svm_data_converter.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import scipy.io\n",
    "import random\n",
    "import csv\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.cluster import KMeans\n",
    "from mpl_toolkits import mplot3d\n",
    "from sklearn.decomposition import PCA\n",
    "from sklearn.discriminant_analysis import LinearDiscriminantAnalysis\n",
    "import os, sys, traceback, getpass, time, re"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 134,
   "metadata": {},
   "outputs": [],
   "source": [
    "def calculate_stat(mat, subject, feature_len):\n",
    "    means = []\n",
    "    stds = []\n",
    "    for i in range(feature_len):\n",
    "        a = []\n",
    "        for j in range(len(mat[0][subject])): #clip\n",
    "            a.append(mat[0][subject][j][-feature_len + i])\n",
    "        a = np.nan_to_num(a)\n",
    "        means.append(np.mean(a, axis=0))\n",
    "        stds.append(np.std(a, axis=0))\n",
    "    return means, stds\n",
    "\n",
    "def standarize_and_clamp(arr, mean, std):\n",
    "    arr = (arr - mean)/std\n",
    "    arr = np.clip(arr, -1, 1)\n",
    "    return arr\n",
    "\n",
    "def normalize(arr):\n",
    "    arr = (arr - np.min(arr))/(np.max(arr) - np.min(arr))\n",
    "    return arr\n",
    "\n",
    "def get_arousal_class(rate, mean):\n",
    "    if rate < mean:\n",
    "        return '-1'\n",
    "    else:\n",
    "        return '+1'\n",
    "\n",
    "def get_valence_class(rate, mean):\n",
    "    if rate < mean:\n",
    "        return '-1'\n",
    "    else:\n",
    "        return '+1'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\User\\AppData\\Local\\conda\\conda\\envs\\tensorflow\\lib\\site-packages\\numpy\\core\\_methods.py:116: RuntimeWarning: overflow encountered in multiply\n",
      "  x = um.multiply(x, x, out=x)\n",
      "C:\\Users\\User\\AppData\\Local\\conda\\conda\\envs\\tensorflow\\lib\\site-packages\\ipykernel_launcher.py:14: RuntimeWarning: invalid value encountered in double_scalars\n",
      "  \n",
      "C:\\Users\\User\\AppData\\Local\\conda\\conda\\envs\\tensorflow\\lib\\site-packages\\numpy\\core\\_methods.py:70: RuntimeWarning: overflow encountered in reduce\n",
      "  ret = umr_sum(arr, axis, dtype, out, keepdims)\n",
      "C:\\Users\\User\\AppData\\Local\\conda\\conda\\envs\\tensorflow\\lib\\site-packages\\numpy\\core\\_methods.py:102: RuntimeWarning: overflow encountered in reduce\n",
      "  arrmean = umr_sum(arr, axis, dtype, keepdims=True)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "features = {'ECG':[], 'GSR':[], 'EEG':[], 'ResNet':[] }\n",
    "\n",
    "# ECG feature\n",
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_ECGFeatures.mat')\n",
    "mat = mat['ECGFeatures_58']\n",
    "mat = np.nan_to_num(mat)\n",
    "tmp_mat = np.zeros((1, 58, 36, 14))\n",
    "for i in range(len(mat[0])): # subject\n",
    "    means, stds = calculate_stat(mat, subject=i, feature_len=14)\n",
    "    for j in range(len(mat[0][i])): # clip\n",
    "        for k in range(len(mat[0][i][j][18:])): # feature\n",
    "            tmp_mat[0][i][j][k] = standarize_and_clamp(mat[0][i][j][-14+k], means[k], stds[k])\n",
    "mat = tmp_mat\n",
    "features['ECG'] = mat\n",
    "plt.plot([x for x in range(len(features['ECG'][0][0][33]))], features['ECG'][0][0][33])\n",
    "plt.show()\n",
    "\n",
    "\n",
    "# GSR feature\n",
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_GSRFeatures.mat')\n",
    "mat = mat['GSRFeatures_58']\n",
    "mat = np.nan_to_num(mat)\n",
    "tmp_mat = np.zeros((1, 58, 36, 31))\n",
    "for i in range(len(mat[0])): # subject\n",
    "    means, stds = calculate_stat(mat, subject=i,feature_len=31)\n",
    "    for j in range(len(mat[0][i])): # clip\n",
    "        for k in range(31):\n",
    "            tmp_mat[0][i][j][k] = standarize_and_clamp(mat[0][i][j][k], means[k], stds[k])\n",
    "mat = tmp_mat\n",
    "features['GSR'] = mat\n",
    "plt.plot([x for x in range(len(features['GSR'][0][0][33]))], features['GSR'][0][0][33])\n",
    "plt.show()\n",
    "\n",
    "\n",
    "# EEG feature\n",
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_EEGFeatures.mat')\n",
    "mat = mat['EEGFeatures_58']\n",
    "mat = np.nan_to_num(mat)\n",
    "tmp_mat = np.zeros((1, 58, 36, 88))\n",
    "for i in range(len(mat[0])): # subject\n",
    "    means, stds = calculate_stat(mat, subject=i, feature_len=88)\n",
    "    for j in range(len(mat[0][i])): # clip\n",
    "        for k in range(88): # feature\n",
    "            tmp_mat[0][i][j][k] = standarize_and_clamp(mat[0][i][j][k], means[k], stds[k])\n",
    "mat = tmp_mat\n",
    "features['EEG'] = mat\n",
    "plt.plot([x for x in range(len(features['EEG'][0][0][33]))], features['EEG'][0][0][33])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x1fb83b19080>]"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# ResNet EEG feature\n",
    "data = np.load('EEG_ResNet_Features.npz')\n",
    "data = data['features']\n",
    "for i in range(2088):\n",
    "    data[i][0] = normalize(data[i][0])\n",
    "features['ResNet'] = data\n",
    "plt.plot([x for x in range(2048)], features['ResNet'][33][0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "metadata": {},
   "outputs": [],
   "source": [
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_Order_Movie.mat')\n",
    "order = mat['PermutationList']\n",
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_SelfReports.mat')\n",
    "rating = mat['Ratings']\n",
    "\n",
    "features['ECG'] = np.nan_to_num(features['ECG'])\n",
    "features['GSR'] = np.nan_to_num(features['GSR'])\n",
    "features['EEG'] = np.nan_to_num(features['EEG'])\n",
    "features['ResNet'] = np.nan_to_num(features['ResNet'])\n",
    "\n",
    "#clusters = [0, 1, 1, 1, 1, 1, 0, 1, 2, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 0, 0, 2, 1, 0, 1, 1, 1, 1, 2, 1, 0, 0, 1, 1, 2, 1, 0,\n",
    " #2, 0, 2, 2, 2, 2, 0, 1, 2, 1, 1, 1, 0, 2, 2, 0, 1, 1, 0, 0, 0]\n",
    "\n",
    "clusters = [2, 0, 1, 2, 2, 2, 2, 2, 1, 1, 1, 0, 0, 0, 2, 2, 1, 2, 1, 1, 0, 1, 0, 2, 1, 0, 0, 1, 2, 2, 2, 2, 2, 2, 0, 0, 0,\n",
    " 1, 2, 2, 0, 2, 1, 2, 0, 0, 1]\n",
    "\n",
    "f = open('svmlib_dataset/train_subjects_id.csv')\n",
    "rows = csv.reader(f)\n",
    "for row in rows:\n",
    "    rows = row\n",
    "    break\n",
    "train_subjects_list = np.array(rows).astype(int)\n",
    "f.close()\n",
    "train_subjects_list -= 1\n",
    "\n",
    "test_subjects_list = np.array([2, 21, 22, 24, 25, 27, 28, 31, 32, 42, 49])\n",
    "\n",
    "f = open('svmlib_dataset/features_permutation.csv')\n",
    "rows = csv.reader(f)\n",
    "for row in rows:\n",
    "    rows = row\n",
    "    break\n",
    "feature_sequence = np.array(rows).astype(int)\n",
    "f.close()\n",
    "\n",
    "f = open('svmlib_dataset/samples_permutation_clip_wise.csv')\n",
    "rows = csv.reader(f)\n",
    "for row in rows:\n",
    "    rows = row\n",
    "    break\n",
    "train_sample_sequence = np.array(rows).astype(int)\n",
    "f.close()\n",
    "\n",
    "f = open('svmlib_dataset/train_clips_id.csv')\n",
    "rows = csv.reader(f)\n",
    "for row in rows:\n",
    "    rows = row\n",
    "    break\n",
    "train_clips_list = np.array(rows).astype(int)\n",
    "f.close()\n",
    "\n",
    "f = open('svmlib_dataset/selected_feature.csv')\n",
    "rows = csv.reader(f)\n",
    "for row in rows:\n",
    "    rows = row\n",
    "    break\n",
    "selected_feature = np.array(rows).astype(int)\n",
    "selected_feature -= 1\n",
    "f.close()\n",
    "\n",
    "'''\n",
    "0 ~ 13: ECG\n",
    "14 ~ 44: GSR\n",
    "45 ~ 132: EEG\n",
    "'''\n",
    "ecgs = []\n",
    "gsrs = []\n",
    "eegs = []\n",
    "\n",
    "# dataset for arousal classifier feature selection\n",
    "subjects = 58\n",
    "instances = 2088\n",
    "clips = 36\n",
    "ones = 0\n",
    "zeros = 0\n",
    "txts_train = []\n",
    "\n",
    "ys = []\n",
    "for i in range(58):\n",
    "    for clip_index in range(33):\n",
    "        #if(clip_index in train_clips_list): continue\n",
    "#         if (i*36 + clip_index) not in train_sample_sequence:\n",
    "#             continue\n",
    "        feature_index = 0\n",
    "        txt = ''\n",
    "        txt += get_valence_class(rating[1][i][order[i][clip_index]-1]) # label\n",
    "        ys.append([int(get_arousal_class(rating[0][i][order[i][clip_index]-1])), int(get_valence_class(rating[1][i][order[i][clip_index]-1]))])\n",
    "        true_feature_index = 0\n",
    "        #print(get_arousal_class(rating[0][i][order[i][clip_index]-1]))\n",
    "        if get_valence_class(rating[1][i][order[i][clip_index]-1])=='+1':\n",
    "            ones += 1\n",
    "        else:\n",
    "            zeros += 1\n",
    "        feature_seq = []\n",
    "        \n",
    "        tmp = []\n",
    "        # ECG\n",
    "        for j in range(14):\n",
    "            tmp.append(features['ECG'][0][i][clip_index][j])\n",
    "        ecgs.append(tmp)\n",
    "        tmp = [tmp]\n",
    "        #tmptmp = pca_ecg.transform(tmp)\n",
    "        tmptmp = tmp\n",
    "        for ecg in tmptmp[0]:\n",
    "            feature_seq.append(ecg)\n",
    "        \n",
    "        \n",
    "        tmp = []\n",
    "        # GSR\n",
    "        for j in range(31):\n",
    "            tmp.append(features['GSR'][0][i][clip_index][j])\n",
    "        gsrs.append(tmp)\n",
    "        tmp = [tmp]\n",
    "        #tmptmp = pca_gsr.transform(tmp)\n",
    "        tmptmp = tmp\n",
    "        for gsr in tmptmp[0]:\n",
    "            feature_seq.append(gsr)\n",
    "            \n",
    "        tmp = []\n",
    "        # EEG\n",
    "        for j in range(88):\n",
    "            tmp.append(features['EEG'][0][i][clip_index][j])\n",
    "        eegs.append(tmp)\n",
    "        tmp = [tmp]\n",
    "        #tmptmp = pca_eeg.transform(tmp)\n",
    "        tmptmp = tmp\n",
    "        for eeg in tmptmp[0]:\n",
    "            feature_seq.append(eeg)\n",
    "        \n",
    "        \n",
    "#         f = [0.0]*133\n",
    "#         for j in range(len(feature_sequence)):\n",
    "#             f[j] = feature_seq[feature_sequence[j]]\n",
    "        \n",
    "        random.shuffle(feature_seq)\n",
    "        for j in range(len(feature_seq)): # selected_feature\n",
    "            txt += ' '\n",
    "            txt += str(feature_index+1) + ':' + str(feature_seq[j])\n",
    "            feature_index += 1    \n",
    "            \n",
    "        txt += '\\n'\n",
    "    \n",
    "        txts_train.append(txt)\n",
    "        \n",
    "random.shuffle(txts_train)\n",
    "        \n",
    "#print(zeros/(ones + zeros))\n",
    "# f = open('svmlib_dataset/valence_train_sample_wise_pca', \"w+\")\n",
    "# for i in range(1671): # len(train_sample_sequence) len(txts_train)\n",
    "#     f.write(txts_train[i]) # txts_train[train_sample_sequence[i]] # txts_train[i]\n",
    "\n",
    "# f.close()\n",
    "\n",
    "# f = open('svmlib_dataset/valence_test_sample_wise_pca', \"w+\")\n",
    "# for i in range(1671, 2088): # len(train_sample_sequence) len(txts_train)\n",
    "#     f.write(txts_train[i]) # txts_train[train_sample_sequence[i]] # txts_train[i]\n",
    "\n",
    "# f.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[ 856 1568 1012 ...  992  396 1029]\n",
      "504\n",
      "468\n",
      "720\n"
     ]
    }
   ],
   "source": [
    "# k means svm\n",
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_Order_Movie.mat')\n",
    "order = mat['PermutationList']\n",
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_SelfReports.mat')\n",
    "rating = mat['Ratings']\n",
    "\n",
    "features['ECG'] = np.nan_to_num(features['ECG'])\n",
    "features['GSR'] = np.nan_to_num(features['GSR'])\n",
    "features['EEG'] = np.nan_to_num(features['EEG'])\n",
    "features['ResNet'] = np.nan_to_num(features['ResNet'])\n",
    "\n",
    "clusters = [2, 0, 1, 2, 2, 2, 2, 2, 1, 1, 1, 0, 0, 0, 2, 2, 1, 2, 1, 1, 0, 1, 0, 2, 1, 0, 0, 1, 2, 2, 2, 2, 2, 2, 0, 0, 0,\n",
    " 1, 2, 2, 0, 2, 1, 2, 0, 0, 1]\n",
    "\n",
    "f = open('svmlib_dataset/train_subjects_id.csv')\n",
    "rows = csv.reader(f)\n",
    "for row in rows:\n",
    "    rows = row\n",
    "    break\n",
    "train_subjects_list = np.array(rows).astype(int)\n",
    "f.close()\n",
    "train_subjects_list -= 1\n",
    "\n",
    "test_subjects_list = np.array([2, 21, 22, 24, 25, 27, 28, 31, 32, 42, 49])\n",
    "\n",
    "f = open('svmlib_dataset/features_permutation.csv')\n",
    "rows = csv.reader(f)\n",
    "for row in rows:\n",
    "    rows = row\n",
    "    break\n",
    "feature_sequence = np.array(rows).astype(int)\n",
    "f.close()\n",
    "\n",
    "f = open('svmlib_dataset/samples_permutation_clip_wise.csv')\n",
    "rows = csv.reader(f)\n",
    "for row in rows:\n",
    "    rows = row\n",
    "    break\n",
    "train_sample_sequence = np.array(rows).astype(int)\n",
    "f.close()\n",
    "print(train_sample_sequence)\n",
    "\n",
    "# f = open('svmlib_dataset/train_subjects_id.csv')\n",
    "# rows = csv.reader(f)\n",
    "# for row in rows:\n",
    "#     rows = row\n",
    "#     break\n",
    "# train_subjects_list = np.array(rows).astype(int)\n",
    "# f.close()\n",
    "\n",
    "f = open('svmlib_dataset/selected_feature.csv')\n",
    "rows = csv.reader(f)\n",
    "for row in rows:\n",
    "    rows = row\n",
    "    break\n",
    "selected_feature = np.array(rows).astype(int)\n",
    "selected_feature -= 1\n",
    "f.close()\n",
    "\n",
    "'''\n",
    "0 ~ 13: ECG\n",
    "14 ~ 44: GSR\n",
    "45 ~ 132: EEG\n",
    "'''\n",
    "\n",
    "# dataset for arousal classifier feature selection\n",
    "subjects = 47\n",
    "instances = 2088\n",
    "clips = 36\n",
    "ones = 0\n",
    "zeros = 0\n",
    "\n",
    "\n",
    "for k in range(3):\n",
    "    txts_train = []\n",
    "    index_1 = 0\n",
    "    for i in train_subjects_list:\n",
    "        #print(index_1, clusters[index_1], k)\n",
    "        if clusters[index_1] != k:\n",
    "            index_1 += 1\n",
    "            continue\n",
    "        for clip_index in range(36):\n",
    "            #if(clip_index in train_clips_list): continue\n",
    "            feature_index = 0\n",
    "            txt = ''\n",
    "            txt += get_valence_class(rating[1][i][order[i][clip_index]-1]) # label\n",
    "            true_feature_index = 0\n",
    "\n",
    "            if get_arousal_class(rating[1][i][order[i][clip_index]-1])=='+1':\n",
    "                ones += 1\n",
    "            else:\n",
    "                zeros += 1\n",
    "            feature_seq = []\n",
    "\n",
    "            # ECG\n",
    "            for j in range(14):\n",
    "                feature_seq.append(features['ECG'][0][i][clip_index][j])\n",
    "\n",
    "            # GSR\n",
    "            for j in range(31):\n",
    "                feature_seq.append(features['GSR'][0][i][clip_index][j])\n",
    "\n",
    "            # EEG\n",
    "            for j in range(88):\n",
    "                feature_seq.append(features['EEG'][0][i][clip_index][j])\n",
    "\n",
    "\n",
    "            f = [0.0]*133\n",
    "            for j in range(len(feature_sequence)):\n",
    "                f[j] = feature_seq[feature_sequence[j]]\n",
    "\n",
    "            for j in range(133): # selected_feature\n",
    "                txt += ' '\n",
    "                txt += str(feature_index+1) + ':' + str(f[j])\n",
    "                feature_index += 1    \n",
    "\n",
    "            txt += '\\n'\n",
    "\n",
    "            txts_train.append(txt)\n",
    "            \n",
    "        index_1 += 1\n",
    "    print(len(txts_train))\n",
    "    random.shuffle(txts_train)\n",
    "\n",
    "    f = open('svmlib_dataset/valence_train_' + str(k+1) + '_cluster', \"w+\")\n",
    "    for i in range(len(txts_train)): # len(train_sample_sequence) len(txts_train)\n",
    "        f.write(txts_train[i]) # txts_train[train_sample_sequence[i]] # txts_train[i]\n",
    "\n",
    "    f.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[ 856 1568 1012 ...  992  396 1029]\n"
     ]
    }
   ],
   "source": [
    "# single subject\n",
    "# k means svm\n",
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_Order_Movie.mat')\n",
    "order = mat['PermutationList']\n",
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_SelfReports.mat')\n",
    "rating = mat['Ratings']\n",
    "\n",
    "features['ECG'] = np.nan_to_num(features['ECG'])\n",
    "features['GSR'] = np.nan_to_num(features['GSR'])\n",
    "features['EEG'] = np.nan_to_num(features['EEG'])\n",
    "features['ResNet'] = np.nan_to_num(features['ResNet'])\n",
    "\n",
    "clusters = [2, 0, 1, 2, 2, 2, 2, 2, 1, 1, 1, 0, 0, 0, 2, 2, 1, 2, 1, 1, 0, 1, 0, 2, 1, 0, 0, 1, 2, 2, 2, 2, 2, 2, 0, 0, 0,\n",
    " 1, 2, 2, 0, 2, 1, 2, 0, 0, 1]\n",
    "\n",
    "f = open('svmlib_dataset/train_subjects_id.csv')\n",
    "rows = csv.reader(f)\n",
    "for row in rows:\n",
    "    rows = row\n",
    "    break\n",
    "train_subjects_list = np.array(rows).astype(int)\n",
    "f.close()\n",
    "train_subjects_list -= 1\n",
    "\n",
    "test_subjects_list = np.array([2, 21, 22, 24, 25, 27, 28, 31, 32, 42, 49])\n",
    "\n",
    "f = open('svmlib_dataset/features_permutation.csv')\n",
    "rows = csv.reader(f)\n",
    "for row in rows:\n",
    "    rows = row\n",
    "    break\n",
    "feature_sequence = np.array(rows).astype(int)\n",
    "f.close()\n",
    "\n",
    "f = open('svmlib_dataset/samples_permutation_clip_wise.csv')\n",
    "rows = csv.reader(f)\n",
    "for row in rows:\n",
    "    rows = row\n",
    "    break\n",
    "train_sample_sequence = np.array(rows).astype(int)\n",
    "f.close()\n",
    "print(train_sample_sequence)\n",
    "\n",
    "# f = open('svmlib_dataset/train_subjects_id.csv')\n",
    "# rows = csv.reader(f)\n",
    "# for row in rows:\n",
    "#     rows = row\n",
    "#     break\n",
    "# train_subjects_list = np.array(rows).astype(int)\n",
    "# f.close()\n",
    "\n",
    "f = open('svmlib_dataset/selected_feature.csv')\n",
    "rows = csv.reader(f)\n",
    "for row in rows:\n",
    "    rows = row\n",
    "    break\n",
    "selected_feature = np.array(rows).astype(int)\n",
    "selected_feature -= 1\n",
    "f.close()\n",
    "\n",
    "'''\n",
    "0 ~ 13: ECG\n",
    "14 ~ 44: GSR\n",
    "45 ~ 132: EEG\n",
    "'''\n",
    "\n",
    "# dataset for arousal classifier feature selection\n",
    "subjects = 47\n",
    "instances = 2088\n",
    "clips = 36\n",
    "ones = 0\n",
    "zeros = 0\n",
    "\n",
    "\n",
    "for k in range(1):\n",
    "    \n",
    "    \n",
    "    for i in test_subjects_list:\n",
    "        txts_test = []\n",
    "        index_1 = 0\n",
    "        for clip_index in range(36):\n",
    "            feature_index = 0\n",
    "            txt = ''\n",
    "            txt += get_valence_class(rating[1][i][order[i][clip_index]-1]) # label\n",
    "            true_feature_index = 0\n",
    "\n",
    "            if get_arousal_class(rating[1][i][order[i][clip_index]-1])=='+1':\n",
    "                ones += 1\n",
    "            else:\n",
    "                zeros += 1\n",
    "            feature_seq = []\n",
    "\n",
    "            # ECG\n",
    "            for j in range(14):\n",
    "                feature_seq.append(features['ECG'][0][i][clip_index][j])\n",
    "\n",
    "            # GSR\n",
    "            for j in range(31):\n",
    "                feature_seq.append(features['GSR'][0][i][clip_index][j])\n",
    "\n",
    "            # EEG\n",
    "            for j in range(88):\n",
    "                feature_seq.append(features['EEG'][0][i][clip_index][j])\n",
    "\n",
    "\n",
    "            f = [0.0]*133\n",
    "            for j in range(len(feature_sequence)):\n",
    "                f[j] = feature_seq[feature_sequence[j]]\n",
    "\n",
    "            for j in range(133): # selected_feature\n",
    "                txt += ' '\n",
    "                txt += str(feature_index+1) + ':' + str(f[j])\n",
    "                feature_index += 1    \n",
    "\n",
    "            txt += '\\n'\n",
    "\n",
    "            txts_test.append(txt)\n",
    "            \n",
    "        index_1 += 1\n",
    "        random.shuffle(txts_test)\n",
    "\n",
    "        f = open('svmlib_dataset/subjects/valence_test_' + str(i) + '_subject', \"w+\")\n",
    "        for i in range(len(txts_test)): # len(train_sample_sequence) len(txts_train)\n",
    "            f.write(txts_test[i]) # txts_train[train_sample_sequence[i]] # txts_train[i]\n",
    "\n",
    "        f.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 89,
   "metadata": {},
   "outputs": [],
   "source": [
    "f = open('svmlib_dataset/arousal_test_sample_wise', \"w+\")\n",
    "for i in range(1671, 2087):\n",
    "    f.write(txts_train[i])\n",
    "f.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "'''\n",
    "to do lists:\n",
    "\n",
    "* subject-wise split >\n",
    "* clip-wise split >\n",
    "* sample-wise split\n",
    "\n",
    "* selected no-resnet >\n",
    "* selected resnet-only\n",
    "* no-personality\n",
    "* with personality\n",
    "\n",
    "* sample a subject of samples for IBCGA (for training and evaluating)\n",
    "\n",
    "* get rid of neural rating\n",
    "* MMAE\n",
    "\n",
    "* k-means with personality\n",
    "\n",
    "'''"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Collecting csv\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  Could not find a version that satisfies the requirement csv (from versions: )\n",
      "No matching distribution found for csv\n",
      "You are using pip version 10.0.1, however version 20.2b1 is available.\n",
      "You should consider upgrading via the 'python -m pip install --upgrade pip' command.\n"
     ]
    }
   ],
   "source": [
    "# cmd\n",
    "! pip install csv"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "4.0"
      ]
     },
     "execution_count": 75,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "rating[0][0][order[0][clip_index]-1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "416\n",
      "416\n",
      "0.5563538689531068\n"
     ]
    }
   ],
   "source": [
    "f = open('./svmlib_dataset/arousal_test_sample_wise')\n",
    "data = f.read()\n",
    "f.close()\n",
    "\n",
    "predicts = []\n",
    "f = open('./libsvm-3.23/windows/tmp')\n",
    "n = f.read()\n",
    "n = n.split('\\n')[1:]\n",
    "for x in n:\n",
    "    if x == '': break\n",
    "    xx = x.split()\n",
    "    predicts.append(float(xx[1]))\n",
    "print(len(predicts))\n",
    "data = data.split('\\n')\n",
    "positive = 0\n",
    "right = 0\n",
    "trues = []\n",
    "for d in data:\n",
    "    for s in d.split():\n",
    "        if s[0] == '+':\n",
    "            trues.append(1)\n",
    "        else:\n",
    "            trues.append(-1)\n",
    "        break\n",
    "print(len(trues))\n",
    "fpr, tpr, _ = roc_curve(trues, predicts)\n",
    "print(auc(fpr, tpr))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.scatter([0, 1], [0, 1])\n",
    "plt.scatter([0, 1], [0, 1])\n",
    "plt.scatter([0, 1], [0, 1])\n",
    "plt.scatter([0, 1], [0, 1])\n",
    "plt.savefig('gg')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 123,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "k\n"
     ]
    }
   ],
   "source": [
    "def _aa():\n",
    "    if _aa is not None:\n",
    "        print('k')\n",
    "    elif _aa is not None:\n",
    "        print('k')\n",
    "    else:\n",
    "        print('k')\n",
    "_aa()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[2 0 1 2 2 2 2 2 1 1 1 0 0 0 2 2 1 2 1 1 0 1 0 2 1 0 0 1 2 2 2 2 2 2 0 0 0\n",
      " 1 2 2 0 2 1 2 0 0 1]\n",
      "[0.2413793103448276, 0.22413793103448276, 0.3448275862068966]\n",
      "[[4.5, 5.3, 3.7], [5.1, 6.0, 4.9], [2.8, 3.5, 2.5], [4.9, 4.7, 2.5], [5.5, 5.0, 4.3], [5.9, 3.9, 3.0], [5.7, 4.8, 4.5], [5.9, 4.9, 5.0], [3.0, 4.3, 3.7], [2.6, 5.7, 3.3], [4.1, 3.9, 3.8], [4.5, 4.5, 4.8], [3.9, 5.7, 4.1], [4.2, 5.6, 4.9], [5.1, 5.1, 3.5], [5.8, 6.9, 4.8], [3.0, 6.1, 2.7], [4.9, 5.9, 4.0], [3.1, 3.4, 4.1], [3.5, 5.4, 3.9], [4.6, 5.1, 4.6], [3.1, 4.6, 4.4], [3.2, 6.1, 5.7], [4.8, 5.4, 3.8], [3.0, 5.9, 3.8], [5.2, 6.4, 5.4], [3.9, 5.4, 4.3], [2.6, 4.2, 2.8], [5.6, 6.5, 2.6], [6.2, 4.8, 4.7], [5.1, 4.8, 3.3], [5.6, 6.0, 4.4], [5.5, 4.5, 3.8], [5.4, 5.6, 4.6], [3.9, 6.3, 4.4], [3.3, 5.8, 5.7], [3.6, 5.5, 5.0], [2.9, 4.0, 4.1], [5.4, 5.2, 5.1], [5.3, 5.2, 3.4], [2.9, 5.6, 5.7], [6.1, 5.3, 2.0], [4.1, 4.3, 2.7], [5.6, 5.7, 4.0], [3.2, 5.2, 5.5], [4.6, 5.4, 5.2], [4.0, 5.0, 4.0]]\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# personality k-means clustering\n",
    "train_subjects_num = 47\n",
    "\n",
    "clusters = 3\n",
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_Personality.mat')\n",
    "mat = mat['Personality']\n",
    "personality = np.array([[0.0]*3]*train_subjects_num)\n",
    "selected_personality = [0, 2, 3]\n",
    "\n",
    "f = open('svmlib_dataset/train_subjects_id.csv')\n",
    "rows = csv.reader(f)\n",
    "for row in rows:\n",
    "    rows = row\n",
    "    break\n",
    "train_subjects_list = np.array(rows).astype(int)\n",
    "train_subjects_list -= 1\n",
    "f.close()\n",
    "\n",
    "index = 0\n",
    "for i in train_subjects_list:\n",
    "    for j in range(len(selected_personality)):\n",
    "        personality[index][j] = mat[i][selected_personality[j]]\n",
    "    index += 1\n",
    "\n",
    "kmeans = KMeans(n_clusters=clusters, random_state=0).fit(personality)\n",
    "print(np.array(kmeans.labels_))\n",
    "distributions = []\n",
    "for i in range(clusters):\n",
    "    distributions.append(sum([1 if x == i else 0 for x in kmeans.labels_])/mat.shape[0])\n",
    "print(distributions)\n",
    "print(personality.tolist())\n",
    "ax = plt.axes(projection='3d')\n",
    "ax.scatter(xs=personality[:, 0], ys=personality[:, 1], zs=personality[:, 2], c=kmeans.labels_)\n",
    "ax.set_xlabel(\"Extraversio\")\n",
    "ax.set_ylabel(\"Conscientiousness\")\n",
    "ax.set_zlabel(\"Emotional Stability\")\n",
    "plt.savefig('personality_train_k_means')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# mmae\n",
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_Order_Movie.mat')\n",
    "order = mat['PermutationList']\n",
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_SelfReports.mat')\n",
    "rating = mat['Ratings']\n",
    "\n",
    "features['ECG'] = np.nan_to_num(features['ECG'])\n",
    "features['GSR'] = np.nan_to_num(features['GSR'])\n",
    "features['EEG'] = np.nan_to_num(features['EEG'])\n",
    "features['ResNet'] = np.nan_to_num(features['ResNet'])\n",
    "\n",
    "ecgs = []\n",
    "gsrs = []\n",
    "eegs = []\n",
    "\n",
    "# dataset for arousal classifier feature selection\n",
    "subjects = 58\n",
    "instances = 2088\n",
    "clips = 36\n",
    "ones = 0\n",
    "zeros = 0\n",
    "\n",
    "\n",
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_Personality.mat')\n",
    "mat = mat['Personality']\n",
    "personality = np.array([[0.0]*3]*58)\n",
    "selected_personality = [0, 2, 3]\n",
    "\n",
    "for i in range(58):\n",
    "    for j in range(len(selected_personality)):\n",
    "        personality[i][j] = mat[i][selected_personality[j]]\n",
    "personality = np.array(personality)\n",
    "for i in range(3):\n",
    "    personality[:,i] = normalize(personality[:,i])\n",
    "    \n",
    "    \n",
    "# data quality\n",
    "f = open('./data_quality/raw_EEG.txt','r')\n",
    "raw_EEG_quality = 36 * [58 * [0]]\n",
    "data = f.read()\n",
    "clips = data.split('\\n')\n",
    "for i in range(36):\n",
    "    j = 0\n",
    "    for text in clips[i].split():\n",
    "        raw_EEG_quality[i][j] = int(text)\n",
    "        j += 1\n",
    "        \n",
    "f = open('./data_quality/ECG.txt','r')\n",
    "ECG_quality = 36 * [58 * [0]]\n",
    "data = f.read()\n",
    "clips = data.split('\\n')\n",
    "for i in range(36):\n",
    "    j = 0\n",
    "    for text in clips[i].split():\n",
    "        ECG_quality[i][j] = int(text)\n",
    "        j += 1\n",
    "        \n",
    "f = open('./data_quality/GSR.txt','r')\n",
    "GSR_quality = 36 * [58 * [0]]\n",
    "data = f.read()\n",
    "clips = data.split('\\n')\n",
    "for i in range(36):\n",
    "    j = 0\n",
    "    for text in clips[i].split():\n",
    "        GSR_quality[i][j] = int(text)\n",
    "        j += 1\n",
    "\n",
    "ys = []\n",
    "excluded_clips = [3, 4, 32] # 33 - 3 = 30\n",
    "for i in range(58): # 0~27 for train, 28~35 for test\n",
    "    txts_train = []\n",
    "    for clip_index in range(36):\n",
    "        # filter cleaned data\n",
    "        #if not(raw_EEG_quality[clip_index][i] <= 4 and ECG_quality[clip_index][i] <= 4 and GSR_quality[clip_index][i] <= 4): continue\n",
    "        \n",
    "        feature_index = 0\n",
    "        txt = ''\n",
    "        txt += str(int(get_arousal_class(rating[0][i][order[i][clip_index]-1]))) # label\n",
    "        ys.append([int(get_arousal_class(rating[0][i][order[i][clip_index]-1])), int(get_valence_class(rating[1][i][order[i][clip_index]-1]))])\n",
    "        true_feature_index = 0\n",
    "        \n",
    "        feature_seq = []\n",
    "        \n",
    "        feature_seq.append(str(int(get_valence_class(rating[1][i][order[i][clip_index]-1]))))\n",
    "        \n",
    "        feature_seq.append(i)\n",
    "        feature_seq.append(clip_index)\n",
    "        \n",
    "        if clip_index < 28:\n",
    "            feature_seq.append('Train')\n",
    "        else:\n",
    "            feature_seq.append('Test')\n",
    "            \n",
    "        feature_seq.append(not(raw_EEG_quality[clip_index][i] <= 4 and ECG_quality[clip_index][i] <= 4 and GSR_quality[clip_index][i] <= 4))\n",
    "        \n",
    "        \n",
    "        \n",
    "        tmp = []\n",
    "        # ECG\n",
    "        for j in range(14):\n",
    "            tmp.append(features['ECG'][0][i][clip_index][j]) # order[i].tolist().index(clip_index+1) // for same-clip arrangement\n",
    "        ecgs.append(tmp)\n",
    "        tmp = [tmp]\n",
    "        #tmptmp = lda_ecg.transform(tmp)\n",
    "        tmptmp = tmp\n",
    "        for ecg in tmptmp[0]:\n",
    "            feature_seq.append(ecg)\n",
    "        \n",
    "        \n",
    "        tmp = []\n",
    "        #GSR\n",
    "        for j in range(31):\n",
    "            tmp.append(features['GSR'][0][i][clip_index][j])\n",
    "        gsrs.append(tmp)\n",
    "        tmp = [tmp]\n",
    "        #tmptmp = lda_gsr.transform(tmp)\n",
    "        tmptmp = tmp\n",
    "        for gsr in tmptmp[0]:\n",
    "            feature_seq.append(gsr)\n",
    "            \n",
    "\n",
    "    \n",
    "        tmp = []\n",
    "        # EEG\n",
    "        for j in range(88):\n",
    "            tmp.append(features['EEG'][0][i][clip_index][j])\n",
    "        eegs.append(tmp)\n",
    "        tmp = [tmp]\n",
    "        #tmptmp = lda_eeg.transform(tmp)\n",
    "        tmptmp = tmp\n",
    "        for eeg in tmptmp[0]:\n",
    "            feature_seq.append(eeg)\n",
    "        \n",
    "        # personality\n",
    "#         for j in range(3):\n",
    "#             feature_seq.append(personality[i][j])\n",
    "        \n",
    "        for j in range(len(feature_seq)): # selected_feature\n",
    "            txt += ' '\n",
    "            txt += str(feature_index+1) + ':' + str(feature_seq[j])\n",
    "            feature_index += 1    \n",
    "            \n",
    "        txt += '\\n'\n",
    "    \n",
    "        txts_train.append(txt)\n",
    "        \n",
    "#random.shuffle(txts_train)\n",
    "        \n",
    "    f = open('svmlib_dataset/mmae'+'_'+str(i), \"w+\")\n",
    "    for i in range(len(txts_train)): # len(train_sample_sequence) len(txts_train)\n",
    "        f.write(txts_train[i]) # txts_train[train_sample_sequence[i]] # txts_train[i]\n",
    "\n",
    "    f.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 119,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\User\\AppData\\Local\\conda\\conda\\envs\\tensorflow\\lib\\site-packages\\sklearn\\discriminant_analysis.py:388: UserWarning: Variables are collinear.\n",
      "  warnings.warn(\"Variables are collinear.\")\n"
     ]
    }
   ],
   "source": [
    "# within subject data handler\n",
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_Order_Movie.mat')\n",
    "order = mat['PermutationList']\n",
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_SelfReports.mat')\n",
    "rating = mat['Ratings']\n",
    "\n",
    "arousal_means_train = [np.mean(rating[0,:,i]) for i in range(28)]\n",
    "valence_means = [np.mean(rating[1,:,i]) for i in range(28)]\n",
    "\n",
    "features['ECG'] = np.nan_to_num(features['ECG'])\n",
    "features['GSR'] = np.nan_to_num(features['GSR'])\n",
    "features['EEG'] = np.nan_to_num(features['EEG'])\n",
    "features['ResNet'] = np.nan_to_num(features['ResNet'])\n",
    "\n",
    "\n",
    "\n",
    "# dataset for arousal classifier feature selection\n",
    "subjects = 58\n",
    "instances = 2088\n",
    "clips = 36\n",
    "ones = 0\n",
    "zeros = 0\n",
    "txts_train = []\n",
    "\n",
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_Personality.mat')\n",
    "mat = mat['Personality']\n",
    "personality = np.array([[0.0]*3]*58)\n",
    "selected_personality = [0, 2, 3]\n",
    "\n",
    "for i in range(58):\n",
    "    for j in range(len(selected_personality)):\n",
    "        personality[i][j] = mat[i][selected_personality[j]]\n",
    "personality = np.array(personality)\n",
    "for i in range(3):\n",
    "    personality[:,i] = normalize(personality[:,i])\n",
    "    \n",
    "    \n",
    "# data quality\n",
    "f = open('./data_quality/raw_EEG.txt','r')\n",
    "raw_EEG_quality = 36 * [58 * [0]]\n",
    "data = f.read()\n",
    "clips = data.split('\\n')\n",
    "for i in range(36):\n",
    "    j = 0\n",
    "    for text in clips[i].split():\n",
    "        raw_EEG_quality[i][j] = int(text)\n",
    "        j += 1\n",
    "        \n",
    "f = open('./data_quality/ECG.txt','r')\n",
    "ECG_quality = 36 * [58 * [0]]\n",
    "data = f.read()\n",
    "clips = data.split('\\n')\n",
    "for i in range(36):\n",
    "    j = 0\n",
    "    for text in clips[i].split():\n",
    "        ECG_quality[i][j] = int(text)\n",
    "        j += 1\n",
    "        \n",
    "f = open('./data_quality/GSR.txt','r')\n",
    "GSR_quality = 36 * [58 * [0]]\n",
    "data = f.read()\n",
    "clips = data.split('\\n')\n",
    "for i in range(36):\n",
    "    j = 0\n",
    "    for text in clips[i].split():\n",
    "        GSR_quality[i][j] = int(text)\n",
    "        j += 1\n",
    "ecgs = []\n",
    "gsrs = []\n",
    "eegs = []\n",
    "ys = []    \n",
    "for i in range(58): \n",
    "    \n",
    "    for clip_index in range(28): # 0~27 for train, 28~35 for test\n",
    "        feature_index = 0\n",
    "        # int(get_arousal_class(rating[0][i][order[i][clip_index]-1])), int(get_valence_class(rating[1][i][order[i][clip_index]-1]))\n",
    "        ys.append([int(get_arousal_class(rating[0][i][order[i][clip_index]-1], arousals_means[order[i][clip_index]-1])), int(get_valence_class(rating[1][i][order[i][clip_index]-1]))])\n",
    "\n",
    "        feature_seq = []\n",
    "\n",
    "        tmp = []\n",
    "        # ECG\n",
    "        for j in range(14):\n",
    "            tmp.append(features['ECG'][0][i][clip_index][j]) # order[i].tolist().index(clip_index+1) // for same-clip arrangement\n",
    "        ecgs.append(tmp)\n",
    "\n",
    "        tmp = []\n",
    "        #GSR\n",
    "        for j in range(31):\n",
    "            tmp.append(features['GSR'][0][i][clip_index][j])\n",
    "        gsrs.append(tmp)\n",
    "\n",
    "        tmp = []\n",
    "        # EEG\n",
    "        for j in range(88):\n",
    "            tmp.append(features['EEG'][0][i][clip_index][j])\n",
    "        eegs.append(tmp)\n",
    "\n",
    "ys = np.array(ys)\n",
    "yss = []\n",
    "for y in ys:\n",
    "    if y[0] == -1 and y[1] == -1:\n",
    "        yss.append(0)\n",
    "    elif y[0] == -1 and y[1] == 1:\n",
    "        yss.append(1)\n",
    "    elif y[0] == 1 and y[1] == -1:\n",
    "        yss.append(2)\n",
    "    else:\n",
    "        yss.append(3)\n",
    "\n",
    "lda_eeg = LinearDiscriminantAnalysis(n_components=2) # n_components=2\n",
    "lda_eeg.fit(np.array(eegs), yss)\n",
    "#print('ratio', lda_eeg.explained_variance_ratio_)\n",
    "\n",
    "lda_ecg = LinearDiscriminantAnalysis(n_components=1)\n",
    "lda_ecg.fit(np.array(ecgs), yss)\n",
    "#print('ratio', lda_ecg.explained_variance_ratio_)\n",
    "\n",
    "lda_gsr = LinearDiscriminantAnalysis(n_components=2)\n",
    "lda_gsr.fit(np.array(gsrs), yss) \n",
    "\n",
    "        \n",
    "        \n",
    "\n",
    "for i in range(58): \n",
    "    \n",
    "    #print('ratio', lda_gsr.explained_variance_ratio_)\n",
    "\n",
    "#     for clip_index in range(28): # 0~27 for train, 28~35 for test\n",
    "        \n",
    "#         feature_index = 0\n",
    "#         ys.append([int(get_arousal_class(rating[0][i][order[i][clip_index]-1])), int(get_valence_class(rating[1][i][order[i][clip_index]-1]))])\n",
    "        \n",
    "#         feature_seq = []\n",
    "\n",
    "#         tmp = []\n",
    "#         # ECG\n",
    "#         for j in range(14):\n",
    "#             tmp.append(features['ECG'][0][i][clip_index][j]) # order[i].tolist().index(clip_index+1) // for same-clip arrangement\n",
    "#         ecgs.append(tmp)\n",
    "        \n",
    "#         tmp = []\n",
    "#         #GSR\n",
    "#         for j in range(31):\n",
    "#             tmp.append(features['GSR'][0][i][clip_index][j])\n",
    "#         gsrs.append(tmp)\n",
    "    \n",
    "#         tmp = []\n",
    "#         # EEG\n",
    "#         for j in range(88):\n",
    "#             tmp.append(features['EEG'][0][i][clip_index][j])\n",
    "#         eegs.append(tmp)\n",
    "        \n",
    "#     ys = np.array(ys)\n",
    "#     yss = []\n",
    "#     for y in ys:\n",
    "#         if y[0] == -1 and y[1] == -1:\n",
    "#             yss.append(0)\n",
    "#         elif y[0] == -1 and y[1] == 1:\n",
    "#             yss.append(1)\n",
    "#         elif y[0] == 1 and y[1] == -1:\n",
    "#             yss.append(2)\n",
    "#         else:\n",
    "#             yss.append(3)\n",
    "\n",
    "#     lda_eeg = LinearDiscriminantAnalysis(n_components=2)\n",
    "#     lda_eeg.fit(np.array(eegs), yss)\n",
    "\n",
    "#     lda_ecg = LinearDiscriminantAnalysis(n_components=2)\n",
    "#     lda_ecg.fit(np.array(ecgs), yss)\n",
    "\n",
    "#     lda_gsr = LinearDiscriminantAnalysis(n_components=2)\n",
    "#     lda_gsr.fit(np.array(gsrs), yss)\n",
    "    \n",
    "    txts_train = []\n",
    "    for clip_index in range(28): # 0~27 for train, 28~35 for test\n",
    "        \n",
    "        feature_index = 0\n",
    "        txt = ''\n",
    "        txt += get_arousal_class(rating[0][i][order[i][clip_index]-1]) # label\n",
    "        true_feature_index = 0\n",
    "        \n",
    "        feature_seq = []\n",
    "\n",
    "        tmp = []\n",
    "        # ECG\n",
    "        for j in range(14):\n",
    "            tmp.append(features['ECG'][0][i][clip_index][j]) # order[i].tolist().index(clip_index+1) // for same-clip arrangement\n",
    "        tmp = [tmp]\n",
    "        #tmptmp = lda_ecg.transform(tmp)\n",
    "        tmptmp = tmp\n",
    "        for ecg in tmptmp[0]:\n",
    "            feature_seq.append(ecg)\n",
    "        \n",
    "        \n",
    "        tmp = []\n",
    "        #GSR\n",
    "        for j in range(31):\n",
    "            tmp.append(features['GSR'][0][i][clip_index][j])\n",
    "        tmp = [tmp]\n",
    "        #tmptmp = lda_gsr.transform(tmp)\n",
    "        tmptmp = tmp\n",
    "        for gsr in tmptmp[0]:\n",
    "            feature_seq.append(gsr)\n",
    "            \n",
    "        tmp = []\n",
    "        # EEG\n",
    "        for j in range(88):\n",
    "            tmp.append(features['EEG'][0][i][clip_index][j])\n",
    "        tmp = [tmp]\n",
    "        #tmptmp = lda_eeg.transform(tmp)\n",
    "        tmptmp = tmp\n",
    "        for eeg in tmptmp[0]:\n",
    "            feature_seq.append(eeg)\n",
    "        \n",
    "        for j in range(len(feature_seq)): # selected_feature\n",
    "            txt += ' '\n",
    "            txt += str(feature_index+1) + ':' + str(feature_seq[j])\n",
    "            feature_index += 1    \n",
    "            \n",
    "        txt += '\\n'\n",
    "    \n",
    "        txts_train.append(txt)\n",
    "        \n",
    "    f = open('svmlib_dataset/subjects/' + str(i+1) + '_arousal_raw_train', \"w+\")\n",
    "    for j in range(len(txts_train)): # len(train_sample_sequence) len(txts_train)\n",
    "        f.write(txts_train[j]) # txts_train[train_sample_sequence[i]] # txts_train[i]\n",
    "\n",
    "    f.close()\n",
    "    txts_train = []\n",
    "    \n",
    "#     ecgs = []\n",
    "#     gsrs = []\n",
    "#     eegs = []\n",
    "#     ys = []\n",
    "#     for clip_index in range(28, 36): # 0~27 for train, 28~35 for test\n",
    "        \n",
    "#         feature_index = 0\n",
    "#         ys.append([int(get_arousal_class(rating[0][i][order[i][clip_index]-1])), int(get_valence_class(rating[1][i][order[i][clip_index]-1]))])\n",
    "        \n",
    "#         feature_seq = []\n",
    "\n",
    "#         tmp = []\n",
    "#         # ECG\n",
    "#         for j in range(14):\n",
    "#             tmp.append(features['ECG'][0][i][clip_index][j]) # order[i].tolist().index(clip_index+1) // for same-clip arrangement\n",
    "#         ecgs.append(tmp)\n",
    "        \n",
    "#         tmp = []\n",
    "#         #GSR\n",
    "#         for j in range(31):\n",
    "#             tmp.append(features['GSR'][0][i][clip_index][j])\n",
    "#         gsrs.append(tmp)\n",
    "    \n",
    "#         tmp = []\n",
    "#         # EEG\n",
    "#         for j in range(88):\n",
    "#             tmp.append(features['EEG'][0][i][clip_index][j])\n",
    "#         eegs.append(tmp)\n",
    "        \n",
    "#     ys = np.array(ys)\n",
    "#     yss = []\n",
    "#     for y in ys:\n",
    "#         if y[0] == -1 and y[1] == -1:\n",
    "#             yss.append(0)\n",
    "#         elif y[0] == -1 and y[1] == 1:\n",
    "#             yss.append(1)\n",
    "#         elif y[0] == 1 and y[1] == -1:\n",
    "#             yss.append(2)\n",
    "#         else:\n",
    "#             yss.append(3)\n",
    "\n",
    "#     lda_eeg = LinearDiscriminantAnalysis(n_components=2)\n",
    "#     lda_eeg.fit(np.array(eegs), yss)\n",
    "\n",
    "#     lda_ecg = LinearDiscriminantAnalysis(n_components=2)\n",
    "#     lda_ecg.fit(np.array(ecgs), yss)\n",
    "\n",
    "#     lda_gsr = LinearDiscriminantAnalysis(n_components=2)\n",
    "#     lda_gsr.fit(np.array(gsrs), yss)\n",
    "        \n",
    "    for clip_index in range(28, 36): # 0~27 for train, 28~35 for test\n",
    "        \n",
    "        feature_index = 0\n",
    "        txt = ''\n",
    "        txt += get_arousal_class(rating[0][i][order[i][clip_index]-1]) # label\n",
    "        true_feature_index = 0\n",
    "        \n",
    "        feature_seq = []\n",
    "\n",
    "        tmp = []\n",
    "        # ECG\n",
    "        for j in range(14):\n",
    "            tmp.append(features['ECG'][0][i][clip_index][j]) # order[i].tolist().index(clip_index+1) // for same-clip arrangement\n",
    "        tmp = [tmp]\n",
    "        #tmptmp = lda_ecg.transform(tmp)\n",
    "        tmptmp = tmp\n",
    "        for ecg in tmptmp[0]:\n",
    "            feature_seq.append(ecg)\n",
    "        \n",
    "        \n",
    "        tmp = []\n",
    "        #GSR\n",
    "        for j in range(31):\n",
    "            tmp.append(features['GSR'][0][i][clip_index][j])\n",
    "        tmp = [tmp]\n",
    "        #tmptmp = lda_gsr.transform(tmp)\n",
    "        tmptmp = tmp\n",
    "        for gsr in tmptmp[0]:\n",
    "            feature_seq.append(gsr)\n",
    "            \n",
    "\n",
    "    \n",
    "        tmp = []\n",
    "        # EEG\n",
    "        for j in range(88):\n",
    "            tmp.append(features['EEG'][0][i][clip_index][j])\n",
    "        tmp = [tmp]\n",
    "        #tmptmp = lda_eeg.transform(tmp)\n",
    "        tmptmp = tmp\n",
    "        for eeg in tmptmp[0]:\n",
    "            feature_seq.append(eeg)\n",
    "        \n",
    "        \n",
    "        for j in range(len(feature_seq)): # selected_feature\n",
    "            txt += ' '\n",
    "            txt += str(feature_index+1) + ':' + str(feature_seq[j])\n",
    "            feature_index += 1    \n",
    "            \n",
    "        txt += '\\n'\n",
    "    \n",
    "        txts_train.append(txt)\n",
    "    \n",
    "    f = open('svmlib_dataset/subjects/' + str(i+1) + '_arousal_raw_test', \"w+\")\n",
    "    for j in range(len(txts_train)): # len(train_sample_sequence) len(txts_train)\n",
    "        f.write(txts_train[j]) # txts_train[train_sample_sequence[i]] # txts_train[i]\n",
    "\n",
    "    f.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 198,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.43478260869565216 0.5641304347826087\n",
      "0.5454545454545454 0.6737373737373736\n",
      "0.5660377358490566 0.6754032258064516\n",
      "0.4081632653061225 0.5463157894736842\n",
      "0.5084745762711865 0.6206896551724137\n",
      "0.5384615384615384 0.7112244897959183\n",
      "0.7647058823529412 0.7981744421906694\n",
      "0.49056603773584906 0.6421370967741935\n",
      "0.5 0.6732558139534883\n",
      "0.5283018867924529 0.5922920892494928\n",
      "0.46153846153846156 0.6112244897959183\n",
      "0.4999999999999999 0.5547368421052631\n",
      "0.5090909090909091 0.7046370967741935\n",
      "0.5084745762711865 0.579107505070994\n",
      "0.4727272727272727 0.6191919191919192\n",
      "0.5217391304347826 0.4794238683127572\n",
      "0.4528301886792452 0.5705263157894737\n",
      "0.2916666666666667 0.4722222222222222\n",
      "0.6557377049180327 0.6838383838383839\n",
      "0.35999999999999993 0.4508547008547008\n",
      "0.5161290322580646 0.5689655172413794\n",
      "0.5185185185185185 0.6409736308316429\n",
      "0.6181818181818183 0.7262626262626263\n",
      "0.46808510638297873 0.6189473684210527\n",
      "0.5000000000000001 0.5282258064516129\n",
      "0.6071428571428571 0.6723790322580645\n",
      "0.5098039215686274 0.5779914529914529\n",
      "0.5357142857142857 0.5466531440162272\n",
      "0.4150943396226415 0.6040816326530613\n",
      "0.5542168674698795 0.6089743589743589\n",
      "[0.43478260869565216, 0.5454545454545454, 0.5660377358490566, 0.4081632653061225, 0.5084745762711865, 0.5384615384615384, 0.7647058823529412, 0.49056603773584906, 0.5, 0.5283018867924529, 0.46153846153846156, 0.4999999999999999, 0.5090909090909091, 0.5084745762711865, 0.4727272727272727, 0.5217391304347826, 0.4528301886792452, 0.2916666666666667, 0.6557377049180327, 0.35999999999999993, 0.5161290322580646, 0.5185185185185185, 0.6181818181818183, 0.46808510638297873, 0.5000000000000001, 0.6071428571428571, 0.5098039215686274, 0.5357142857142857, 0.4150943396226415, 0.5542168674698795] [0.5641304347826087, 0.6737373737373736, 0.6754032258064516, 0.5463157894736842, 0.6206896551724137, 0.7112244897959183, 0.7981744421906694, 0.6421370967741935, 0.6732558139534883, 0.5922920892494928, 0.6112244897959183, 0.5547368421052631, 0.7046370967741935, 0.579107505070994, 0.6191919191919192, 0.4794238683127572, 0.5705263157894737, 0.4722222222222222, 0.6838383838383839, 0.4508547008547008, 0.5689655172413794, 0.6409736308316429, 0.7262626262626263, 0.6189473684210527, 0.5282258064516129, 0.6723790322580645, 0.5779914529914529, 0.5466531440162272, 0.6040816326530613, 0.6089743589743589]\n"
     ]
    }
   ],
   "source": [
    "import importlib\n",
    "import utils.utils\n",
    "importlib.reload(utils.utils)\n",
    "resource_path = './libsvm-3.23/windows/'\n",
    "dataset_paths = ['./svmlib_dataset/valence_train_lda_ecg', './svmlib_dataset/valence_train_lda_gsr', './svmlib_dataset/valence_train_lda_eeg']\n",
    "names = ['valence_train_lda_ecg', 'valence_train_lda_gsr', 'valence_train_lda_eeg']\n",
    "f1_scores = [0, 0.055, 0.40]\n",
    "alphas = [0.3, 0.3, 0.4]\n",
    "optimized_alphas = [random.gauss(mu=1, sigma=0.2) for x in range(3)]\n",
    "k = 30\n",
    "cs = [512, 512, 512]\n",
    "gs = [0.0078125, 0.0078125, 0.0078125]\n",
    "f1s, aucs = utils.utils.fusion_predict(resource_path, dataset_paths, names, f1_scores, alphas, optimized_alphas, k, cs, gs)\n",
    "print(f1s, aucs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 199,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.5087213244701858\n",
      "0.6105526108331197\n"
     ]
    }
   ],
   "source": [
    "print(sum(f1s)/30)\n",
    "print(sum(aucs)/30)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 112,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[1, 1, -1, 1, 1, 1, 1, -1, 1, 1, -1, 1, 1, 1, -1, 1, -1, -1, -1, -1, -1, 1, -1, -1, 1, -1, -1, -1] [1, 1, -1, 1, 1, 1, 1, -1, 1, 1, -1, 1, 1, 1, -1, 1, -1, -1, -1, -1, -1, 1, -1, -1, 1, -1, -1, -1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, -1, 1, 1, 1, -1, -1, -1, -1, -1, -1, 1, -1, -1, 1] [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, -1, 1, 1, 1, -1, -1, -1, -1, -1, -1, 1, -1, -1, 1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[-1, 1, 1, -1, 1, -1, 1, -1, 1, 1, -1, -1, -1, 1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 1, 1, -1, -1, -1] [-1, 1, 1, -1, 1, -1, 1, -1, 1, 1, -1, -1, -1, 1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 1, 1, -1, -1, -1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[-1, -1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, -1, 1, 1, 1, 1, 1, -1, 1, 1, -1, 1, 1, -1, 1, 1, 1] [-1, -1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, -1, 1, 1, 1, 1, 1, -1, 1, 1, -1, 1, 1, -1, 1, 1, 1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[-1, 1, -1, 1, -1, -1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, 1, -1, 1, -1, -1, -1, 1, -1, 1, -1, -1, 1] [-1, 1, 1, 1, -1, -1, 1, 1, -1, 1, 1, -1, -1, 1, 1, -1, 1, -1, 1, -1, -1, -1, 1, -1, 1, -1, -1, 1]\n",
      "f1:  0.962962962962963\n",
      "acc:  0.9642857142857143\n",
      "\n",
      "[1, -1, -1, 1, 1, -1, 1, 1, 1, 1, 1, -1, 1, 1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, -1, 1, -1, 1] [1, -1, -1, 1, 1, -1, 1, 1, 1, 1, 1, -1, 1, 1, -1, 1, 1, 1, -1, 1, 1, -1, -1, 1, -1, 1, -1, 1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[1, -1, 1, -1, -1, -1, 1, -1, -1, 1, -1, 1, -1, 1, 1, -1, -1, 1, 1, 1, -1, 1, -1, 1, -1, -1, 1, 1] [1, -1, 1, 1, -1, -1, 1, -1, -1, 1, -1, 1, -1, 1, 1, -1, -1, 1, 1, 1, -1, 1, -1, 1, -1, -1, 1, 1]\n",
      "f1:  0.9655172413793104\n",
      "acc:  0.9642857142857143\n",
      "\n",
      "[1, 1, 1, 1, 1, 1, -1, -1, -1, -1, 1, -1, 1, -1, -1, 1, -1, 1, -1, 1, -1, -1, -1, 1, 1, -1, 1, 1] [1, 1, 1, 1, 1, 1, -1, -1, -1, -1, 1, -1, 1, -1, -1, 1, -1, 1, -1, 1, -1, -1, -1, 1, 1, -1, -1, 1]\n",
      "f1:  0.9655172413793104\n",
      "acc:  0.9642857142857143\n",
      "\n",
      "[1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, -1, 1, 1, 1, 1, 1, 1, -1, 1, -1, -1, 1, -1, -1] [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, -1, -1, 1, -1, -1]\n",
      "f1:  0.9565217391304348\n",
      "acc:  0.9285714285714286\n",
      "\n",
      "[-1, 1, -1, 1, -1, -1, 1, 1, -1, 1, 1, 1, 1, 1, -1, 1, 1, -1, 1, 1, -1, 1, -1, -1, -1, 1, 1, -1] [-1, 1, -1, 1, -1, -1, 1, 1, -1, 1, 1, 1, 1, 1, -1, 1, 1, -1, 1, 1, -1, 1, -1, -1, -1, 1, 1, -1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[-1, 1, 1, 1, 1, -1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, -1, -1, -1, 1, -1, -1, 1, -1, -1, 1, -1, -1] [-1, 1, 1, 1, 1, -1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, -1, -1, -1, 1, -1, -1, 1, -1, -1, 1, -1, -1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[1, 1, 1, -1, 1, 1, -1, -1, -1, 1, -1, 1, 1, -1, 1, 1, 1, -1, 1, 1, -1, 1, 1, 1, -1, -1, -1, 1] [1, 1, 1, -1, 1, 1, -1, -1, -1, 1, -1, 1, 1, -1, 1, 1, 1, -1, 1, 1, -1, 1, 1, 1, -1, -1, -1, 1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[-1, 1, 1, 1, 1, 1, -1, 1, 1, 1, 1, -1, -1, 1, -1, 1, 1, 1, 1, 1, -1, -1, 1, -1, 1, -1, -1, 1] [-1, 1, 1, 1, 1, 1, -1, 1, 1, 1, 1, -1, -1, 1, -1, 1, 1, 1, 1, 1, 1, -1, 1, -1, 1, -1, -1, 1]\n",
      "f1:  0.972972972972973\n",
      "acc:  0.9642857142857143\n",
      "\n",
      "[1, 1, -1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, -1, -1, 1, -1, 1, 1, 1, 1, 1, 1] [1, 1, -1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, -1, 1, -1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.9795918367346939\n",
      "acc:  0.9642857142857143\n",
      "\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[1;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[1;32m<ipython-input-112-899bcb3cbe8b>\u001b[0m in \u001b[0;36m<module>\u001b[1;34m()\u001b[0m\n\u001b[0;32m     12\u001b[0m \u001b[0mpreds\u001b[0m \u001b[1;33m=\u001b[0m \u001b[1;33m[\u001b[0m\u001b[1;33m]\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m     13\u001b[0m \u001b[1;32mfor\u001b[0m \u001b[0mi\u001b[0m \u001b[1;32min\u001b[0m \u001b[0mrange\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;36m58\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m---> 14\u001b[1;33m     \u001b[0mpreds\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mtrues\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mutils\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mutils\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mf1_score_evaluate\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mresource_path\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mdataset_path\u001b[0m\u001b[1;33m+\u001b[0m\u001b[0mstr\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mi\u001b[0m\u001b[1;33m+\u001b[0m\u001b[1;36m1\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m+\u001b[0m\u001b[1;34m'_arousal_lda_train'\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mstr\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mi\u001b[0m\u001b[1;33m+\u001b[0m\u001b[1;36m1\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m+\u001b[0m\u001b[1;34m'_arousal_lda_train'\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mk\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m     15\u001b[0m     \u001b[0mpreds\u001b[0m \u001b[1;33m=\u001b[0m \u001b[1;33m[\u001b[0m\u001b[1;33m-\u001b[0m\u001b[1;36m1\u001b[0m \u001b[1;32mif\u001b[0m \u001b[0mx\u001b[0m\u001b[1;33m<=\u001b[0m\u001b[1;36m0.5\u001b[0m \u001b[1;32melse\u001b[0m \u001b[1;36m1\u001b[0m \u001b[1;32mfor\u001b[0m \u001b[0mx\u001b[0m \u001b[1;32min\u001b[0m \u001b[0mpreds\u001b[0m\u001b[1;33m]\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m     16\u001b[0m     \u001b[0mf1\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mf1_score\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mtrues\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mpreds\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32mD:\\Courses\\2020-2 Biology Machine Learning\\ascertain-project\\utils\\utils.py\u001b[0m in \u001b[0;36mf1_score_evaluate\u001b[1;34m(resource_path, dataset_path, dataset_name, k, c, g, b)\u001b[0m\n\u001b[0;32m    354\u001b[0m         \u001b[0mcmdline\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mget_test_cmd\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mresource_path\u001b[0m \u001b[1;33m+\u001b[0m \u001b[1;34m'svm-predict.exe'\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mtest_name\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mtrain\u001b[0m \u001b[1;33m+\u001b[0m \u001b[1;34m'.model'\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mtest\u001b[0m \u001b[1;33m+\u001b[0m \u001b[1;34m'_output'\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    355\u001b[0m         \u001b[0mprocess\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0msubprocess\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mPopen\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mcmdline\u001b[0m\u001b[1;33m,\u001b[0m\u001b[0mshell\u001b[0m\u001b[1;33m=\u001b[0m\u001b[1;32mTrue\u001b[0m\u001b[1;33m,\u001b[0m\u001b[0mstdout\u001b[0m\u001b[1;33m=\u001b[0m\u001b[0msubprocess\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mPIPE\u001b[0m\u001b[1;33m,\u001b[0m\u001b[0mstderr\u001b[0m\u001b[1;33m=\u001b[0m\u001b[0msubprocess\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mPIPE\u001b[0m\u001b[1;33m,\u001b[0m\u001b[0mstdin\u001b[0m\u001b[1;33m=\u001b[0m\u001b[0msubprocess\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mPIPE\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m--> 356\u001b[1;33m         \u001b[0mprocess\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mwait\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m    357\u001b[0m         \u001b[0mresult\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mprocess\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mstdout\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m    358\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;32m~\\AppData\\Local\\conda\\conda\\envs\\tensorflow\\lib\\subprocess.py\u001b[0m in \u001b[0;36mwait\u001b[1;34m(self, timeout, endtime)\u001b[0m\n\u001b[0;32m   1006\u001b[0m             \u001b[1;32mif\u001b[0m \u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mreturncode\u001b[0m \u001b[1;32mis\u001b[0m \u001b[1;32mNone\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1007\u001b[0m                 result = _winapi.WaitForSingleObject(self._handle,\n\u001b[1;32m-> 1008\u001b[1;33m                                                     timeout_millis)\n\u001b[0m\u001b[0;32m   1009\u001b[0m                 \u001b[1;32mif\u001b[0m \u001b[0mresult\u001b[0m \u001b[1;33m==\u001b[0m \u001b[0m_winapi\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mWAIT_TIMEOUT\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m   1010\u001b[0m                     \u001b[1;32mraise\u001b[0m \u001b[0mTimeoutExpired\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mself\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0margs\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mtimeout\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "# train leave-one-out cross validation\n",
    "import importlib\n",
    "import utils.utils\n",
    "importlib.reload(utils.utils)\n",
    "resource_path = './libsvm-3.23/windows/'\n",
    "dataset_path = './svmlib_dataset/subjects/'\n",
    "dataset_name = ''\n",
    "k = 28\n",
    "f1_means = []\n",
    "acc_means = []\n",
    "trues = []\n",
    "preds = []\n",
    "for i in range(58):\n",
    "    preds, trues = utils.utils.f1_score_evaluate(resource_path, dataset_path+str(i+1)+'_arousal_lda_train', str(i+1)+'_arousal_lda_train', k)\n",
    "    preds = [-1 if x<=0.5 else 1 for x in preds]\n",
    "    f1 = f1_score(trues, preds)\n",
    "    print(trues, preds)\n",
    "    print('f1: ', f1)\n",
    "    f1_means.append(f1)\n",
    "    acc = [1 if preds[x]==trues[x] else 0 for x in range(28)]\n",
    "    print('acc: ', sum(acc)/28)\n",
    "    print()\n",
    "    acc_means.append(sum(acc)/28)\n",
    "print(sum(f1_means)/len(f1_means))\n",
    "print(sum(acc_means)/len(acc_means))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 120,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[1, -1, 1, -1, 1, -1, 1, 1] [1, 1, -1, 1, -1, -1, 1, 1]\n",
      "f1:  0.6\n",
      "acc:  0.5\n",
      "\n",
      "[-1, 1, -1, -1, -1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.6666666666666666\n",
      "acc:  0.5\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\User\\AppData\\Local\\conda\\conda\\envs\\tensorflow\\lib\\site-packages\\sklearn\\metrics\\classification.py:1135: UndefinedMetricWarning: F-score is ill-defined and being set to 0.0 due to no predicted samples.\n",
      "  'precision', 'predicted', average, warn_for)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[1, 1, -1, -1, -1, -1, 1, 1] [-1, -1, -1, -1, -1, -1, -1, -1]\n",
      "f1:  0.0\n",
      "acc:  0.5\n",
      "\n",
      "[1, 1, 1, 1, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[1, 1, 1, -1, -1, 1, 1, -1] [1, -1, -1, 1, 1, -1, -1, -1]\n",
      "f1:  0.25\n",
      "acc:  0.25\n",
      "\n",
      "[1, -1, -1, 1, 1, -1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.7692307692307693\n",
      "acc:  0.625\n",
      "\n",
      "[1, -1, -1, -1, -1, 1, 1, 1] [1, -1, 1, 1, 1, 1, 1, -1]\n",
      "f1:  0.6\n",
      "acc:  0.5\n",
      "\n",
      "[-1, 1, 1, -1, 1, -1, 1, -1] [1, 1, 1, 1, -1, 1, 1, 1]\n",
      "f1:  0.5454545454545454\n",
      "acc:  0.375\n",
      "\n",
      "[-1, -1, -1, -1, -1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.5454545454545454\n",
      "acc:  0.375\n",
      "\n",
      "[1, 1, -1, -1, -1, -1, -1, -1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.4\n",
      "acc:  0.25\n",
      "\n",
      "[1, -1, 1, 1, -1, -1, 1, 1] [1, 1, 1, 1, 1, 1, -1, 1]\n",
      "f1:  0.6666666666666666\n",
      "acc:  0.5\n",
      "\n",
      "[1, 1, 1, -1, 1, -1, 1, -1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.7692307692307693\n",
      "acc:  0.625\n",
      "\n",
      "[1, -1, -1, -1, 1, 1, 1, -1] [1, 1, 1, 1, 1, 1, -1, 1]\n",
      "f1:  0.5454545454545454\n",
      "acc:  0.375\n",
      "\n",
      "[-1, 1, 1, 1, 1, -1, 1, -1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.7692307692307693\n",
      "acc:  0.625\n",
      "\n",
      "[-1, -1, 1, 1, 1, 1, -1, 1] [1, -1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.8333333333333333\n",
      "acc:  0.75\n",
      "\n",
      "[1, -1, 1, -1, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.8571428571428571\n",
      "acc:  0.75\n",
      "\n",
      "[1, 1, -1, 1, 1, 1, 1, 1] [1, -1, 1, -1, 1, 1, -1, 1]\n",
      "f1:  0.6666666666666666\n",
      "acc:  0.5\n",
      "\n",
      "[1, 1, 1, -1, 1, 1, -1, -1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.7692307692307693\n",
      "acc:  0.625\n",
      "\n",
      "[1, -1, -1, 1, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.8571428571428571\n",
      "acc:  0.75\n",
      "\n",
      "[1, 1, 1, 1, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[-1, 1, 1, -1, -1, 1, -1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.6666666666666666\n",
      "acc:  0.5\n",
      "\n",
      "[-1, 1, 1, 1, -1, 1, 1, 1] [-1, -1, -1, 1, -1, -1, 1, 1]\n",
      "f1:  0.6666666666666666\n",
      "acc:  0.625\n",
      "\n",
      "[-1, -1, 1, -1, 1, 1, -1, -1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.5454545454545454\n",
      "acc:  0.375\n",
      "\n",
      "[-1, 1, 1, 1, -1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.8571428571428571\n",
      "acc:  0.75\n",
      "\n",
      "[1, -1, 1, 1, 1, -1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.8571428571428571\n",
      "acc:  0.75\n",
      "\n",
      "[1, 1, 1, 1, 1, 1, -1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.9333333333333333\n",
      "acc:  0.875\n",
      "\n",
      "[1, 1, -1, 1, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.9333333333333333\n",
      "acc:  0.875\n",
      "\n",
      "[-1, -1, -1, -1, -1, -1, -1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.2222222222222222\n",
      "acc:  0.125\n",
      "\n",
      "[1, 1, 1, 1, -1, 1, 1, -1] [-1, -1, -1, -1, -1, -1, -1, -1]\n",
      "f1:  0.0\n",
      "acc:  0.25\n",
      "\n",
      "[1, 1, -1, 1, 1, -1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.8571428571428571\n",
      "acc:  0.75\n",
      "\n",
      "[1, 1, 1, 1, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[-1, -1, -1, -1, -1, -1, -1, 1] [-1, -1, -1, -1, 1, -1, -1, -1]\n",
      "f1:  0.0\n",
      "acc:  0.75\n",
      "\n",
      "[-1, 1, 1, 1, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.9333333333333333\n",
      "acc:  0.875\n",
      "\n",
      "[1, 1, 1, 1, 1, -1, -1, -1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.7692307692307693\n",
      "acc:  0.625\n",
      "\n",
      "[-1, 1, -1, 1, 1, -1, -1, -1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.5454545454545454\n",
      "acc:  0.375\n",
      "\n",
      "[1, 1, -1, 1, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.9333333333333333\n",
      "acc:  0.875\n",
      "\n",
      "[-1, 1, -1, 1, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.8571428571428571\n",
      "acc:  0.75\n",
      "\n",
      "[1, -1, -1, -1, 1, -1, -1, -1] [1, -1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.4444444444444445\n",
      "acc:  0.375\n",
      "\n",
      "[1, 1, 1, 1, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[-1, 1, -1, 1, -1, -1, -1, 1] [-1, -1, -1, -1, -1, -1, -1, -1]\n",
      "f1:  0.0\n",
      "acc:  0.625\n",
      "\n",
      "[1, 1, 1, 1, -1, -1, 1, 1] [1, 1, -1, -1, -1, -1, 1, -1]\n",
      "f1:  0.6666666666666666\n",
      "acc:  0.625\n",
      "\n",
      "[-1, -1, 1, -1, 1, 1, -1, -1] [1, 1, 1, 1, 1, 1, 1, -1]\n",
      "f1:  0.6\n",
      "acc:  0.5\n",
      "\n",
      "[-1, 1, 1, -1, 1, -1, 1, -1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.6666666666666666\n",
      "acc:  0.5\n",
      "\n",
      "[-1, -1, 1, -1, -1, 1, -1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.5454545454545454\n",
      "acc:  0.375\n",
      "\n",
      "[-1, 1, 1, 1, -1, -1, -1, -1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.5454545454545454\n",
      "acc:  0.375\n",
      "\n",
      "[-1, -1, -1, 1, -1, -1, -1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.4\n",
      "acc:  0.25\n",
      "\n",
      "[1, -1, -1, 1, 1, -1, 1, -1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.6666666666666666\n",
      "acc:  0.5\n",
      "\n",
      "[1, -1, 1, 1, -1, -1, -1, -1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.5454545454545454\n",
      "acc:  0.375\n",
      "\n",
      "[1, 1, 1, 1, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[1, 1, 1, 1, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[1, 1, 1, 1, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[1, -1, 1, 1, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.9333333333333333\n",
      "acc:  0.875\n",
      "\n",
      "[-1, 1, 1, 1, 1, 1, -1, 1] [-1, -1, -1, 1, -1, -1, -1, -1]\n",
      "f1:  0.2857142857142857\n",
      "acc:  0.375\n",
      "\n",
      "[1, 1, 1, 1, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[1, 1, 1, 1, 1, 1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  1.0\n",
      "acc:  1.0\n",
      "\n",
      "[1, 1, -1, -1, -1, 1, -1, 1] [-1, -1, -1, -1, -1, 1, -1, -1]\n",
      "f1:  0.4\n",
      "acc:  0.625\n",
      "\n",
      "[1, -1, 1, 1, -1, 1, -1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.7692307692307693\n",
      "acc:  0.625\n",
      "\n",
      "[1, -1, -1, -1, -1, -1, 1, 1] [1, 1, 1, 1, 1, 1, 1, 1]\n",
      "f1:  0.5454545454545454\n",
      "acc:  0.375\n",
      "\n",
      "0.6672939129835682\n",
      "0.6185344827586207\n"
     ]
    }
   ],
   "source": [
    "# independent test\n",
    "import importlib\n",
    "import utils.utils\n",
    "importlib.reload(utils.utils)\n",
    "resource_path = './libsvm-3.23/windows/'\n",
    "dataset_path = './svmlib_dataset/subjects/'\n",
    "dataset_name = ''\n",
    "f1_means = []\n",
    "acc_means = []\n",
    "trues = []\n",
    "preds = []\n",
    "for i in range(58):\n",
    "    preds, trues = utils.utils.independent_test(resource_path, dataset_path, str(i+1)+'_arousal_raw_train', str(i+1)+'_arousal_raw_test')\n",
    "    preds = [-1 if x<=0.5 else 1 for x in preds]\n",
    "    f1 = f1_score(trues, preds)\n",
    "    print(trues, preds)\n",
    "    print('f1: ', f1)\n",
    "    f1_means.append(f1)\n",
    "    acc = [1 if preds[x]==trues[x] else 0 for x in range(8)]\n",
    "    print('acc: ', sum(acc)/8)\n",
    "    print()\n",
    "    acc_means.append(sum(acc)/8)\n",
    "print(sum(f1_means)/len(f1_means))\n",
    "print(sum(acc_means)/len(acc_means))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 181,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.7693599216610778 0.7218045112781956\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\User\\AppData\\Local\\conda\\conda\\envs\\tensorflow\\lib\\site-packages\\sklearn\\metrics\\classification.py:1135: UndefinedMetricWarning: F-score is ill-defined and being set to 0.0 due to no predicted samples.\n",
      "  'precision', 'predicted', average, warn_for)\n"
     ]
    }
   ],
   "source": [
    "f1s = []\n",
    "accs = []\n",
    "for i in range(58):\n",
    "    if i == 49: continue\n",
    "    data = open('./MultimodalAutoencoder/Results/backup/embedding/mmae_intelligent_noise_no_classification_'+str(i)+'.txt').read().split('\\n')\n",
    "    preds = data[36].replace('[','')\n",
    "    preds = preds.replace(']', '').split()\n",
    "    preds = np.asarray(preds).astype(int)\n",
    "    trues = data[37].replace('[','')\n",
    "    trues = trues.replace(']', '').split()\n",
    "    trues = np.asarray(trues).astype(int)\n",
    "    f1 = f1_score(trues, preds)\n",
    "    #f1 = (f1 + f1_score(trues, preds, pos_label=0))/2\n",
    "    acc = sum([1 if trues[x]==preds[x] else 0 for x in range(28)])/28\n",
    "    accs.append(acc)\n",
    "    f1s.append(f1)\n",
    "print(sum(f1s)/len(f1s), sum(accs)/len(accs))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 132,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[3.9 4.4 2.9 2.7 3.3 3.3 4.9 4.5 3.1 5.1 5. ]\n",
      "[3.9 4.4 2.9 2.7 3.3 3.3 4.9 4.5 3.1 5.1 5. ]\n",
      "[3.9 4.4 2.9 2.7 3.3 3.3 4.9 4.5 3.1 5.1 5. ]\n",
      "21 ./svmlib_dataset/subjects/arousal_test_21_subject\n",
      "2 :  0.5833333333333334\n",
      "22 ./svmlib_dataset/subjects/arousal_test_22_subject\n",
      "21 :  0.5555555555555556\n",
      "24 ./svmlib_dataset/subjects/arousal_test_24_subject\n",
      "22 :  0.5833333333333334\n",
      "25 ./svmlib_dataset/subjects/arousal_test_25_subject\n",
      "24 :  0.8055555555555556\n",
      "27 ./svmlib_dataset/subjects/arousal_test_27_subject\n",
      "25 :  0.5\n",
      "28 ./svmlib_dataset/subjects/arousal_test_28_subject\n",
      "27 :  0.5\n",
      "2 ./svmlib_dataset/subjects/arousal_test_2_subject\n",
      "28 :  0.3611111111111111\n",
      "31 ./svmlib_dataset/subjects/arousal_test_31_subject\n",
      "31 :  0.25\n",
      "32 ./svmlib_dataset/subjects/arousal_test_32_subject\n",
      "32 :  0.75\n",
      "42 ./svmlib_dataset/subjects/arousal_test_42_subject\n",
      "42 :  0.5277777777777778\n",
      "49 ./svmlib_dataset/subjects/arousal_test_49_subject\n",
      "49 :  0.9722222222222222\n",
      "0.5808080808080809\n"
     ]
    }
   ],
   "source": [
    "import utils.utils\n",
    "root = './svmlib_dataset/subjects/'\n",
    "\n",
    "# handle data\n",
    "test_subjects_num = 11\n",
    "mat = scipy.io.loadmat('./ASCERTAIN_Features/Dt_Personality.mat')\n",
    "mat = mat['Personality']\n",
    "personality = np.array([[0.0]*3]*test_subjects_num)\n",
    "selected_personality = [0, 2, 3]\n",
    "\n",
    "test_subjects_list = np.array([2, 21, 22, 24, 25, 27, 28, 31, 32, 42, 49])\n",
    "\n",
    "index = 0\n",
    "for i in test_subjects_list:\n",
    "    for j in range(len(selected_personality)):\n",
    "        personality[index][j] = mat[i][selected_personality[j]]\n",
    "    index += 1\n",
    "personality = np.array(personality)\n",
    "for i in range(3):\n",
    "    print(personality[:, 0])\n",
    "###    \n",
    "    \n",
    "index = 0\n",
    "accuracies = []\n",
    "for datasets in os.listdir(root):\n",
    "    accuracy = utils.utils.weighted_predict(datasets.split('_')[2], personality[index], './libsvm-3.23/windows/', './svmlib_dataset/subjects/'+datasets)\n",
    "    print(str(test_subjects_list[index]), ': ', accuracy)\n",
    "    accuracies.append(accuracy)\n",
    "    index += 1\n",
    "print(sum(accuracies)/11)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.40163954 0.31612079]\n",
      "[0.66466361]\n",
      "[0.47456971 0.38986016]\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\User\\AppData\\Local\\conda\\conda\\envs\\tensorflow\\lib\\site-packages\\sklearn\\discriminant_analysis.py:388: UserWarning: Variables are collinear.\n",
      "  warnings.warn(\"Variables are collinear.\")\n"
     ]
    }
   ],
   "source": [
    "# pca test\n",
    "# pca_eeg = PCA(n_components=0.9)\n",
    "# pca_eeg.fit(eegs)\n",
    "# print(pca_eeg.explained_variance_ratio_)\n",
    "# print(pca_eeg.explained_variance_)\n",
    "# print()\n",
    "\n",
    "# pca_ecg = PCA(n_components=0.95)\n",
    "# pca_ecg.fit(ecgs)\n",
    "# print(pca_ecg.explained_variance_ratio_)\n",
    "# print(pca_ecg.explained_variance_)\n",
    "# print()\n",
    "\n",
    "# pca_gsr = PCA(n_components=0.95)\n",
    "# pca_gsr.fit(gsrs)\n",
    "# print(pca_gsr.explained_variance_ratio_)\n",
    "# print(pca_gsr.explained_variance_)\n",
    "\n",
    "# lda test\n",
    "ys = np.array(ys)\n",
    "yss = []\n",
    "for y in ys:\n",
    "    if y[0] == -1 and y[1] == -1:\n",
    "        yss.append(0)\n",
    "    elif y[0] == -1 and y[1] == 1:\n",
    "        yss.append(1)\n",
    "    elif y[0] == 1 and y[1] == -1:\n",
    "        yss.append(2)\n",
    "    else:\n",
    "        yss.append(3)\n",
    "\n",
    "lda_eeg = LinearDiscriminantAnalysis(n_components=2)\n",
    "lda_eeg.fit(np.array(eegs), yss)\n",
    "print(lda_eeg.explained_variance_ratio_)\n",
    "\n",
    "lda_ecg = LinearDiscriminantAnalysis(n_components=1)\n",
    "lda_ecg.fit(np.array(ecgs), yss)\n",
    "print(lda_ecg.explained_variance_ratio_)\n",
    "\n",
    "lda_gsr = LinearDiscriminantAnalysis(n_components=2)\n",
    "lda_gsr.fit(np.array(gsrs), yss)\n",
    "print(lda_gsr.explained_variance_ratio_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "import importlib\n",
    "import utils.utils\n",
    "importlib.reload(utils.utils)\n",
    "output_path = './svmlib_dataset/mmae.csv'\n",
    "svm_path = './svmlib_dataset/mmae'\n",
    "modality_names = ['ECG', 'GSR', 'EEG']\n",
    "modality_nums = [14, 31, 88]\n",
    "for i in range(58):\n",
    "    utils.utils.libsvm_to_mmae_csv('./svmlib_dataset/mmae_subjects/mmae_' + str(i) + '.csv', svm_path+'_'+str(i), modality_names, modality_nums)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'True'"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "str(True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.8979591836734693"
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.metrics import f1_score\n",
    "preds = '0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0'\n",
    "trues = '0 0 0 0 0 0 0 1 0 0 0 1 0 1 0 0 0 0 0 1 0 1 0 0 0 1 0 0 0 1 0 1 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 1 0 1 0 0 0 0 0 0'\n",
    "preds = np.asarray(preds.split()).astype(int)\n",
    "preds = [1 if x==0 else 0 for x in preds]\n",
    "trues = np.asarray(trues.split()).astype(int)\n",
    "trues = [1 if x==0 else 0 for x in trues]\n",
    "f1_score(trues, preds)`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 133,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[[0, 0], [0, 0]], [[0, 0], [0, 0]], [[0, 0], [0, 0]]]\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([1, 0, 0])"
      ]
     },
     "execution_count": 133,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "xxx = [[[0,0],[0,0]]]*3\n",
    "print(xxx)\n",
    "xxx = np.asarray(xxx)\n",
    "def _x(xx):\n",
    "    xx[0][1] = 1\n",
    "_x(xxx[:,1])\n",
    "xxx[:,1,1]"
   ]
  },
  {
   "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": "3.5.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}