ClassificatioN.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "classes = ['CALsuburb', 'MIThighway', 'MITopencountry', 'PARoffice', 'livingroom', \\\n",
    "            'MITcoast', 'MITinsidecity', 'MITstreet', 'bedroom', \\\n",
    "            'MITforest', 'MITmountain', 'MITtallbuilding', 'kitchen']\n",
    "import numpy as np\n",
    "import os\n",
    "import collections\n",
    "import csv"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Option 1: \n",
    "# Load X_train, X_test, Y_train, Y-test, npy\n",
    "# Load the Training and test sets seperately\n",
    "# y_train is assumed to have the same label order as in the above classes\n",
    "\n",
    "# CHANGE THE BELOW FILE NAMES"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(3083, 10)\n",
      "(776, 10)\n",
      "(3083, 1)\n",
      "(776, 1)\n"
     ]
    }
   ],
   "source": [
    "X_train = np.load(\"features/X_train-greyscale-nmf-10-tol20.0.npy\")\n",
    "\n",
    "X_test = np.load(\"features/X_test-greyscale-nmf-10-tol20.0.npy\")\n",
    "y_train = np.load(\"features/y_train.npy\")\n",
    "y_test = np.load(\"features/y_test.npy\")\n",
    "\n",
    "print X_train.shape\n",
    "print X_test.shape\n",
    "print y_train.shape\n",
    "print y_test.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Option 2: load X and y first as csv, then split into training and test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "\n",
    "def load_csv_formatted(csvfile):\n",
    "    \"\"\"load csv file according to the format we specified in sharelatex\n",
    "    assume each row is a sample\n",
    "    assume each column is feature\n",
    "    return X, y, featurename\n",
    "    \"\"\"\n",
    "\n",
    "\n",
    "    df = pd.read_csv(csvfile)\n",
    "    matr = df[df.keys()[4:]]\n",
    "    feature_name = df.keys()[4][:-1]\n",
    "    \n",
    "    X = matr.as_matrix()\n",
    "    labels = df['Class_Label']\n",
    "    y = labels.as_matrix()\n",
    "    length = y.shape[0]\n",
    "    \n",
    "    return X, y, feature_name"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# CHANGE THE FILENAME BELOW TO LOAD"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'load_csv_formatted' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-3-ea631a4b12ee>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mX\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfeature_name\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mload_csv_formatted\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m\"features/test_feature.csv\"\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m: name 'load_csv_formatted' is not defined"
     ]
    }
   ],
   "source": [
    "X, y, feature_name = load_csv_formatted(\"features/test_feature.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.cross_validation import StratifiedKFold\n",
    "\n",
    "#DO NOT CHANGE\n",
    "num_folds = 5\n",
    "rand_seed = 123\n",
    "train_indices, test_indices = next(iter(StratifiedKFold(y.flatten(), n_folds=num_folds, random_state=rand_seed)))\n",
    "\n",
    "X_train, y_train, X_test, y_test = X[train_indices], y[train_indices], X[test_indices], y[test_indices]\n",
    "\n",
    "print \"SHape of training\", X_train.shape\n",
    "print \"SHape of training labels\", y_train.shape\n",
    "print \"SHape of test\", X_test.shape\n",
    "print \"SHape of test labels\", y_test.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Option 3: load X and y as npy, then split into trianing and test set (change the filename below)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(3083, 500)\n",
      "(776, 500)\n",
      "(3083, 1)\n",
      "(776, 1)\n"
     ]
    }
   ],
   "source": [
    "X = np.load(\"features/X.npy\")\n",
    "y = np.load(\"features/y.npy\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.cross_validation import StratifiedKFold\n",
    "\n",
    "#DO NOT CHANGE\n",
    "num_folds = 5\n",
    "rand_seed = 123\n",
    "train_indices, test_indices = next(iter(StratifiedKFold(greyscale_labels.flatten(), n_folds=num_folds, random_state=rand_seed)))\n",
    "\n",
    "X_train, y_train, X_test, y_test = X[train_indices], y[train_indices], X[test_indices], y[test_indices]\n",
    "\n",
    "print \"SHape of training\", X_train.shape\n",
    "print \"SHape of training labels\", y_train.shape\n",
    "print \"SHape of test\", X_test.shape\n",
    "print \"SHape of test labels\", y_test.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# After the loading options, binarize the labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Let's binarize y so we can do multiclass classification\n",
      "Shape of Y_train after binarized is  (3083, 13)\n",
      "First elements of Y_train are [[0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]\n",
      " [0 0 0 0 0 0 0 0 1 0 0 0 0]]\n",
      "First elements of old_y_train are  [8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8]\n"
     ]
    }
   ],
   "source": [
    "from sklearn.preprocessing import LabelBinarizer\n",
    "\n",
    "print \"Let's binarize y so we can do multiclass classification\"\n",
    "lb = LabelBinarizer()\n",
    "lb.fit(y_train)\n",
    "y_train_bin = lb.transform(y_train)\n",
    "y_test_bin = lb.transform(y_test)\n",
    "print \"Shape of Y_train after binarized is \", y_train_bin.shape\n",
    "print \"First elements of Y_train are\", y_train_bin[0:20]\n",
    "\n",
    "old_y_train = lb.inverse_transform(y_train_bin)\n",
    "print \"First elements of old_y_train are \", old_y_train[0:20]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Normalize all the features in X_train and X_test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "from sklearn import preprocessing\n",
    "upper_train_index = len(X_train)\n",
    "X = np.concatenate((X_train, X_test))\n",
    "X = preprocessing.scale(X, axis=0)\n",
    "\n",
    "X_train = X[:upper_train_index]\n",
    "X_test = X[upper_train_index:]\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Declare the base classification code"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from sklearn import metrics\n",
    "import time\n",
    "\n",
    "\n",
    "\n",
    "# This function trains and returns a dicitonary of results\n",
    "def train(clf, X_train, X_test, y_train, y_test, used_label_binarizer, classes=range(13), name=None):\n",
    "    \"\"\"\n",
    "    Returns: a dict of errors/performance\n",
    "    \"\"\"\n",
    "    t = time.time()\n",
    "    r = collections.OrderedDict()\n",
    "    if name == None:\n",
    "        name = str(clf)\n",
    "    \n",
    "\n",
    "    clf.fit(X_train, y_train)\n",
    "    y_pred = clf.predict(X_test)\n",
    "\n",
    "    \n",
    "    y_test_multi = used_label_binarizer.inverse_transform(y_test)\n",
    "    y_pred_multi = used_label_binarizer.inverse_transform(y_pred)\n",
    "    confusion_matrix = metrics.confusion_matrix(y_test_multi, y_pred_multi)\n",
    "    \n",
    "    \n",
    "    precisions = metrics.precision_score(y_test_multi, y_pred_multi, average=None)\n",
    "    recalls = metrics.recall_score(y_test_multi, y_pred_multi, average=None)\n",
    "    f1scores = metrics.f1_score(y_test_multi, y_pred_multi, average=None)\n",
    "    \n",
    "    precision = metrics.precision_score(y_test_multi, y_pred_multi, average='weighted')\n",
    "    recall = metrics.recall_score(y_test_multi, y_pred_multi, average='weighted')\n",
    "    f1score = metrics.f1_score(y_test_multi, y_pred_multi, average='weighted')    \n",
    "    \n",
    "    auc = metrics.roc_auc_score(y_test, y_pred, average='weighted')\n",
    "    \n",
    "    r['Name'] = clf\n",
    "    r['Precision'] = np.round(precision,2)\n",
    "    r['Recall'] = np.round(recall,2)\n",
    "    r['F1Score'] = np.round(f1score,2)\n",
    "    r[\"Pred\"] = y_pred\n",
    "    r[\"Test\"] = y_test\n",
    "    r['Pred_Multi'] = y_pred_multi\n",
    "    r['Test_Multi'] = y_test_multi\n",
    "    r['Precisions'] = precisions\n",
    "    r['Recalls'] = recalls\n",
    "    r['F1Scores'] = f1scores\n",
    "    for i in range(len(classes)): \n",
    "        r['Precisions_' + str(classes[i])] = precisions[i]\n",
    "        r['Recall_' + str(classes[i])] = recalls[i]\n",
    "        r['F1Score_' + str(classes[i])] = f1scores[i]\n",
    "    \n",
    "    r['Confusion Matrix'] = confusion_matrix\n",
    "    r['Time'] = np.round(time.time() - t, 2)\n",
    "    r['CLF'] = clf\n",
    "    r['Name'] = name\n",
    "    r['AUC'] = auc\n",
    "    try:\n",
    "        r['Param'] = clf.get_params(deep=True)\n",
    "    except AttributeError:\n",
    "        r['Param'] = None\n",
    "    \n",
    "    return r"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Declare and make the results directory"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "mkdir: results: File exists\r\n"
     ]
    }
   ],
   "source": [
    "results_directory = \"results\"\n",
    "\n",
    "!mkdir $results_directory\n",
    "outfile =  results_directory + '/classification_results.csv'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Run all the classifiers. \n",
    "# YOU CAN CHANGE THE BELOW PARAMETERS IF YOU'D LIKE.\n",
    "# Note if it takes too long, you can kill the cell and you'll still have results from the earlier classifiers written out in the results csv file, as well as contained in the results list\n",
    "# Each result in the list is a dictionary of attributes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "KNN5\n",
      "Precision 0.02 Recall 0.09 F1Score 0.02 AUC 0.498750946154 Time 33.21\n",
      "Best result is F1-score  0.02 for classifier  OneVsRestClassifier(estimator=KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',\n",
      "           metric_params=None, n_neighbors=5, p=2, weights='uniform'),\n",
      "          n_jobs=1)\n",
      "Results written to  results/classification_results.csv\n",
      "KNN10\n",
      "Precision 0.01 Recall 0.09 F1Score 0.02 AUC 0.499516752577 Time 35.49\n",
      "Best result is F1-score  0.02 for classifier  OneVsRestClassifier(estimator=KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',\n",
      "           metric_params=None, n_neighbors=5, p=2, weights='uniform'),\n",
      "          n_jobs=1)\n",
      "Results written to  results/classification_results.csv\n",
      "KNN15\n",
      "Precision 0.01 Recall 0.09 F1Score 0.02 AUC 0.500197692127 Time 35.38\n",
      "Best result is F1-score  0.02 for classifier  OneVsRestClassifier(estimator=KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',\n",
      "           metric_params=None, n_neighbors=5, p=2, weights='uniform'),\n",
      "          n_jobs=1)\n",
      "Results written to  results/classification_results.csv\n",
      "SVCRBFC10.0gamma1.0\n",
      "Precision 0.08 Recall 0.07 F1Score 0.01 AUC 0.501288659794 Time 113.47\n",
      "Best result is F1-score  0.02 for classifier  OneVsRestClassifier(estimator=KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',\n",
      "           metric_params=None, n_neighbors=5, p=2, weights='uniform'),\n",
      "          n_jobs=1)\n",
      "Results written to  results/classification_results.csv\n",
      "SVCRBFC10.0gamma0.01\n",
      "Precision 0.26 Recall 0.1 F1Score 0.06 AUC 0.515812741474 Time 134.8\n",
      "Best result is F1-score  0.06 for classifier  OneVsRestClassifier(estimator=SVC(C=10.0, cache_size=200, class_weight=None, coef0=0.0, degree=3,\n",
      "  gamma=0.01, kernel='rbf', max_iter=-1, probability=False,\n",
      "  random_state=None, shrinking=True, tol=0.001, verbose=False),\n",
      "          n_jobs=1)\n",
      "Results written to  results/classification_results.csv\n",
      "SVCRBFC10.0gamma0.002\n",
      "Precision 0.28 Recall 0.14 F1Score 0.12 AUC 0.540761044535 Time 108.07\n",
      "Best result is F1-score  0.12 for classifier  OneVsRestClassifier(estimator=SVC(C=10.0, cache_size=200, class_weight=None, coef0=0.0, degree=3,\n",
      "  gamma=0.002, kernel='rbf', max_iter=-1, probability=False,\n",
      "  random_state=None, shrinking=True, tol=0.001, verbose=False),\n",
      "          n_jobs=1)\n",
      "Results written to  results/classification_results.csv\n",
      "SVCRBFC1.0gamma1.0\n",
      "Precision 0.08 Recall 0.07 F1Score 0.01 AUC 0.501288659794 Time 112.98\n",
      "Best result is F1-score  0.12 for classifier  OneVsRestClassifier(estimator=SVC(C=10.0, cache_size=200, class_weight=None, coef0=0.0, degree=3,\n",
      "  gamma=0.002, kernel='rbf', max_iter=-1, probability=False,\n",
      "  random_state=None, shrinking=True, tol=0.001, verbose=False),\n",
      "          n_jobs=1)\n",
      "Results written to  results/classification_results.csv\n",
      "SVCRBFC1.0gamma0.01\n",
      "Precision 0.19 Recall 0.08 F1Score 0.03 AUC 0.50626522638 Time 104.59\n",
      "Best result is F1-score  0.12 for classifier  OneVsRestClassifier(estimator=SVC(C=10.0, cache_size=200, class_weight=None, coef0=0.0, degree=3,\n",
      "  gamma=0.002, kernel='rbf', max_iter=-1, probability=False,\n",
      "  random_state=None, shrinking=True, tol=0.001, verbose=False),\n",
      "          n_jobs=1)\n",
      "Results written to  results/classification_results.csv\n",
      "SVCRBFC1.0gamma0.002\n",
      "Precision 0.12 Recall 0.07 F1Score 0.02 AUC 0.503146052798 Time 98.84\n",
      "Best result is F1-score  0.12 for classifier  OneVsRestClassifier(estimator=SVC(C=10.0, cache_size=200, class_weight=None, coef0=0.0, degree=3,\n",
      "  gamma=0.002, kernel='rbf', max_iter=-1, probability=False,\n",
      "  random_state=None, shrinking=True, tol=0.001, verbose=False),\n",
      "          n_jobs=1)\n",
      "Results written to  results/classification_results.csv\n",
      "SVCRBFC0.01gamma1.0\n",
      "Precision 0.0 Recall 0.06 F1Score 0.01 AUC 0.5 Time 163.53\n",
      "Best result is F1-score  0.12 for classifier  OneVsRestClassifier(estimator=SVC(C=10.0, cache_size=200, class_weight=None, coef0=0.0, degree=3,\n",
      "  gamma=0.002, kernel='rbf', max_iter=-1, probability=False,\n",
      "  random_state=None, shrinking=True, tol=0.001, verbose=False),\n",
      "          n_jobs=1)\n",
      "Results written to  results/classification_results.csv\n",
      "SVCRBFC0.01gamma0.01\n",
      "Precision 0.0 Recall 0.06 F1Score 0.01 AUC 0.5 Time 130.49\n",
      "Best result is F1-score  0.12 for classifier  OneVsRestClassifier(estimator=SVC(C=10.0, cache_size=200, class_weight=None, coef0=0.0, degree=3,\n",
      "  gamma=0.002, kernel='rbf', max_iter=-1, probability=False,\n",
      "  random_state=None, shrinking=True, tol=0.001, verbose=False),\n",
      "          n_jobs=1)\n",
      "Results written to  results/classification_results.csv\n",
      "SVCRBFC0.01gamma0.002\n",
      "Precision 0.0 Recall 0.06 F1Score 0.01 AUC 0.5 Time 117.1\n",
      "Best result is F1-score  0.12 for classifier  OneVsRestClassifier(estimator=SVC(C=10.0, cache_size=200, class_weight=None, coef0=0.0, degree=3,\n",
      "  gamma=0.002, kernel='rbf', max_iter=-1, probability=False,\n",
      "  random_state=None, shrinking=True, tol=0.001, verbose=False),\n",
      "          n_jobs=1)\n",
      "Results written to  results/classification_results.csv\n",
      "LSVC\n",
      "Precision 0.17 Recall 0.12 F1Score 0.11 AUC 0.544803010546 Time 44.09\n",
      "Best result is F1-score  0.12 for classifier  OneVsRestClassifier(estimator=SVC(C=10.0, cache_size=200, class_weight=None, coef0=0.0, degree=3,\n",
      "  gamma=0.002, kernel='rbf', max_iter=-1, probability=False,\n",
      "  random_state=None, shrinking=True, tol=0.001, verbose=False),\n",
      "          n_jobs=1)\n",
      "Results written to  results/classification_results.csv\n",
      "DT5\n",
      "Precision 0.1 Recall 0.08 F1Score 0.04 AUC 0.515021691967 Time 3.65\n",
      "Best result is F1-score  0.12 for classifier  OneVsRestClassifier(estimator=SVC(C=10.0, cache_size=200, class_weight=None, coef0=0.0, degree=3,\n",
      "  gamma=0.002, kernel='rbf', max_iter=-1, probability=False,\n",
      "  random_state=None, shrinking=True, tol=0.001, verbose=False),\n",
      "          n_jobs=1)\n",
      "Results written to  results/classification_results.csv\n",
      "DT10\n",
      "Precision 0.19 Recall 0.14 F1Score 0.14 AUC 0.545895881101 Time 5.68\n",
      "Best result is F1-score  0.14 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=10,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "DT50\n",
      "Precision 0.18 Recall 0.18 F1Score 0.18 AUC 0.55482787568 Time 7.06\n",
      "Best result is F1-score  0.18 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=50,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "DT100\n",
      "Precision 0.19 Recall 0.19 F1Score 0.19 AUC 0.561938766584 Time 7.37\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "DT15\n",
      "Precision 0.31 Recall 0.12 F1Score 0.11 AUC 0.541459493758 Time 18.69\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "DT110\n",
      "Precision 0.24 Recall 0.11 F1Score 0.11 AUC 0.540821439738 Time 37.87\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "DT150\n",
      "Precision 0.19 Recall 0.14 F1Score 0.14 AUC 0.557086043685 Time 78.27\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "DT1100\n",
      "Precision 0.2 Recall 0.14 F1Score 0.14 AUC 0.560792474312 Time 68.16\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "RFmax_depth5_num_learners100\n",
      "Precision 0.0 Recall 0.06 F1Score 0.01 AUC 0.5 Time 8.3\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "RFmax_depth10_num_learners100\n",
      "Precision 0.0 Recall 0.06 F1Score 0.01 AUC 0.5 Time 15.51\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "RFmax_depth50_num_learners100\n",
      "Precision 0.14 Recall 0.07 F1Score 0.02 AUC 0.502511422212 Time 21.43\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "RFmax_depth100_num_learners100\n",
      "Precision 0.08 Recall 0.07 F1Score 0.01 AUC 0.501288659794 Time 21.56\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "RF1max_depth5_num_learners100\n",
      "Precision 0.0 Recall 0.06 F1Score 0.01 AUC 0.5 Time 49.93\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "RF1max_depth10_num_learners100\n",
      "Precision 0.1 Recall 0.06 F1Score 0.01 AUC 0.500644329897 Time 99.32\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "RF1max_depth50_num_learners100\n",
      "Precision 0.12 Recall 0.07 F1Score 0.02 AUC 0.50237962746 Time 146.6\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "RF1max_depth100_num_learners100\n",
      "Precision 0.11 Recall 0.07 F1Score 0.02 AUC 0.501801194939 Time 134.67\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "LR\n",
      "Precision 0.21 Recall 0.13 F1Score 0.13 AUC 0.548934192048 Time 5.59\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "AB\n",
      "Precision 0.34 Recall 0.17 F1Score 0.18 AUC 0.567843917793 Time 182.27\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "Dummy\n",
      "Precision 0.08 Recall 0.07 F1Score 0.06 AUC 0.499760932128 Time 0.09\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n",
      "Dummy1\n",
      "Precision 0.1 Recall 0.08 F1Score 0.08 AUC 0.505392419934 Time 0.44\n",
      "Best result is F1-score  0.19 for classifier  DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=100,\n",
      "            max_features=None, max_leaf_nodes=None, min_samples_leaf=1,\n",
      "            min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
      "            random_state=None, splitter='best')\n",
      "Results written to  results/classification_results.csv\n"
     ]
    }
   ],
   "source": [
    "# run over all classifiers\n",
    "\n",
    "from sklearn.cross_validation import StratifiedKFold\n",
    "from sklearn import metrics\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "from sklearn.ensemble import RandomForestClassifier, AdaBoostClassifier\n",
    "from sklearn import svm\n",
    "from sklearn.multiclass import OneVsRestClassifier\n",
    "from sklearn.naive_bayes import GaussianNB\n",
    "from sklearn.naive_bayes import MultinomialNB\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.tree import DecisionTreeClassifier\n",
    "from sklearn.dummy import DummyClassifier\n",
    "import time\n",
    "\n",
    "\n",
    "results = []\n",
    "\n",
    "# CLASSIFICATION PARAMETERS\n",
    "neighbors = [5, 10, 15]\n",
    "knns = [OneVsRestClassifier(KNeighborsClassifier(n_neighbors = n)) for n in neighbors]\n",
    "\n",
    "\n",
    "num_features = len(X_train[0])\n",
    "Cs = [10.0, 1.0, 0.01]\n",
    "gammas = [1.0, 0.01, 1.0/num_features]\n",
    "\n",
    "svcrbfs = []\n",
    "for C in Cs:\n",
    "    for gamma in gammas:\n",
    "        svcrbf = OneVsRestClassifier(svm.SVC(C=C, gamma=gamma))\n",
    "        svcrbfs.append(svcrbf)\n",
    "\n",
    "lr = OneVsRestClassifier(LogisticRegression())\n",
    "        \n",
    "lsvc = OneVsRestClassifier(svm.LinearSVC())\n",
    "\n",
    "maxLearners = 100\n",
    "maxDepths = [5, 10, 50, 100]\n",
    "dts = [DecisionTreeClassifier(max_depth = maxDepth) for maxDepth in maxDepths]\n",
    "dt1s = [OneVsRestClassifier(DecisionTreeClassifier(max_depth = maxDepth)) for maxDepth in maxDepths]\n",
    "rfs = [RandomForestClassifier(n_estimators = maxLearners, max_depth = maxDepth, warm_start = False) for maxDepth in maxDepths]\n",
    "rf1s = [OneVsRestClassifier(RandomForestClassifier(n_estimators = maxLearners, max_depth = maxDepth, warm_start = False)) for maxDepth in maxDepths]\n",
    "\n",
    "\n",
    "gnb = GaussianNB()\n",
    "\n",
    "\n",
    "\n",
    "ab1 = OneVsRestClassifier(AdaBoostClassifier())\n",
    "\n",
    "dummy = DummyClassifier()\n",
    "dummy1 = OneVsRestClassifier(DummyClassifier())\n",
    "\n",
    "\n",
    "# ----------------------------------------------------------------------------------------------------\n",
    "# Perform the training and result writing below\n",
    "# ----------------------------------------------------------------------------------------------------\n",
    "\n",
    "for knn in knns:\n",
    "    name = 'KNN' + str(knn.get_params()['estimator'].get_params()['n_neighbors'])\n",
    "    print name\n",
    "    result = train(knn, X_train, X_test, y_train_bin, y_test_bin, used_label_binarizer=lb, classes=classes, name=name)\n",
    "    print \"Precision\", result['Precision'], \"Recall\", result['Recall'], \"F1Score\", result['F1Score'], \"AUC\", result['AUC'], \"Time\", result['Time']\n",
    "    results.append(result)\n",
    "    results = sorted(results, key = lambda entry: entry['F1Score'], reverse=True)\n",
    "    print \"Best result is F1-score \", results[0]['F1Score'], \"for classifier \", results[0]['CLF']\n",
    "    with open(outfile, 'w') as csvfile:\n",
    "        writer = csv.DictWriter(csvfile, fieldnames=results[0].keys())\n",
    "        writer.writeheader()\n",
    "        for row in results:\n",
    "            writer.writerow(row)\n",
    "        print \"Results written to \", outfile\n",
    "\n",
    "            \n",
    "for svcrbf in svcrbfs:\n",
    "    name = 'SVCRBF' + 'C' + str(svcrbf.get_params()['estimator'].get_params()['C']) + 'gamma' + str(svcrbf.get_params()['estimator'].get_params()['gamma'])\n",
    "    print name\n",
    "    result = train(svcrbf, X_train, X_test, y_train_bin, y_test_bin, used_label_binarizer=lb, classes=classes, name=name)\n",
    "    print \"Precision\", result['Precision'], \"Recall\", result['Recall'], \"F1Score\", result['F1Score'], \"AUC\", result['AUC'], \"Time\", result['Time']\n",
    "    results.append(result)\n",
    "    results = sorted(results, key = lambda entry: entry['F1Score'], reverse=True)\n",
    "    print \"Best result is F1-score \", results[0]['F1Score'], \"for classifier \", results[0]['CLF']\n",
    "    with open(outfile, 'w') as csvfile:\n",
    "        writer = csv.DictWriter(csvfile, fieldnames=results[0].keys())\n",
    "        writer.writeheader()\n",
    "        for row in results:\n",
    "            writer.writerow(row)\n",
    "        print \"Results written to \", outfile\n",
    "\n",
    "\n",
    "name = 'LSVC'\n",
    "print name\n",
    "result = train(lsvc, X_train, X_test, y_train_bin, y_test_bin, used_label_binarizer=lb, classes=classes, name=name)\n",
    "print \"Precision\", result['Precision'], \"Recall\", result['Recall'], \"F1Score\", result['F1Score'], \"AUC\", result['AUC'], \"Time\", result['Time']\n",
    "results.append(result)\n",
    "results = sorted(results, key = lambda entry: entry['F1Score'], reverse=True)\n",
    "print \"Best result is F1-score \", results[0]['F1Score'], \"for classifier \", results[0]['CLF']\n",
    "with open(outfile, 'w') as csvfile:\n",
    "    writer = csv.DictWriter(csvfile, fieldnames=results[0].keys())\n",
    "    writer.writeheader()\n",
    "    for row in results:\n",
    "        writer.writerow(row)\n",
    "    print \"Results written to \", outfile\n",
    "\n",
    "\n",
    "for dt in dts:\n",
    "    name = 'DT' + str(dt.get_params()['max_depth'])\n",
    "    print name\n",
    "    result = train(dt, X_train, X_test, y_train_bin, y_test_bin, used_label_binarizer=lb, classes=classes, name=name)\n",
    "    print \"Precision\", result['Precision'], \"Recall\", result['Recall'], \"F1Score\", result['F1Score'], \"AUC\", result['AUC'], \"Time\", result['Time']\n",
    "    results.append(result)\n",
    "    results = sorted(results, key = lambda entry: entry['F1Score'], reverse=True)\n",
    "    print \"Best result is F1-score \", results[0]['F1Score'], \"for classifier \", results[0]['CLF']\n",
    "    with open(outfile, 'w') as csvfile:\n",
    "        writer = csv.DictWriter(csvfile, fieldnames=results[0].keys())\n",
    "        writer.writeheader()\n",
    "        for row in results:\n",
    "            writer.writerow(row)\n",
    "        print \"Results written to \", outfile\n",
    "\n",
    "for dt1 in dt1s:\n",
    "    name = 'DT1' + str(dt1.get_params()['estimator'].get_params()['max_depth'])\n",
    "    print name\n",
    "    result = train(dt1, X_train, X_test, y_train_bin, y_test_bin, used_label_binarizer=lb, classes=classes, name=name)\n",
    "    print \"Precision\", result['Precision'], \"Recall\", result['Recall'], \"F1Score\", result['F1Score'], \"AUC\", result['AUC'], \"Time\", result['Time']\n",
    "    results.append(result)\n",
    "    results = sorted(results, key = lambda entry: entry['F1Score'], reverse=True)\n",
    "    print \"Best result is F1-score \", results[0]['F1Score'], \"for classifier \", results[0]['CLF']\n",
    "    with open(outfile, 'w') as csvfile:\n",
    "        writer = csv.DictWriter(csvfile, fieldnames=results[0].keys())\n",
    "        writer.writeheader()\n",
    "        for row in results:\n",
    "            writer.writerow(row)\n",
    "        print \"Results written to \", outfile\n",
    "    \n",
    "for rf in rfs:\n",
    "    name = 'RF' + 'max_depth' + str(rf.get_params()['max_depth']) + '_num_learners' + str(rf.get_params()['n_estimators'])\n",
    "    print name\n",
    "    result = train(rf, X_train, X_test, y_train_bin, y_test_bin, used_label_binarizer=lb, classes=classes, name=name)\n",
    "    print \"Precision\", result['Precision'], \"Recall\", result['Recall'], \"F1Score\", result['F1Score'], \"AUC\", result['AUC'], \"Time\", result['Time']\n",
    "    results.append(result)\n",
    "    results = sorted(results, key = lambda entry: entry['F1Score'], reverse=True)\n",
    "    print \"Best result is F1-score \", results[0]['F1Score'], \"for classifier \", results[0]['CLF']\n",
    "    with open(outfile, 'w') as csvfile:\n",
    "        writer = csv.DictWriter(csvfile, fieldnames=results[0].keys())\n",
    "        writer.writeheader()\n",
    "        for row in results:\n",
    "            writer.writerow(row)\n",
    "        print \"Results written to \", outfile\n",
    "    \n",
    "for rf1 in rf1s:\n",
    "    name = 'RF1' + 'max_depth' + str(rf1.get_params()['estimator'].get_params()['max_depth']) + '_num_learners' + str(rf1.get_params()['estimator'].get_params()['n_estimators'])\n",
    "    print name\n",
    "    result = train(rf1,X_train, X_test, y_train_bin, y_test_bin, used_label_binarizer=lb, classes=classes, name=name)\n",
    "    print \"Precision\", result['Precision'], \"Recall\", result['Recall'], \"F1Score\", result['F1Score'], \"AUC\", result['AUC'], \"Time\", result['Time']\n",
    "    results.append(result)\n",
    "    results = sorted(results, key = lambda entry: entry['F1Score'], reverse=True)\n",
    "    print \"Best result is F1-score \", results[0]['F1Score'], \"for classifier \", results[0]['CLF']\n",
    "    with open(outfile, 'w') as csvfile:\n",
    "        writer = csv.DictWriter(csvfile, fieldnames=results[0].keys())\n",
    "        writer.writeheader()\n",
    "        for row in results:\n",
    "            writer.writerow(row)\n",
    "        print \"Results written to \", outfile\n",
    "    \n",
    "# name = 'GNB'\n",
    "# print name\n",
    "# result = train(gnb, X_train, X_test, y_train_bin, y_test_bin, used_label_binarizer=lb, classes=classes, name=name)\n",
    "# print \"Precision\", result['Precision'], \"Recall\", result['Recall'], \"F1Score\", result['F1Score'], \"AUC\", result['AUC'], \"Time\", result['Time']\n",
    "# results.append(result)\n",
    "# with open(outfile, 'w') as csvfile:\n",
    "#     writer = csv.DictWriter(csvfile, fieldnames=results[0].keys())\n",
    "#     writer.writeheader()\n",
    "#     for row in results:\n",
    "#         writer.writerow(row)\n",
    "#     print \"Results written to \", outfile\n",
    "\n",
    "name = 'LR'\n",
    "print name\n",
    "result = train(lr, X_train, X_test, y_train_bin, y_test_bin, used_label_binarizer=lb, classes=classes, name=name)\n",
    "print \"Precision\", result['Precision'], \"Recall\", result['Recall'], \"F1Score\", result['F1Score'], \"AUC\", result['AUC'], \"Time\", result['Time']\n",
    "results.append(result)\n",
    "results = sorted(results, key = lambda entry: entry['F1Score'], reverse=True)\n",
    "print \"Best result is F1-score \", results[0]['F1Score'], \"for classifier \", results[0]['CLF']\n",
    "with open(outfile, 'w') as csvfile:\n",
    "    writer = csv.DictWriter(csvfile, fieldnames=results[0].keys())\n",
    "    writer.writeheader()\n",
    "    for row in results:\n",
    "        writer.writerow(row)\n",
    "    print \"Results written to \", outfile\n",
    "\n",
    "\n",
    "name = 'AB'\n",
    "print name\n",
    "result = train(ab1, X_train, X_test, y_train_bin, y_test_bin, used_label_binarizer=lb, classes=classes, name=name)\n",
    "print \"Precision\", result['Precision'], \"Recall\", result['Recall'], \"F1Score\", result['F1Score'], \"AUC\", result['AUC'], \"Time\", result['Time']\n",
    "results.append(result)\n",
    "results = sorted(results, key = lambda entry: entry['F1Score'], reverse=True)\n",
    "print \"Best result is F1-score \", results[0]['F1Score'], \"for classifier \", results[0]['CLF']\n",
    "with open(outfile, 'w') as csvfile:\n",
    "    writer = csv.DictWriter(csvfile, fieldnames=results[0].keys())\n",
    "    writer.writeheader()\n",
    "    for row in results:\n",
    "        writer.writerow(row)\n",
    "    print \"Results written to \", outfile\n",
    "\n",
    "name = 'Dummy'\n",
    "print name\n",
    "result = train(dummy, X_train, X_test, y_train_bin, y_test_bin, used_label_binarizer=lb, classes=classes, name=name)\n",
    "print \"Precision\", result['Precision'], \"Recall\", result['Recall'], \"F1Score\", result['F1Score'], \"AUC\", result['AUC'], \"Time\", result['Time']\n",
    "results.append(result)\n",
    "results = sorted(results, key = lambda entry: entry['F1Score'], reverse=True)\n",
    "print \"Best result is F1-score \", results[0]['F1Score'], \"for classifier \", results[0]['CLF']\n",
    "with open(outfile, 'w') as csvfile:\n",
    "    writer = csv.DictWriter(csvfile, fieldnames=results[0].keys())\n",
    "    writer.writeheader()\n",
    "    for row in results:\n",
    "        writer.writerow(row)\n",
    "    print \"Results written to \", outfile\n",
    "\n",
    "name = 'Dummy1'\n",
    "print name\n",
    "result = train(dummy1, X_train, X_test, y_train_bin, y_test_bin, used_label_binarizer=lb, classes=classes, name=name)\n",
    "print \"Precision\", result['Precision'], \"Recall\", result['Recall'], \"F1Score\", result['F1Score'], \"AUC\", result['AUC'], \"Time\", result['Time']\n",
    "results.append(result)\n",
    "results = sorted(results, key = lambda entry: entry['F1Score'], reverse=True)\n",
    "print \"Best result is F1-score \", results[0]['F1Score'], \"for classifier \", results[0]['CLF']\n",
    "with open(outfile, 'w') as csvfile:\n",
    "    writer = csv.DictWriter(csvfile, fieldnames=results[0].keys())\n",
    "    writer.writeheader()\n",
    "    for row in results:\n",
    "        writer.writerow(row)\n",
    "    print \"Results written to \", outfile\n",
    "    \n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Find the important Features using random forest gini feature importsnces"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "from sklearn.ensemble import RandomForestClassifier\n",
    "import csv\n",
    "\n",
    "feat_file_prefix = \"results/feature_importances_\"\n",
    "maxLearners = 100\n",
    "maxDepth = 10\n",
    "\n",
    "labels = range(13)\n",
    "\n",
    "for i in range(13):\n",
    "    \n",
    "    this_y_train = (y_train == i).flatten()\n",
    "    forest = RandomForestClassifier(n_estimators=maxLearners, max_depth = maxDepth)\n",
    "    forest.fit(X_train, this_y_train)\n",
    "    importances = forest.feature_importances_\n",
    "    indices = np.argsort(importances)[::-1]\n",
    "\n",
    "    with open(feat_file_prefix + str(classes[i]) + \".csv\", 'w') as csvfile:\n",
    "        writer = csv.writer(csvfile)\n",
    "        writer.writerow(['Index', 'Importance'])\n",
    "        for f in range(len(X_train[0])):\n",
    "            writer.writerow([indices[f], importances[indices[f]]])\n",
    "    print \"Feature importances written to \", feat_file_prefix + str(classes[i]) + \".csv\"\n",
    "        "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Write a confusion matrix out to csv and plot it."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "%matplotlib inline\n",
    "def plot_confusion_matrix(cm, title='Confusion matrix', cmap=plt.cm.Blues, classes=classes):\n",
    "    # First normalize into recalls.\n",
    "    cm = np.array(cm) * 1.0 / np.sum(cm, axis=1)\n",
    "    \n",
    "    plt.figure(figsize=(10,10))\n",
    "    plt.imshow(cm, interpolation='nearest', cmap=cmap)\n",
    "    plt.title(title)\n",
    "    plt.colorbar()\n",
    "    tick_marks = np.arange(len(classes))\n",
    "    plt.xticks(tick_marks, classes, rotation=45)\n",
    "    plt.yticks(tick_marks, classes)\n",
    "    plt.tight_layout()\n",
    "    plt.ylabel('True label')\n",
    "    plt.xlabel('Predicted label')\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Choose which classifier and name of confusion file below, first"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "SVC RBF\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAq8AAALECAYAAAAmUtkEAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAElEQVQ1rwYeAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x10aba3050>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# look at confusion matrices of the best\n",
    "\n",
    "\n",
    "# CHOOSE YOUR CLASSIFIER AND FILE NAME\n",
    "clf = OneVsRestClassifier(svm.SVC())\n",
    "confusefile = 'results/????????????????????????.csv'\n",
    "\n",
    "\n",
    "print (\"SVC RBF\")\n",
    "r = train(clf, X_train, X_test, y_train_bin, y_test_bin, used_label_binarizer=lb)\n",
    "\n",
    "confuse = r['Confusion Matrix']\n",
    "\n",
    "plot_confusion_matrix(confuse)\n",
    "\n",
    "with open(confusefile, 'w') as csvfile:\n",
    "    writer = csv.writer(csvfile)\n",
    "    writer.writerow([''] + classes)\n",
    "    for i in range(len(confuse)):\n",
    "        writer.writerow([classes[i]] + list(confuse[i]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "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": "2.7.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}