img_row\n",
    "    img = vec.reshape((img_row, img_col))\n",
    "    \n",
    "    io.imshow(img)\n",
    "    io.show()\n",
    "    \n",
    "# check predict\n",
    "def chkpredict(actual_class, predict_class):\n",
    "    if(actual_class==predict_class):\n",
    "        print(\"(Correct prediction)\")\n",
    "    else:\n",
    "        print(\"(Incorrect prediction)\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f6fdc0acc90>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# show image\n",
    "vec2img(digit_data[img_number,1:])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python2.7/dist-packages/sklearn/neural_network/multilayer_perceptron.py:563: ConvergenceWarning: Stochastic Optimizer: Maximum iterations reached and the optimization hasn't converged yet.\n",
      "  % (), ConvergenceWarning)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "MLPClassifier(activation='relu', alpha=1, batch_size='auto', beta_1=0.9,\n",
       "       beta_2=0.999, early_stopping=False, epsilon=1e-08,\n",
       "       hidden_layer_sizes=(15, 30), learning_rate='constant',\n",
       "       learning_rate_init=0.001, max_iter=20, momentum=0.9,\n",
       "       nesterovs_momentum=True, power_t=0.5, random_state=None,\n",
       "       shuffle=True, solver='adam', tol=0.0001, validation_fraction=0.1,\n",
       "       verbose=False, warm_start=False)"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.neural_network import MLPClassifier\n",
    "\n",
    "X = digit_data[:,1:]\n",
    "y = digit_data[:,0]\n",
    "\n",
    "# create MLP model\n",
    "clf_mlp = MLPClassifier(alpha=1, hidden_layer_sizes=(15,30), max_iter=20, verbose=False)\n",
    "clf_mlp.fit(X,y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training set score: 1.000000\n",
      "Training set loss: 0.212084\n"
     ]
    }
   ],
   "source": [
    "print(\"Training set score: %f\" % clf_mlp.score(X, y))\n",
    "print(\"Training set loss: %f\" % clf_mlp.loss_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('Actual', 2.0)\n",
      "('Predict', array([ 2.]))\n",
      "(Correct prediction)\n"
     ]
    }
   ],
   "source": [
    "# Prediction\n",
    "\n",
    "print(\"Actual\", y[img_number])\n",
    "predict_class = clf_mlp.predict([X[img_number,:]])\n",
    "print(\"Predict\", predict_class)\n",
    "chkpredict(y[img_number], predict_class)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# MLP"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Confusion matrix\n",
      "[[100   0   0   0   0   0   0   0   0   0]\n",
      " [  0 100   0   0   0   0   0   0   0   0]\n",
      " [  0   0  84   0   0   0   0   0   0   0]\n",
      " [  0   0   0  91   0   0   0   0   0   0]\n",
      " [  0   0   0   0  93   0   0   0   0   0]\n",
      " [  0   0   0   0   0  92   0   0   0   0]\n",
      " [  0   0   0   0   0   0 109   0   0   0]\n",
      " [  0   0   0   0   0   0   0  94   0   0]\n",
      " [  0   0   0   0   0   0   0   0 101   0]\n",
      " [  0   0   0   0   0   0   0   0   0 114]]\n",
      "\n",
      "Accuracy score\n",
      "100.0\n",
      "\n",
      "Precision & Recall\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "        0.0       1.00      1.00      1.00       100\n",
      "        1.0       1.00      1.00      1.00       100\n",
      "        2.0       1.00      1.00      1.00        84\n",
      "        3.0       1.00      1.00      1.00        91\n",
      "        4.0       1.00      1.00      1.00        93\n",
      "        5.0       1.00      1.00      1.00        92\n",
      "        6.0       1.00      1.00      1.00       109\n",
      "        7.0       1.00      1.00      1.00        94\n",
      "        8.0       1.00      1.00      1.00       101\n",
      "        9.0       1.00      1.00      1.00       114\n",
      "\n",
      "avg / total       1.00      1.00      1.00       978\n",
      "\n"
     ]
    }
   ],
   "source": [
    "#link: https://www.kdnuggets.com/2016/10/beginners-guide-neural-networks-python-scikit-learn.html/2\n",
    "# Data Preprocessing\n",
    "from sklearn.neural_network import MLPClassifier\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from sklearn.metrics import classification_report, confusion_matrix, accuracy_score\n",
    "\n",
    "\n",
    "scaler = StandardScaler()\n",
    "\n",
    "X = digit_data[:,1:]\n",
    "y = digit_data[:,0]\n",
    "\n",
    "# Fit only to the training data\n",
    "scaler.fit(X)\n",
    "X_train = scaler.transform(X)\n",
    "\n",
    "mlp = MLPClassifier(hidden_layer_sizes=(30,30), max_iter=200)\n",
    "mlp.fit(X_train,y)\n",
    "\n",
    "# predict\n",
    "predictions = mlp.predict(X_train)\n",
    "\n",
    "# Predictions and Evaluation\n",
    "print(\"Confusion matrix\")\n",
    "print(confusion_matrix(y,predictions))\n",
    "print(\"\\nAccuracy score\")      \n",
    "print(accuracy_score(y,predictions)*100)\n",
    "print(\"\\nPrecision & Recall\")\n",
    "print(classification_report(y,predictions))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "3\n",
      "784\n"
     ]
    }
   ],
   "source": [
    "# MLP Weight and biases\n",
    "\n",
    "#coefs_ is a list of weight matrices, where weight matrix at index i represents the weights between layer i and layer i+1.\n",
    "\n",
    "#intercepts_ is a list of bias vectors, where the vector at index i represents the bias values added to layer i+1.\n",
    "\n",
    "print len(mlp.coefs_)\n",
    "#print mlp.coefs_\n",
    "\n",
    "print len(mlp.coefs_[0])\n",
    "#mlp.intercepts_[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f6fbb351310>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#http://scikit-learn.org/stable/auto_examples/neural_networks/plot_mnist_filters.html#sphx-glr-auto-examples-neural-networks-plot-mnist-filters-py\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.neural_network import MLPClassifier\n",
    "\n",
    "fig, axes = plt.subplots(4, 4)\n",
    "# use global min / max to ensure all weights are shown on the same scale\n",
    "vmin, vmax = mlp.coefs_[0].min(), mlp.coefs_[0].max()\n",
    "for coef, ax in zip(mlp.coefs_[0].T, axes.ravel()):\n",
    "    ax.matshow(coef.reshape(28, 28), cmap=plt.cm.gray, vmin=.5 * vmin,\n",
    "               vmax=.5 * vmax)\n",
    "    ax.set_xticks(())\n",
    "    ax.set_yticks(())\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# MLP parameters\n",
    "```\n",
    "    MLPClassifier(activation='relu', alpha=1, batch_size='auto', beta_1=0.9,  \n",
    "       beta_2=0.999, early_stopping=False, epsilon=1e-08,  \n",
    "       hidden_layer_sizes=(15, 30), learning_rate='constant',  \n",
    "       learning_rate_init=0.001, max_iter=20, momentum=0.9,  \n",
    "       nesterovs_momentum=True, power_t=0.5, random_state=None,  \n",
    "       shuffle=True, solver='adam', tol=0.0001, validation_fraction=0.1,  \n",
    "       verbose=False, warm_start=False)  \n",
    "```  \n",
    "\n",
    "grid search: http://scikit-learn.org/stable/auto_examples/model_selection/plot_randomized_search.html#sphx-glr-auto-examples-model-selection-plot-randomized-search-py  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The best parameters are {'learning_rate': 'invscaling', 'hidden_layer_sizes': (20,)} with a score of 1.00\n",
      "\n",
      "GridSearchCV took 12.61 seconds for 8 candidate parameter settings.\n",
      "Model with rank: 1\n",
      "Mean validation score: 1.000 (std: 0.000)\n",
      "Parameters: {'learning_rate': 'invscaling', 'hidden_layer_sizes': (20,)}\n",
      "\n",
      "Model with rank: 1\n",
      "Mean validation score: 1.000 (std: 0.000)\n",
      "Parameters: {'learning_rate': 'adaptive', 'hidden_layer_sizes': (20,)}\n",
      "\n",
      "Model with rank: 1\n",
      "Mean validation score: 1.000 (std: 0.000)\n",
      "Parameters: {'learning_rate': 'invscaling', 'hidden_layer_sizes': (10,)}\n",
      "\n"
     ]
    }
   ],
   "source": [
    "from sklearn.model_selection import GridSearchCV\n",
    "from sklearn.model_selection import RandomizedSearchCV\n",
    "from time import time\n",
    "\n",
    "# Utility function to report best scores\n",
    "def report(results, n_top=3):\n",
    "    for i in range(1, n_top + 1):\n",
    "        candidates = np.flatnonzero(results['rank_test_score'] == i)\n",
    "        for candidate in candidates:\n",
    "            print(\"Model with rank: {0}\".format(i))\n",
    "            print(\"Mean validation score: {0:.3f} (std: {1:.3f})\".format(\n",
    "                  results['mean_test_score'][candidate],\n",
    "                  results['std_test_score'][candidate]))\n",
    "            print(\"Parameters: {0}\".format(results['params'][candidate]))\n",
    "            print(\"\")\n",
    "            \n",
    "param_grid = {\"learning_rate\":[\"constant\", \"invscaling\", \"adaptive\", \"constant\"],\n",
    "              \"hidden_layer_sizes\": [(20,), (10,)]}\n",
    "              \n",
    "    \n",
    "# run grid search\n",
    "grid_search = GridSearchCV(mlp, param_grid=param_grid)\n",
    "start = time()\n",
    "grid_search.fit(X_train, y)\n",
    "\n",
    "print(\"The best parameters are %s with a score of %0.2f\"\n",
    "      % (grid_search.best_params_, grid_search.best_score_))\n",
    "print(\"\\nGridSearchCV took %.2f seconds for %d candidate parameter settings.\"\n",
    "      % (time() - start, len(grid_search.cv_results_['params'])))\n",
    "report(grid_search.cv_results_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The best parameters are {'learning_rate': 'invscaling', 'hidden_layer_sizes': (20,)} with a score of 1.00\n",
      "\n",
      "RandomizedSearchCV took 7.16 seconds for 5 candidates parameter settings.\n",
      "Model with rank: 1\n",
      "Mean validation score: 1.000 (std: 0.000)\n",
      "Parameters: {'learning_rate': 'invscaling', 'hidden_layer_sizes': (20,)}\n",
      "\n",
      "Model with rank: 2\n",
      "Mean validation score: 0.985 (std: 0.022)\n",
      "Parameters: {'learning_rate': 'constant', 'hidden_layer_sizes': (20,)}\n",
      "\n",
      "Model with rank: 3\n",
      "Mean validation score: 0.970 (std: 0.042)\n",
      "Parameters: {'learning_rate': 'constant', 'hidden_layer_sizes': (20,)}\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# run randomized search\n",
    "n_iter_search = 5\n",
    "random_search = RandomizedSearchCV(mlp, param_distributions=param_grid,\n",
    "                                   n_iter=n_iter_search)\n",
    "\n",
    "start = time()\n",
    "random_search.fit(X_train, y)\n",
    "print(\"The best parameters are %s with a score of %0.2f\"\n",
    "      % (random_search.best_params_, random_search.best_score_))\n",
    "print(\"\\nRandomizedSearchCV took %.2f seconds for %d candidates\"\n",
    "      \" parameter settings.\" % ((time() - start), n_iter_search))\n",
    "report(random_search.cv_results_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}