MS using SVM on lesion load.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For MS training we have 3 datasets: train, validation and holdout"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import nibabel as nib\n",
    "\n",
    "from scipy import interp\n",
    "from sklearn.utils import shuffle\n",
    "from sklearn.model_selection import GroupShuffleSplit\n",
    "from sklearn.metrics import confusion_matrix, roc_auc_score, roc_curve, auc\n",
    "from sklearn.model_selection import KFold\n",
    "from sklearn.svm import SVC\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import os\n",
    "import time\n",
    "import h5py\n",
    "\n",
    "from config import *\n",
    "from utils import specificity, sensitivity, balanced_accuracy, shuffle_data, normalize_float"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Start timing\n",
    "start_time = time.time()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "zero_one_normalize = False\n",
    "dtype = np.float32"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# load hdf5 files and extract columns\n",
    "train_h5 = h5py.File('/analysis/share/Ritter/MS/CIS/train_dataset_FLAIR_lesions_filled.h5', 'r')\n",
    "holdout_h5 = h5py.File('/analysis/share/Ritter/MS/CIS/holdout_dataset_FLAIR_lesions_filled.h5', 'r')\n",
    "\n",
    "# loading only labels from original file\n",
    "y_train = train_h5['y']\n",
    "y_holdout = holdout_h5['y']\n",
    "\n",
    "train_lesions_h5 = h5py.File('/analysis/share/Ritter/MS/CIS/train_dataset_lesions.h5', 'r')\n",
    "holdout_lesions_h5 = h5py.File('/analysis/share/Ritter/MS/CIS/holdout_dataset_lesions.h5', 'r')\n",
    "\n",
    "lesion_masks_train = train_lesions_h5['masks']\n",
    "lesion_masks_holdout = holdout_lesions_h5['masks']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Convert to lesion volume"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# convert data to numpy arrays using lesions masks\n",
    "X_train = np.array(lesion_masks_train, dtype=dtype)\n",
    "y_train = np.array(y_train)\n",
    "\n",
    "X_holdout = np.array(lesion_masks_holdout, dtype=dtype)\n",
    "y_holdout = np.array(y_holdout)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Total datset length: 124\n",
      "Number of healthy controls: 61\n",
      "Number of MS patients: 63\n"
     ]
    }
   ],
   "source": [
    "print(\"Total datset length: {}\".format(len(y_train)))\n",
    "print(\"Number of healthy controls: {}\".format(len(y_train[y_train==0.])))\n",
    "print(\"Number of MS patients: {}\".format(len(y_train[y_train==1.])))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# sum over all dimensions\n",
    "X_train = np.sum(X_train, axis=(1, 2, 3)).reshape(-1, 1)\n",
    "X_holdout = np.sum(X_holdout, axis=(1, 2, 3)).reshape(-1, 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7f96f948b2e8>"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "_, bins, _ = plt.hist(X_train[y_train==1.], bins=20, alpha=0.5, range=[0, 8000])\n",
    "_ = plt.hist(X_train[y_train==0.], bins=bins, alpha=0.5, range=[0, 8000])\n",
    "plt.legend([\"MS\", \"HC\"])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Normalization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def normalize(train, test):\n",
    "    # get training set moments\n",
    "    mean = np.mean(train)\n",
    "    std = np.std(train)\n",
    "    # apply on train and test\n",
    "    train = (train - mean)/std\n",
    "    test = (test - mean)/std\n",
    "    return train, test"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.model_selection import GridSearchCV\n",
    "from sklearn import preprocessing\n",
    "from sklearn.pipeline import make_pipeline\n",
    "\n",
    "def svc_param_selection(X, y, n_folds):\n",
    "    Cs = [0.001, 0.01, 0.1, 1, 10]\n",
    "    kernels = ['linear', 'rbf']\n",
    "    param_grid = {'svc__C': Cs,\n",
    "                  'svc__kernel': kernels}\n",
    "    # use standard scaler for preprocessing\n",
    "    scaler = preprocessing.StandardScaler()\n",
    "    pipeline = make_pipeline(scaler, SVC(gamma='auto'))\n",
    "    grid_search = GridSearchCV(pipeline, param_grid, cv=n_folds, n_jobs=10)\n",
    "    grid_search.fit(X, y)\n",
    "    grid_search.best_params_\n",
    "    return grid_search.best_params_, grid_search.cv_results_"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fold 0\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/fabiane/.local/lib/python3.6/site-packages/sklearn/model_selection/_search.py:841: DeprecationWarning: The default of the `iid` parameter will change from True to False in version 0.22 and will be removed in 0.24. This will change numeric results when test-set sizes are unequal.\n",
      "  DeprecationWarning)\n",
      "/home/fabiane/.local/lib/python3.6/site-packages/sklearn/svm/base.py:196: FutureWarning: The default value of gamma will change from 'auto' to 'scale' in version 0.22 to account better for unscaled features. Set gamma explicitly to 'auto' or 'scale' to avoid this warning.\n",
      "  \"avoid this warning.\", FutureWarning)\n",
      "/home/fabiane/.local/lib/python3.6/site-packages/sklearn/model_selection/_search.py:841: DeprecationWarning: The default of the `iid` parameter will change from True to False in version 0.22 and will be removed in 0.24. This will change numeric results when test-set sizes are unequal.\n",
      "  DeprecationWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best paramter value: {'svc__C': 10, 'svc__kernel': 'rbf'}\n",
      "######## Training set results ########\n",
      "Balanced accuracy 99.11 %\n",
      "Sensitivity 98.21 %\n",
      "Specificity 100.00 %\n",
      "######## Validation set results ########\n",
      "Balanced accuracy 81.82 %\n",
      "Sensitivity 100.00 %\n",
      "Specificity 63.64 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 83.77\n",
      "Fold 1\n",
      "Best paramter value: {'svc__C': 10, 'svc__kernel': 'linear'}\n",
      "######## Training set results ########\n",
      "Balanced accuracy 80.62 %\n",
      "Sensitivity 69.09 %\n",
      "Specificity 92.16 %\n",
      "######## Validation set results ########\n",
      "Balanced accuracy 81.25 %\n",
      "Sensitivity 62.50 %\n",
      "Specificity 100.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 95.00\n",
      "Fold 2\n",
      "Best paramter value: {'svc__C': 1, 'svc__kernel': 'rbf'}\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.04 %\n",
      "Sensitivity 100.00 %\n",
      "Specificity 96.08 %\n",
      "######## Validation set results ########\n",
      "Balanced accuracy 73.75 %\n",
      "Sensitivity 87.50 %\n",
      "Specificity 60.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 76.88\n",
      "Fold 3\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/fabiane/.local/lib/python3.6/site-packages/sklearn/model_selection/_search.py:841: DeprecationWarning: The default of the `iid` parameter will change from True to False in version 0.22 and will be removed in 0.24. This will change numeric results when test-set sizes are unequal.\n",
      "  DeprecationWarning)\n",
      "/home/fabiane/.local/lib/python3.6/site-packages/sklearn/svm/base.py:196: FutureWarning: The default value of gamma will change from 'auto' to 'scale' in version 0.22 to account better for unscaled features. Set gamma explicitly to 'auto' or 'scale' to avoid this warning.\n",
      "  \"avoid this warning.\", FutureWarning)\n",
      "/home/fabiane/.local/lib/python3.6/site-packages/sklearn/model_selection/_search.py:841: DeprecationWarning: The default of the `iid` parameter will change from True to False in version 0.22 and will be removed in 0.24. This will change numeric results when test-set sizes are unequal.\n",
      "  DeprecationWarning)\n",
      "/home/fabiane/.local/lib/python3.6/site-packages/sklearn/svm/base.py:196: FutureWarning: The default value of gamma will change from 'auto' to 'scale' in version 0.22 to account better for unscaled features. Set gamma explicitly to 'auto' or 'scale' to avoid this warning.\n",
      "  \"avoid this warning.\", FutureWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best paramter value: {'svc__C': 10, 'svc__kernel': 'rbf'}\n",
      "######## Training set results ########\n",
      "Balanced accuracy 99.02 %\n",
      "Sensitivity 98.04 %\n",
      "Specificity 100.00 %\n",
      "######## Validation set results ########\n",
      "Balanced accuracy 70.83 %\n",
      "Sensitivity 91.67 %\n",
      "Specificity 50.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 77.08\n",
      "Fold 4\n",
      "Best paramter value: {'svc__C': 10, 'svc__kernel': 'rbf'}\n",
      "######## Training set results ########\n",
      "Balanced accuracy 99.04 %\n",
      "Sensitivity 100.00 %\n",
      "Specificity 98.08 %\n",
      "######## Validation set results ########\n",
      "Balanced accuracy 83.33 %\n",
      "Sensitivity 100.00 %\n",
      "Specificity 66.67 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 90.12\n",
      "Fold 5\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/fabiane/.local/lib/python3.6/site-packages/sklearn/model_selection/_search.py:841: DeprecationWarning: The default of the `iid` parameter will change from True to False in version 0.22 and will be removed in 0.24. This will change numeric results when test-set sizes are unequal.\n",
      "  DeprecationWarning)\n",
      "/home/fabiane/.local/lib/python3.6/site-packages/sklearn/svm/base.py:196: FutureWarning: The default value of gamma will change from 'auto' to 'scale' in version 0.22 to account better for unscaled features. Set gamma explicitly to 'auto' or 'scale' to avoid this warning.\n",
      "  \"avoid this warning.\", FutureWarning)\n",
      "/home/fabiane/.local/lib/python3.6/site-packages/sklearn/model_selection/_search.py:841: DeprecationWarning: The default of the `iid` parameter will change from True to False in version 0.22 and will be removed in 0.24. This will change numeric results when test-set sizes are unequal.\n",
      "  DeprecationWarning)\n",
      "/home/fabiane/.local/lib/python3.6/site-packages/sklearn/svm/base.py:196: FutureWarning: The default value of gamma will change from 'auto' to 'scale' in version 0.22 to account better for unscaled features. Set gamma explicitly to 'auto' or 'scale' to avoid this warning.\n",
      "  \"avoid this warning.\", FutureWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best paramter value: {'svc__C': 10, 'svc__kernel': 'rbf'}\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.13 %\n",
      "Sensitivity 98.11 %\n",
      "Specificity 98.15 %\n",
      "######## Validation set results ########\n",
      "Balanced accuracy 85.71 %\n",
      "Sensitivity 100.00 %\n",
      "Specificity 71.43 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 80.00\n",
      "Fold 6\n",
      "Best paramter value: {'svc__C': 10, 'svc__kernel': 'linear'}\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/fabiane/.local/lib/python3.6/site-packages/sklearn/model_selection/_search.py:841: DeprecationWarning: The default of the `iid` parameter will change from True to False in version 0.22 and will be removed in 0.24. This will change numeric results when test-set sizes are unequal.\n",
      "  DeprecationWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "######## Training set results ########\n",
      "Balanced accuracy 80.49 %\n",
      "Sensitivity 68.52 %\n",
      "Specificity 92.45 %\n",
      "######## Validation set results ########\n",
      "Balanced accuracy 88.89 %\n",
      "Sensitivity 77.78 %\n",
      "Specificity 100.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 97.22\n",
      "Training Time: 0.0h:0.0m:20.59779405593872s\n",
      "######## Final results ########\n",
      "Validation balanced accuracies: \n",
      " [0.8181818181818181, 0.8125, 0.7375, 0.7083333333333333, 0.8333333333333333, 0.8571428571428572, 0.8888888888888888]\n",
      "Validation balanced accuracies mean: 0.8079828901257472\n",
      "Validation final sensitivities: \n",
      " [1.0, 0.625, 0.875, 0.9166666666666666, 1.0, 1.0, 0.7777777777777778]\n",
      "Validation final sensitivities' mean: 0.8849206349206348\n",
      "Validation final specificities: \n",
      " [0.6363636363636364, 1.0, 0.6, 0.5, 0.6666666666666666, 0.7142857142857143, 1.0]\n",
      "Validation final specificities' mean: 0.7310451453308595\n",
      "Mean ROC AUC score 85.72\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "kf = KFold(n_splits=7)\n",
    "fold = 0\n",
    "\n",
    "best_params = []\n",
    "train_balanced_accuracies = []\n",
    "train_sensitivities = []\n",
    "train_specificities = []\n",
    "\n",
    "val_balanced_accuracies = []\n",
    "val_sensitivities = []\n",
    "val_specificities = []\n",
    "\n",
    "auc_scores = []\n",
    "tprs = []\n",
    "mean_fpr = np.linspace(0, 1, 100)\n",
    "\n",
    "# shuffle the data once\n",
    "X_train, y_train = shuffle_data(X_train, y_train)\n",
    "\n",
    "# nested cross-validation\n",
    "for train_idx, test_idx in kf.split(X_train):\n",
    "    print(\"Fold %i\" %fold)\n",
    "    fold += 1\n",
    "    \n",
    "    # Start inner cross-validation\n",
    "    best_param, cv_result = svc_param_selection(\n",
    "        X_train[train_idx],\n",
    "        y_train[train_idx],\n",
    "        n_folds=5)\n",
    "    \n",
    "    print(\"Best paramter value: {}\".format(best_param))\n",
    "    \n",
    "    model = SVC(kernel=best_param[\"svc__kernel\"], C=best_param[\"svc__C\"])\n",
    "    model.fit(X_train[train_idx], y_train[train_idx])\n",
    "    \n",
    "    # training set results\n",
    "    train_pred = model.predict(X_train[train_idx])\n",
    "    train_bal_acc = balanced_accuracy(y_train[train_idx], train_pred)\n",
    "    train_sens = sensitivity(y_train[train_idx], train_pred)\n",
    "    train_spec = specificity(y_train[train_idx], train_pred)\n",
    "    # val set results\n",
    "    val_pred = model.predict(X_train[test_idx])\n",
    "    val_scores = model.decision_function(X_train[test_idx])\n",
    "    val_bal_acc = balanced_accuracy(y_train[test_idx], val_pred)\n",
    "    val_sens = sensitivity(y_train[test_idx], val_pred)\n",
    "    val_spec = specificity(y_train[test_idx], val_pred)\n",
    "    \n",
    "    roc_auc = roc_auc_score(y_train[test_idx], val_scores)\n",
    "    fpr, tpr, thresholds = roc_curve(y_train[test_idx], val_scores)   \n",
    "    \n",
    "    # Store results\n",
    "    best_params.append(best_param)\n",
    "    train_balanced_accuracies.append(train_bal_acc)\n",
    "    train_sensitivities.append(train_sens)\n",
    "    train_specificities.append(train_spec)\n",
    "    val_balanced_accuracies.append(val_bal_acc)\n",
    "    val_sensitivities.append(val_sens)\n",
    "    val_specificities.append(val_spec)\n",
    "    \n",
    "    auc_scores.append(roc_auc)\n",
    "    # interpolate with diagonal to get comparable results\n",
    "    tprs.append(interp(mean_fpr, fpr, tpr))\n",
    "    tprs[-1][0] = 0.0 # correct lowest value after interpolation\n",
    "\n",
    "    # Print results\n",
    "    print(\"######## Training set results ########\")\n",
    "    print(\"Balanced accuracy {:.2f} %\".format(train_bal_acc*100))\n",
    "    print(\"Sensitivity {:.2f} %\".format(train_sens*100))\n",
    "    print(\"Specificity {:.2f} %\".format(train_spec*100))\n",
    "    print(\"######## Validation set results ########\")\n",
    "    print(\"Balanced accuracy {:.2f} %\".format(val_bal_acc*100))\n",
    "    print(\"Sensitivity {:.2f} %\".format(val_sens*100))\n",
    "    print(\"Specificity {:.2f} %\".format(val_spec*100))\n",
    "    print(\"Area Under the Receiver Operating Curve (ROC AUC score) {:.2f}\".format(roc_auc*100))\n",
    "    plt.plot(fpr, tpr, lw=1, alpha=0.3,\n",
    "             label='ROC fold %d (AUC = %0.2f)' % (fold, roc_auc))\n",
    "    \n",
    "training_time = time.time() - start_time\n",
    "print(\"Training Time: {}h:{}m:{}s\".format(\n",
    "            training_time//3600, (training_time//60)%60, training_time%60))\n",
    "# Print results\n",
    "print(\"######## Final results ########\")\n",
    "print(\"Validation balanced accuracies: \\n {}\".format(val_balanced_accuracies))\n",
    "print(\"Validation balanced accuracies mean: {}\".format(np.mean(val_balanced_accuracies)))\n",
    "print(\"Validation final sensitivities: \\n {}\".format(val_sensitivities))\n",
    "print(\"Validation final sensitivities' mean: {}\".format(np.mean(val_sensitivities)))\n",
    "print(\"Validation final specificities: \\n {}\".format(val_specificities))\n",
    "print(\"Validation final specificities' mean: {}\".format(np.mean(val_specificities)))\n",
    "print(\"Mean ROC AUC score {:.2f}\".format(np.mean(auc_scores)*100))\n",
    "\n",
    "# Plot ROC Curves\n",
    "plt.plot([0, 1], [0, 1], linestyle='--', lw=2, color='r',\n",
    "         label='Chance', alpha=.8)\n",
    "\n",
    "mean_tpr = np.mean(tprs, axis=0)\n",
    "mean_tpr[-1] = 1.0 # correct max value after interpolation and mean\n",
    "mean_auc = auc(mean_fpr, mean_tpr)\n",
    "#assert(mean_auc == np.mean(auc_scores))\n",
    "\n",
    "std_auc = np.std(auc_scores)\n",
    "plt.plot(mean_fpr, mean_tpr, color='b',\n",
    "         label=r'Mean ROC (AUC = %0.2f $\\pm$ %0.2f)' % (mean_auc, std_auc),\n",
    "         lw=2, alpha=.8)\n",
    "\n",
    "std_tpr = np.std(tprs, axis=0)\n",
    "tprs_upper = np.minimum(mean_tpr + std_tpr, 1)\n",
    "tprs_lower = np.maximum(mean_tpr - std_tpr, 0)\n",
    "plt.fill_between(mean_fpr, tprs_lower, tprs_upper, color='grey', alpha=.2,\n",
    "                 label=r'$\\pm$ 1 std. dev.')\n",
    "\n",
    "plt.xlim([-0.05, 1.05])\n",
    "plt.ylim([-0.05, 1.05])\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('Receiver operating characteristic example')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "training_time = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "counter = {}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def majority_vote(best_params):\n",
    "    \"\"\"\n",
    "    Find the most often used combination \n",
    "    of parameters.\n",
    "    \"\"\"\n",
    "    assert(len(best_params)>=1)\n",
    "    counter = {}\n",
    "    # count unique value list\n",
    "    for i in range(len(best_params)):\n",
    "        # turn values into key\n",
    "        new_key = \"\"\n",
    "        for x in list(best_params[i].values()):\n",
    "            new_key = new_key + str(x) + \"_\"\n",
    "        if new_key in counter.keys():\n",
    "            counter[new_key] += 1\n",
    "        else:\n",
    "            counter[new_key] = 1\n",
    "    # select most frequent value list\n",
    "    majority_param = max(counter, key=lambda key: counter[key])\n",
    "    # reformat to list\n",
    "    majority_param = majority_param[:-1].split(\"_\")\n",
    "    # reformat to dictionary\n",
    "    result = {}\n",
    "    for key, value in zip(best_params[0].keys(), majority_param):\n",
    "        result[key] = value\n",
    "    return result"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'svc__C': '10', 'svc__kernel': 'rbf'}\n"
     ]
    }
   ],
   "source": [
    "majority_param = majority_vote(best_params)\n",
    "print(majority_param)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Evaluation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Train on the entire training set with the best parameters from above and test on the holdout dataset for final performance."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trial 0\n",
      "######## Training set results ########\n",
      "Balanced accuracy 80.85 %\n",
      "Sensitivity 68.25 %\n",
      "Specificity 93.44 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 88.46 %\n",
      "Sensitivity 76.92 %\n",
      "Specificity 100.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 94.62\n",
      "Trial 1\n",
      "######## Training set results ########\n",
      "Balanced accuracy 80.85 %\n",
      "Sensitivity 68.25 %\n",
      "Specificity 93.44 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 88.46 %\n",
      "Sensitivity 76.92 %\n",
      "Specificity 100.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 94.62\n",
      "Trial 2\n",
      "######## Training set results ########\n",
      "Balanced accuracy 80.85 %\n",
      "Sensitivity 68.25 %\n",
      "Specificity 93.44 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 88.46 %\n",
      "Sensitivity 76.92 %\n",
      "Specificity 100.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 94.62\n",
      "Trial 3\n",
      "######## Training set results ########\n",
      "Balanced accuracy 80.85 %\n",
      "Sensitivity 68.25 %\n",
      "Specificity 93.44 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 88.46 %\n",
      "Sensitivity 76.92 %\n",
      "Specificity 100.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 94.62\n",
      "Trial 4\n",
      "######## Training set results ########\n",
      "Balanced accuracy 80.85 %\n",
      "Sensitivity 68.25 %\n",
      "Specificity 93.44 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 88.46 %\n",
      "Sensitivity 76.92 %\n",
      "Specificity 100.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 94.62\n",
      "Trial 5\n",
      "######## Training set results ########\n",
      "Balanced accuracy 80.85 %\n",
      "Sensitivity 68.25 %\n",
      "Specificity 93.44 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 88.46 %\n",
      "Sensitivity 76.92 %\n",
      "Specificity 100.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 94.62\n",
      "Trial 6\n",
      "######## Training set results ########\n",
      "Balanced accuracy 80.85 %\n",
      "Sensitivity 68.25 %\n",
      "Specificity 93.44 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 88.46 %\n",
      "Sensitivity 76.92 %\n",
      "Specificity 100.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 94.62\n",
      "Trial 7\n",
      "######## Training set results ########\n",
      "Balanced accuracy 80.85 %\n",
      "Sensitivity 68.25 %\n",
      "Specificity 93.44 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 88.46 %\n",
      "Sensitivity 76.92 %\n",
      "Specificity 100.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 94.62\n",
      "Trial 8\n",
      "######## Training set results ########\n",
      "Balanced accuracy 80.85 %\n",
      "Sensitivity 68.25 %\n",
      "Specificity 93.44 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 88.46 %\n",
      "Sensitivity 76.92 %\n",
      "Specificity 100.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 94.62\n",
      "Trial 9\n",
      "######## Training set results ########\n",
      "Balanced accuracy 80.85 %\n",
      "Sensitivity 68.25 %\n",
      "Specificity 93.44 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 88.46 %\n",
      "Sensitivity 76.92 %\n",
      "Specificity 100.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 94.62\n",
      "Training Time: 0.0h:0.0m:21.26303744316101s\n",
      "######## Final results ########\n",
      "Holdout balanced accuracies: \n",
      " [0.8846153846153846, 0.8846153846153846, 0.8846153846153846, 0.8846153846153846, 0.8846153846153846, 0.8846153846153846, 0.8846153846153846, 0.8846153846153846, 0.8846153846153846, 0.8846153846153846]\n",
      "Holdout balanced accuracies mean: 0.8846153846153847\n",
      "Holdout final sensitivities: \n",
      " [0.7692307692307693, 0.7692307692307693, 0.7692307692307693, 0.7692307692307693, 0.7692307692307693, 0.7692307692307693, 0.7692307692307693, 0.7692307692307693, 0.7692307692307693, 0.7692307692307693]\n",
      "Holdout final sensitivities' mean: 0.7692307692307693\n",
      "Holdout final specificities: \n",
      " [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]\n",
      "Holdout final specificities' mean: 1.0\n",
      "Mean ROC AUC score 94.62\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# training args\n",
    "kernel = majority_param[\"svc__kernel\"]\n",
    "C = float(majority_param[\"svc__C\"])\n",
    "model = SVC(kernel=kernel, C=C)\n",
    "\n",
    "num_trials = 10\n",
    "\n",
    "train_balanced_accuracies = []\n",
    "train_sensitivities = []\n",
    "train_specificities = []\n",
    "\n",
    "holdout_balanced_accuracies = []\n",
    "holdout_sensitivities = []\n",
    "holdout_specificities = []\n",
    "\n",
    "auc_scores = []\n",
    "tprs = []\n",
    "mean_fpr = np.linspace(0, 1, 100)\n",
    "\n",
    "for i in range(num_trials):\n",
    "    print(\"Trial %i\" %i)\n",
    "    \n",
    "    # shuffle the data each time\n",
    "    X_train, y_train = shuffle_data(X_train, y_train)\n",
    "    \n",
    "    # normalize\n",
    "    X_train, X_holdout = normalize(X_train, X_holdout)\n",
    "    # Start training\n",
    "    model.fit(X_train, y_train)\n",
    "    # training set results\n",
    "    train_pred = model.predict(X_train)\n",
    "    train_bal_acc = balanced_accuracy(y_train, train_pred)\n",
    "    train_sens = sensitivity(y_train, train_pred)\n",
    "    train_spec = specificity(y_train, train_pred)\n",
    "    # holdout set results\n",
    "    holdout_pred = model.predict(X_holdout)\n",
    "    holdout_scores = model.decision_function(X_holdout)\n",
    "    holdout_bal_acc = balanced_accuracy(y_holdout, holdout_pred)\n",
    "    holdout_sens = sensitivity(y_holdout, holdout_pred)\n",
    "    holdout_spec = specificity(y_holdout, holdout_pred)\n",
    "    \n",
    "    roc_auc = roc_auc_score(y_holdout, holdout_scores)\n",
    "    fpr, tpr, thresholds = roc_curve(y_holdout, holdout_scores)   \n",
    "    \n",
    "    # Store results\n",
    "    train_balanced_accuracies.append(train_bal_acc)\n",
    "    train_sensitivities.append(train_sens)\n",
    "    train_specificities.append(train_spec)\n",
    "    holdout_balanced_accuracies.append(holdout_bal_acc)\n",
    "    holdout_sensitivities.append(holdout_sens)\n",
    "    holdout_specificities.append(holdout_spec)\n",
    "    \n",
    "    auc_scores.append(roc_auc)\n",
    "    # interpolate with diagonal to get comparable results\n",
    "    tprs.append(interp(mean_fpr, fpr, tpr))\n",
    "    tprs[-1][0] = 0.0 # correct lowest value after interpolation\n",
    "\n",
    "    # Print results\n",
    "    print(\"######## Training set results ########\")\n",
    "    print(\"Balanced accuracy {:.2f} %\".format(train_bal_acc*100))\n",
    "    print(\"Sensitivity {:.2f} %\".format(train_sens*100))\n",
    "    print(\"Specificity {:.2f} %\".format(train_spec*100))\n",
    "    print(\"######## Holdout set results ########\")\n",
    "    print(\"Balanced accuracy {:.2f} %\".format(holdout_bal_acc*100))\n",
    "    print(\"Sensitivity {:.2f} %\".format(holdout_sens*100))\n",
    "    print(\"Specificity {:.2f} %\".format(holdout_spec*100))\n",
    "    print(\"Area Under the Receiver Operating Curve (ROC AUC score) {:.2f}\".format(roc_auc*100))\n",
    "    plt.plot(fpr, tpr, lw=1, alpha=0.3,\n",
    "             label='ROC trial %d (AUC = %0.2f)' % (i, roc_auc))\n",
    "    \n",
    "training_time = time.time() - start_time\n",
    "print(\"Training Time: {}h:{}m:{}s\".format(\n",
    "            training_time//3600, (training_time//60)%60, training_time%60))\n",
    "# Print results\n",
    "print(\"######## Final results ########\")\n",
    "print(\"Holdout balanced accuracies: \\n {}\".format(holdout_balanced_accuracies))\n",
    "print(\"Holdout balanced accuracies mean: {}\".format(np.mean(holdout_balanced_accuracies)))\n",
    "print(\"Holdout final sensitivities: \\n {}\".format(holdout_sensitivities))\n",
    "print(\"Holdout final sensitivities' mean: {}\".format(np.mean(holdout_sensitivities)))\n",
    "print(\"Holdout final specificities: \\n {}\".format(holdout_specificities))\n",
    "print(\"Holdout final specificities' mean: {}\".format(np.mean(holdout_specificities)))\n",
    "print(\"Mean ROC AUC score {:.2f}\".format(np.mean(auc_scores)*100))\n",
    "\n",
    "# Plot ROC Curves\n",
    "plt.plot([0, 1], [0, 1], linestyle='--', lw=2, color='r',\n",
    "         label='Chance', alpha=.8)\n",
    "\n",
    "mean_tpr = np.mean(tprs, axis=0)\n",
    "mean_tpr[-1] = 1.0 # correct max value after interpolation and mean\n",
    "mean_auc = auc(mean_fpr, mean_tpr)\n",
    "#assert(mean_auc == np.mean(auc_scores))\n",
    "\n",
    "std_auc = np.std(auc_scores)\n",
    "plt.plot(mean_fpr, mean_tpr, color='b',\n",
    "         label=r'Mean ROC (AUC = %0.2f $\\pm$ %0.2f)' % (mean_auc, std_auc),\n",
    "         lw=2, alpha=.8)\n",
    "\n",
    "std_tpr = np.std(tprs, axis=0)\n",
    "tprs_upper = np.minimum(mean_tpr + std_tpr, 1)\n",
    "tprs_lower = np.maximum(mean_tpr - std_tpr, 0)\n",
    "plt.fill_between(mean_fpr, tprs_lower, tprs_upper, color='grey', alpha=.2,\n",
    "                 label=r'$\\pm$ 1 std. dev.')\n",
    "\n",
    "plt.xlim([-0.05, 1.05])\n",
    "plt.ylim([-0.05, 1.05])\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('Receiver operating characteristic example')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "total_time = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Time: 0.0h:0.0m:21.26303744316101s\n",
      "Total time elapsed: 0.0h:0.0m:21.559102535247803s\n"
     ]
    }
   ],
   "source": [
    "print(\"Training Time: {}h:{}m:{}s\".format(\n",
    "            training_time//3600, (training_time//60)%60, training_time%60))\n",
    "print(\"Total time elapsed: {}h:{}m:{}s\".format(\n",
    "            total_time//3600, (total_time//60)%60, total_time%60))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "quit()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python (postal)",
   "language": "python",
   "name": "postal"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}