MS using SVM on FLAIR.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, LinearSVC\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, normalize_float, shuffle_data"
   ]
  },
  {
   "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": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# load hdf5 files and extract columns\n",
    "train_h5 = h5py.File('/analysis/ritter/Dataset/MS/CIS/train_dataset_FLAIR_lesions_filled.h5', 'r')\n",
    "holdout_h5 = h5py.File('/analysis/ritter/Dataset/MS/CIS/holdout_dataset_FLAIR_lesions_filled.h5', 'r')\n",
    "\n",
    "X_train, y_train = train_h5['X'], train_h5['y']\n",
    "X_holdout, y_holdout = holdout_h5['X'], holdout_h5['y']\n",
    "\n",
    "train_lesions_h5 = h5py.File('/analysis/ritter/Dataset/MS/CIS/train_dataset_lesions.h5', 'r')\n",
    "holdout_lesions_h5 = h5py.File('/analysis/ritter/Dataset/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": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# convert data to numpy arrays\n",
    "X_train = np.array(X_train, dtype=dtype)\n",
    "y_train = np.array(y_train)\n",
    "\n",
    "X_holdout = np.array(X_holdout, dtype=dtype)\n",
    "y_holdout = np.array(y_holdout)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "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": 7,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "plt.imshow(X_train[0][:,:,49], cmap='gray')\n",
    "plt.title(y_train[0])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Normalization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "plt.hist(X_train[0].ravel())\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# try 0-1 scaling\n",
    "if zero_one_normalize:\n",
    "    X_train = normalize_float(X_train)\n",
    "    X_holdout = normalize_float(X_holdout)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "plt.hist(X_train[0].ravel())\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# flatten image dimensions\n",
    "X_train = X_train.reshape(X_train.shape[0], -1)\n",
    "X_holdout = X_holdout.reshape(X_holdout.shape[0], -1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(124, 1050624)"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "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",
    "    param_grid = {'svc__C': Cs}\n",
    "    # use standard scaler for preprocessing\n",
    "    scaler = preprocessing.StandardScaler(with_mean=True, with_std=False)\n",
    "    pipeline = make_pipeline(scaler, SVC(kernel='linear', gamma='auto'))\n",
    "    grid_search = GridSearchCV(pipeline, param_grid, cv=n_folds, n_jobs=5)\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": 14,
   "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"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best parameter value: {'svc__C': 0.001}\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.04 %\n",
      "Sensitivity 96.08 %\n",
      "Specificity 100.00 %\n",
      "######## Validation set results ########\n",
      "Balanced accuracy 79.17 %\n",
      "Sensitivity 75.00 %\n",
      "Specificity 83.33 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 86.11\n",
      "Fold 1\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": [
      "Best parameter value: {'svc__C': 0.001}\n",
      "######## Training set results ########\n",
      "Balanced accuracy 100.00 %\n",
      "Sensitivity 100.00 %\n",
      "Specificity 100.00 %\n",
      "######## Validation set results ########\n",
      "Balanced accuracy 72.08 %\n",
      "Sensitivity 71.43 %\n",
      "Specificity 72.73 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 68.83\n",
      "Fold 2\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": [
      "Best parameter value: {'svc__C': 0.001}\n",
      "######## Training set results ########\n",
      "Balanced accuracy 100.00 %\n",
      "Sensitivity 100.00 %\n",
      "Specificity 100.00 %\n",
      "######## Validation set results ########\n",
      "Balanced accuracy 34.62 %\n",
      "Sensitivity 0.00 %\n",
      "Specificity 69.23 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 27.69\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"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best parameter value: {'svc__C': 0.001}\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.08 %\n",
      "Sensitivity 96.15 %\n",
      "Specificity 100.00 %\n",
      "######## Validation set results ########\n",
      "Balanced accuracy 53.25 %\n",
      "Sensitivity 63.64 %\n",
      "Specificity 42.86 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 63.64\n",
      "Fold 4\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": [
      "Best parameter value: {'svc__C': 0.001}\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.18 %\n",
      "Sensitivity 96.36 %\n",
      "Specificity 100.00 %\n",
      "######## Validation set results ########\n",
      "Balanced accuracy 53.75 %\n",
      "Sensitivity 37.50 %\n",
      "Specificity 70.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 65.00\n",
      "Fold 5\n",
      "Best parameter value: {'svc__C': 0.001}\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.15 %\n",
      "Sensitivity 96.30 %\n",
      "Specificity 100.00 %\n",
      "######## Validation set results ########\n",
      "Balanced accuracy 64.58 %\n",
      "Sensitivity 66.67 %\n",
      "Specificity 62.50 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 63.89\n",
      "Fold 6\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": [
      "Best parameter value: {'svc__C': 0.001}\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.08 %\n",
      "Sensitivity 96.15 %\n",
      "Specificity 100.00 %\n",
      "######## Validation set results ########\n",
      "Balanced accuracy 51.52 %\n",
      "Sensitivity 36.36 %\n",
      "Specificity 66.67 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 71.21\n",
      "Training Time: 0.0h:33.0m:58.539198875427246s\n",
      "######## Final results ########\n",
      "Validation balanced accuracies: \n",
      " [0.7916666666666667, 0.7207792207792207, 0.34615384615384615, 0.5324675324675324, 0.5375, 0.6458333333333333, 0.5151515151515151]\n",
      "Validation balanced accuracies mean: 0.5842217306503021\n",
      "Validation final sensitivities: \n",
      " [0.75, 0.7142857142857143, 0.0, 0.6363636363636364, 0.375, 0.6666666666666666, 0.36363636363636365]\n",
      "Validation final sensitivities' mean: 0.5008503401360545\n",
      "Validation final specificities: \n",
      " [0.8333333333333334, 0.7272727272727273, 0.6923076923076923, 0.42857142857142855, 0.7, 0.625, 0.6666666666666666]\n",
      "Validation final specificities' mean: 0.6675931211645497\n",
      "Mean ROC AUC score 63.77\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 parameter value: {}\".format(best_param))\n",
    "    \n",
    "    scaler = preprocessing.StandardScaler(with_mean=True, with_std=False)\n",
    "    model = SVC(kernel=\"linear\", C=best_param[\"svc__C\"])\n",
    "    pipeline = make_pipeline(scaler, model)\n",
    "    pipeline.fit(X_train[train_idx], y_train[train_idx])\n",
    "    \n",
    "    # training set results\n",
    "    train_pred = pipeline.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 = pipeline.predict(X_train[test_idx])\n",
    "    val_scores = pipeline.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": 15,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "training_time = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "counter = {}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "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": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'svc__C': '0.001'}\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": 19,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Trial 0\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.41 %\n",
      "Sensitivity 96.83 %\n",
      "Specificity 100.00 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 66.92 %\n",
      "Sensitivity 53.85 %\n",
      "Specificity 80.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 66.92\n",
      "Trial 1\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.41 %\n",
      "Sensitivity 96.83 %\n",
      "Specificity 100.00 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 66.92 %\n",
      "Sensitivity 53.85 %\n",
      "Specificity 80.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 66.92\n",
      "Trial 2\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.41 %\n",
      "Sensitivity 96.83 %\n",
      "Specificity 100.00 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 66.92 %\n",
      "Sensitivity 53.85 %\n",
      "Specificity 80.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 66.92\n",
      "Trial 3\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.41 %\n",
      "Sensitivity 96.83 %\n",
      "Specificity 100.00 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 66.92 %\n",
      "Sensitivity 53.85 %\n",
      "Specificity 80.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 66.92\n",
      "Trial 4\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.41 %\n",
      "Sensitivity 96.83 %\n",
      "Specificity 100.00 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 66.92 %\n",
      "Sensitivity 53.85 %\n",
      "Specificity 80.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 66.92\n",
      "Trial 5\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.41 %\n",
      "Sensitivity 96.83 %\n",
      "Specificity 100.00 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 66.92 %\n",
      "Sensitivity 53.85 %\n",
      "Specificity 80.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 66.92\n",
      "Trial 6\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.41 %\n",
      "Sensitivity 96.83 %\n",
      "Specificity 100.00 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 66.92 %\n",
      "Sensitivity 53.85 %\n",
      "Specificity 80.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 66.92\n",
      "Trial 7\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.41 %\n",
      "Sensitivity 96.83 %\n",
      "Specificity 100.00 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 66.92 %\n",
      "Sensitivity 53.85 %\n",
      "Specificity 80.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 66.92\n",
      "Trial 8\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.41 %\n",
      "Sensitivity 96.83 %\n",
      "Specificity 100.00 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 66.92 %\n",
      "Sensitivity 53.85 %\n",
      "Specificity 80.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 66.92\n",
      "Trial 9\n",
      "######## Training set results ########\n",
      "Balanced accuracy 98.41 %\n",
      "Sensitivity 96.83 %\n",
      "Specificity 100.00 %\n",
      "######## Holdout set results ########\n",
      "Balanced accuracy 66.92 %\n",
      "Sensitivity 53.85 %\n",
      "Specificity 80.00 %\n",
      "Area Under the Receiver Operating Curve (ROC AUC score) 66.92\n",
      "Training Time: 0.0h:51.0m:46.11532998085022s\n",
      "######## Final results ########\n",
      "Holdout balanced accuracies: \n",
      " [0.6692307692307693, 0.6692307692307693, 0.6692307692307693, 0.6692307692307693, 0.6692307692307693, 0.6692307692307693, 0.6692307692307693, 0.6692307692307693, 0.6692307692307693, 0.6692307692307693]\n",
      "Holdout balanced accuracies mean: 0.6692307692307693\n",
      "Holdout final sensitivities: \n",
      " [0.5384615384615384, 0.5384615384615384, 0.5384615384615384, 0.5384615384615384, 0.5384615384615384, 0.5384615384615384, 0.5384615384615384, 0.5384615384615384, 0.5384615384615384, 0.5384615384615384]\n",
      "Holdout final sensitivities' mean: 0.5384615384615384\n",
      "Holdout final specificities: \n",
      " [0.8, 0.8, 0.8, 0.8, 0.8, 0.8, 0.8, 0.8, 0.8, 0.8]\n",
      "Holdout final specificities' mean: 0.8\n",
      "Mean ROC AUC score 66.92\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 = \"linear\"\n",
    "C = float(majority_param[\"svc__C\"])\n",
    "scaler = preprocessing.StandardScaler(with_mean=True, with_std=False)\n",
    "model = SVC(kernel=kernel, C=C, cache_size=2000)\n",
    "pipeline = make_pipeline(scaler, model)\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\n",
    "    X_train, y_train = shuffle_data(X_train, y_train)\n",
    "    \n",
    "    # Start training\n",
    "    pipeline.fit(X_train, y_train)\n",
    "    # training set results\n",
    "    train_pred = pipeline.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 = pipeline.predict(X_holdout)\n",
    "    holdout_scores = pipeline.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": 20,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "total_time = time.time() - start_time"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Time: 0.0h:51.0m:46.11532998085022s\n",
      "Total time elapsed: 0.0h:51.0m:46.45267581939697s\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": 22,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "quit()"
   ]
  }
 ],
 "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
}