Supervised classification.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h1>Supervised Learning on the labelled data</h1>\n",
    "<h3>Here, we will use the supervised classification technique for the binary classification of the image patches extracted using RST as CMB or Non-CMB. The classifier used here will be Support Vector Machine(SVM) with Radial Basis Function(RBF) kernel.</h3>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Current Working Directory: /Users/shwetankpanwar/Documents/BTPData/python_scripts/cmb_segmentation\n",
      "The file structure creation failed. Structure already exists!\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from pfca import init_path\n",
    "from pfca.visuals.img2D import imgplot\n",
    "from pfca.core.preprocessing import *\n",
    "from pfca.core.processing import *\n",
    "from pfca import file_read as f\n",
    "import ants\n",
    "import skimage.feature as feature\n",
    "from time import time\n",
    "from pfca.exp.results import roi_snipping\n",
    "from pfca.exp.dataset import *\n",
    "from pfca import init_path"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "raw_dir, nifti_dir = init_path()\n",
    "cur_path = os.getcwd()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<p>After the initialization, we will import the data stored in the pickle files and merge them in order to increase our sample size</p>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def import_dataset(dataset):\n",
    "    import os\n",
    "    import pandas as pd\n",
    "    cur_path = os.getcwd()\n",
    "    ds = pd.read_pickle(cur_path + '/datasets/learning/' + dataset)\n",
    "    return ds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "pdataset = import_dataset('rst_output_all_subjects_2_2_0.1')          #Data from positive RST\n",
    "ndataset = import_dataset('rst_output_all_subjects_negative_2_2_0.1') #Data from negative RST"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "labelled_data = pdataset.append(ndataset, ignore_index=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Total no of labelled datapoints: 2599\n",
      "Total no of target datapoints: 123\n",
      "Total no of non-target datapoints: 2476\n"
     ]
    }
   ],
   "source": [
    "print(\"Total no of labelled datapoints: \" + str(len(labelled_data)))\n",
    "print(\"Total no of target datapoints: \" + str(len(labelled_data.loc[\n",
    "    labelled_data['label'] == 1\n",
    "])))\n",
    "print(\"Total no of non-target datapoints: \" + str(len(labelled_data.loc[\n",
    "    labelled_data['label'] == 0\n",
    "])))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h3>First, we will try to evaluate the performance on the original non-augmented dataset and see how our classifier performs</h3>\n",
    "<p>The results will be used to see if the data augmentation improves performance or not.</p>\n",
    "<p>The evaluation will be done based on k-fold cross validation</p>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Function to perform k-fold splitting in the dataset\n",
    "def k_fold_split(dataset, k, balanced = True):\n",
    "        import pandas as pd\n",
    "        import numpy as np\n",
    "        ds = dataset.copy()\n",
    "        targets = ds.loc[ds['label'] == 1]         #Target data\n",
    "        non_targets = ds.loc[ds['label'] == 0]     #Non target data\n",
    "\n",
    "        n_targets = len(targets)\n",
    "        n_non_t = len(non_targets)\n",
    "\n",
    "        n_comm = min(n_targets, n_non_t)\n",
    "        list_target = targets.index.values\n",
    "        list_non_targets = non_targets.index.values\n",
    "\n",
    "        if balanced == True:  # It will produce a balanced class with equal no of targets and non-targets\n",
    "            # Decategorizing data in case of appending\n",
    "            n_session = int(n_comm/k)         # no of target points to select\n",
    "            ds['session'] = np.nan\n",
    "\n",
    "            trainees = np.random.choice(list_target, size=(k, n_session), replace=False)\n",
    "            trainees2 = np.random.choice(list_non_targets, size=(k, n_session), replace=False)\n",
    "\n",
    "            for i in range(k):\n",
    "                all_train = np.append(trainees[i,:], trainees2[i,:])\n",
    "\n",
    "                ds.at[all_train, ['session']] = 'fold_' + str(i+1)  # Labelling the datapoints as training points\n",
    "            return ds, all_train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "splitted_data, lst = k_fold_split(labelled_data, 4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Total no of datapoints in one fold session: 30\n"
     ]
    }
   ],
   "source": [
    "print(\"Total no of datapoints in one fold session: \" + str(len(splitted_data.loc[\n",
    "    (splitted_data['session'] == 'fold_4') & (splitted_data['label'] == 1)\n",
    "])))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Adding the raw feature column in the dataframe\n",
    "import sys"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h3>Data Normalization</h3>\n",
    "<p>Because non-normalized data can mess up with the transfer function of our non-linear classifier</p>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>RST_peak</th>\n",
       "      <th>image_patch</th>\n",
       "      <th>label</th>\n",
       "      <th>patient_name</th>\n",
       "      <th>session</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>[144, 181, 99]</td>\n",
       "      <td>[[[1272.2008, 1251.0284, 1266.076, 1333.9814, ...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>fold_2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>[141, 107, 39]</td>\n",
       "      <td>[[[0.0, 0.0, 79.63042, 351.2794, 523.55176, 65...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>[138, 120, 27]</td>\n",
       "      <td>[[[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 168...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>[136, 134, 139]</td>\n",
       "      <td>[[[1643.6931, 1681.0558, 1701.4177, 1736.2515,...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>[126, 79, 71]</td>\n",
       "      <td>[[[1812.4193, 1821.4623, 1826.1653, 1805.4312,...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>[125, 76, 75]</td>\n",
       "      <td>[[[1681.3533, 1696.9663, 1738.3156, 1780.3088,...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>[122, 82, 80]</td>\n",
       "      <td>[[[1856.1481, 1847.7457, 1856.007, 1877.7863, ...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>[120, 131, 40]</td>\n",
       "      <td>[[[1264.1499, 1140.9651, 1209.2004, 1521.9618,...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>[119, 139, 54]</td>\n",
       "      <td>[[[946.6912, 1141.942, 1500.9038, 2212.3945, 2...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>[119, 85, 80]</td>\n",
       "      <td>[[[1805.539, 1808.1421, 1817.4318, 1832.182, 1...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          RST_peak                                        image_patch  label  \\\n",
       "0   [144, 181, 99]  [[[1272.2008, 1251.0284, 1266.076, 1333.9814, ...      0   \n",
       "1   [141, 107, 39]  [[[0.0, 0.0, 79.63042, 351.2794, 523.55176, 65...      0   \n",
       "2   [138, 120, 27]  [[[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 168...      0   \n",
       "3  [136, 134, 139]  [[[1643.6931, 1681.0558, 1701.4177, 1736.2515,...      0   \n",
       "4    [126, 79, 71]  [[[1812.4193, 1821.4623, 1826.1653, 1805.4312,...      0   \n",
       "5    [125, 76, 75]  [[[1681.3533, 1696.9663, 1738.3156, 1780.3088,...      0   \n",
       "6    [122, 82, 80]  [[[1856.1481, 1847.7457, 1856.007, 1877.7863, ...      0   \n",
       "7   [120, 131, 40]  [[[1264.1499, 1140.9651, 1209.2004, 1521.9618,...      0   \n",
       "8   [119, 139, 54]  [[[946.6912, 1141.942, 1500.9038, 2212.3945, 2...      0   \n",
       "9    [119, 85, 80]  [[[1805.539, 1808.1421, 1817.4318, 1832.182, 1...      0   \n",
       "\n",
       "  patient_name session  \n",
       "0           12  fold_2  \n",
       "1           12     NaN  \n",
       "2           12     NaN  \n",
       "3           12     NaN  \n",
       "4           12     NaN  \n",
       "5           12     NaN  \n",
       "6           12     NaN  \n",
       "7           12     NaN  \n",
       "8           12     NaN  \n",
       "9           12     NaN  "
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "splitted_data[0:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "#normalize the data based on their datatype\n",
    "#Lets know what's the datatype of the image data\n",
    "def normalize_image(dataset):\n",
    "    k = dataset.copy()\n",
    "    for i in range(len(k)):\n",
    "        max_pix = np.max(k.iloc[i]['image_patch'])\n",
    "        min_pix = np.min(k.iloc[i]['image_patch'])\n",
    "        k.loc[i , ['image_patch']] = (k.loc[i,['image_patch']] - min_pix)/(max_pix - min_pix)\n",
    "    return k    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "#normalized_dataset\n",
    "normalized_dataset = normalize_image(splitted_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>RST_peak</th>\n",
       "      <th>image_patch</th>\n",
       "      <th>label</th>\n",
       "      <th>patient_name</th>\n",
       "      <th>session</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>[144, 181, 99]</td>\n",
       "      <td>[[[0.6995006, 0.6878593, 0.696133, 0.7334698, ...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>fold_2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>[141, 107, 39]</td>\n",
       "      <td>[[[0.0, 0.0, 0.04216997, 0.18602741, 0.2772579...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>[138, 120, 27]</td>\n",
       "      <td>[[[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>[136, 134, 139]</td>\n",
       "      <td>[[[0.69480616, 0.7105998, 0.719207, 0.73393154...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>[126, 79, 71]</td>\n",
       "      <td>[[[0.70013624, 0.7052303, 0.7078795, 0.6961996...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>[125, 76, 75]</td>\n",
       "      <td>[[[0.7176489, 0.7276675, 0.7542006, 0.78114694...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>[122, 82, 80]</td>\n",
       "      <td>[[[0.7279448, 0.72366834, 0.72787297, 0.738957...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>[120, 131, 40]</td>\n",
       "      <td>[[[0.5096289, 0.42495853, 0.4718597, 0.6868344...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>[119, 139, 54]</td>\n",
       "      <td>[[[0.11030873, 0.1833937, 0.31775787, 0.584078...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>[119, 85, 80]</td>\n",
       "      <td>[[[0.8183661, 0.8199103, 0.82542056, 0.8341700...</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          RST_peak                                        image_patch  label  \\\n",
       "0   [144, 181, 99]  [[[0.6995006, 0.6878593, 0.696133, 0.7334698, ...      0   \n",
       "1   [141, 107, 39]  [[[0.0, 0.0, 0.04216997, 0.18602741, 0.2772579...      0   \n",
       "2   [138, 120, 27]  [[[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0...      0   \n",
       "3  [136, 134, 139]  [[[0.69480616, 0.7105998, 0.719207, 0.73393154...      0   \n",
       "4    [126, 79, 71]  [[[0.70013624, 0.7052303, 0.7078795, 0.6961996...      0   \n",
       "5    [125, 76, 75]  [[[0.7176489, 0.7276675, 0.7542006, 0.78114694...      0   \n",
       "6    [122, 82, 80]  [[[0.7279448, 0.72366834, 0.72787297, 0.738957...      0   \n",
       "7   [120, 131, 40]  [[[0.5096289, 0.42495853, 0.4718597, 0.6868344...      0   \n",
       "8   [119, 139, 54]  [[[0.11030873, 0.1833937, 0.31775787, 0.584078...      0   \n",
       "9    [119, 85, 80]  [[[0.8183661, 0.8199103, 0.82542056, 0.8341700...      0   \n",
       "\n",
       "  patient_name session  \n",
       "0           12  fold_2  \n",
       "1           12     NaN  \n",
       "2           12     NaN  \n",
       "3           12     NaN  \n",
       "4           12     NaN  \n",
       "5           12     NaN  \n",
       "6           12     NaN  \n",
       "7           12     NaN  \n",
       "8           12     NaN  \n",
       "9           12     NaN  "
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "normalized_dataset[0:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "def k_fold_training(dataset, k, classifier):\n",
    "    import pfca.exp.dataset\n",
    "    import pfca.exp.results \n",
    "    import numpy as np\n",
    "    from sklearn.metrics import confusion_matrix\n",
    "    from matplotlib import pyplot as plt\n",
    "    #Rather than rewriting the code, we will be utilizing the same code written for data_dictionary class\n",
    "    datas = pfca.exp.dataset.data_dictionary()\n",
    "    datas.dataset = splitted_data\n",
    "    datas.splitted = True           #So that feature_matrix_raw() method works\n",
    "    datas.dataset = datas.dataset.rename(columns = {\"image_patch\" : \"image\"})\n",
    "    for i in range(k):\n",
    "        print(\"Starting the testing on fold_\" + str(i+1) + \" data.\")\n",
    "        ###########Data Acquisition##########################\n",
    "        ks_t = np.arange(k)\n",
    "        y_t = np.array([])\n",
    "        ks_values = np.delete(ks_t, [i], None)\n",
    "        x_t, _ = datas.feature_matrix_raw('fold_' + str(i+1))\n",
    "        x_zero = np.zeros((1, x_t.shape[1]))\n",
    "        \n",
    "        for j in ks_values:\n",
    "            x_temp, y_temp = datas.feature_matrix_raw('fold_' + str(j+1))\n",
    "            x_zero = np.vstack((x_zero, x_temp))\n",
    "            y_t = np.append(y_t, y_temp)    \n",
    "        \n",
    "        X = np.delete(x_zero, 0, 0)\n",
    "        Y = y_t\n",
    "        print(X.shape)\n",
    "        print(Y.shape)\n",
    "        X_test, Y_test = datas.feature_matrix_raw('fold_' + str(i+1))\n",
    "        #######################################################\n",
    "        #sys.exit()\n",
    "        classifier.fit(X, Y)\n",
    "        y_pred = classifier.predict(X_test)\n",
    "        tn, fp, fn, tp = (confusion_matrix(Y_test, y_pred)).ravel()\n",
    "\n",
    "        confusion_mat = np.array([[tp, fp],\n",
    "                                  [fn, tn]])\n",
    "        fig, ax = plt.subplots()\n",
    "\n",
    "        im, cbar = pfca.exp.results.heatmap(confusion_mat, ['Predicted True', 'Predicted False'], ['Actual True', 'Actual False'], ax=ax,\n",
    "                           cmap=\"Blues\", cbarlabel=\" [No. of samples]\")\n",
    "        texts = pfca.exp.results.annotate_heatmap(im, valfmt=\"{x:.2f}\")\n",
    "        fig.tight_layout()\n",
    "        plt.show()\n",
    "        print(\"Sensitivity : \", (tp / (fn + tp)))\n",
    "        print(\"Specificity : \", (fp / (fp + tn)))\n",
    "    return classifier    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Starting the testing on fold_1 data.\n",
      "(60, 1000)\n",
      "(60,)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Sensitivity :  1.0\n",
      "Specificity :  1.0\n",
      "Starting the testing on fold_2 data.\n",
      "(60, 1000)\n",
      "(60,)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Sensitivity :  1.0\n",
      "Specificity :  1.0\n"
     ]
    }
   ],
   "source": [
    "from sklearn.svm import SVC\n",
    "svm_classifier = SVC(gamma = 'auto')\n",
    "trained_svm = k_fold_training(normalized_dataset, 2, svm_classifier)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<h3>Trying to develop a general feature matrix for testing the classifiers in MATLAB</h3>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "from pfca.exp.dataset import *\n",
    "datas = data_dictionary()\n",
    "datas.dataset = labelled_data\n",
    "datas.splitted = True           #So that feature_matrix_raw() method works\n",
    "datas.dataset = datas.dataset.rename(columns = {\"image_patch\" : \"image\"})\n",
    "datas.dataset['session'] = 'whole'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "X, Y = datas.feature_matrix_raw('whole')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2599, 1000)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2599,)"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Y.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "#Saving the data in .mat files for loading them on MATLAB\n",
    "from scipy.io import savemat"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "feature_matrix = {}\n",
    "feature_matrix['feature_matrix'] = X\n",
    "labels = {}\n",
    "labels['label'] = Y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "savemat('feature_matrix.mat', feature_matrix)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "savemat('labels.mat', labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}