m@94: {
m@94:  "cells": [
m@94:   {
m@94:    "cell_type": "code",
m@94:    "execution_count": 6,
m@94:    "metadata": {},
m@94:    "outputs": [
m@94:     {
m@94:      "name": "stdout",
m@94:      "output_type": "stream",
m@94:      "text": [
m@94:       "The autoreload extension is already loaded. To reload it, use:\n",
m@94:       "  %reload_ext autoreload\n"
m@94:      ]
m@94:     }
m@94:    ],
m@94:    "source": [
m@94:     "import numpy as np\n",
m@94:     "import pandas as pd\n",
m@94:     "from sklearn.model_selection import train_test_split\n",
m@94:     "from collections import Counter\n",
m@94:     "\n",
m@94:     "%matplotlib inline\n",
m@94:     "import matplotlib.pyplot as plt\n",
m@94:     "\n",
m@94:     "%load_ext autoreload\n",
m@94:     "%autoreload 2\n",
m@94:     "\n",
m@94:     "import sys\n",
m@94:     "sys.path.append('../')\n",
m@94:     "import scripts.classification as classification\n",
m@94:     "import scripts.outliers as outliers"
m@94:    ]
m@94:   },
m@94:   {
m@94:    "cell_type": "markdown",
m@94:    "metadata": {},
m@94:    "source": [
m@94:     "## Sample 80% of the dataset, for 10 times"
m@94:    ]
m@94:   },
m@94:   {
m@94:    "cell_type": "markdown",
m@94:    "metadata": {},
m@94:    "source": [
m@94:     "Let's sample only 80% of the recordings each time (in a stratified manner) so that the set of recordings considered for each country is changed every time."
m@94:    ]
m@94:   },
m@94:   {
m@94:    "cell_type": "code",
m@94:    "execution_count": null,
m@94:    "metadata": {
m@94:     "collapsed": true
m@94:    },
m@94:    "outputs": [],
m@94:    "source": [
m@94:     "results_file = '../data/lda_data_8.pickle'\n",
m@94:     "n_iters = 10\n",
m@94:     "for n in range(n_iters):\n",
m@94:     "    print \"iteration %d\" % n\n",
m@94:     "    print results_file\n",
m@94:     "    X, Y, Yaudio = classification.load_data_from_pickle(results_file)\n",
m@94:     "    # get only 80% of the dataset.. to vary the choice of outliers\n",
m@94:     "    X, _, Y, _ = train_test_split(X, Y, train_size=0.8, stratify=Y)\n",
m@94:     "    print X.shape, Y.shape\n",
m@94:     "    # outliers\n",
m@94:     "    print \"detecting outliers...\"\n",
m@94:     "    df_global, threshold, MD = outliers.get_outliers_df(X, Y, chi2thr=0.999)\n",
m@94:     "    outliers.print_most_least_outliers_topN(df_global, N=10)\n",
m@94:     "    \n",
m@94:     "    # write output\n",
m@94:     "    print \"writing file\"\n",
m@94:     "    df_global.to_csv('../data/outliers_'+str(n)+'.csv', index=False)"
m@94:    ]
m@94:   },
m@94:   {
m@94:    "cell_type": "code",
m@94:    "execution_count": 3,
m@94:    "metadata": {},
m@94:    "outputs": [],
m@94:    "source": [
m@94:     "n_iters = 10\n",
m@94:     "ranked_countries = pd.DataFrame()\n",
m@94:     "ranked_outliers = pd.DataFrame()\n",
m@94:     "for n in range(n_iters):\n",
m@94:     "    df_global = pd.read_csv('../data/outliers_'+str(n)+'.csv')\n",
m@94:     "    df_global = df_global.sort_values('Outliers', axis=0, ascending=False).reset_index()\n",
m@94:     "    ranked_countries = pd.concat([ranked_countries, df_global['Country']], axis=1)\n",
m@94:     "    ranked_outliers = pd.concat([ranked_outliers, df_global['Outliers']], axis=1)\n",
m@94:     "ranked_countries_arr = ranked_countries.get_values()"
m@94:    ]
m@94:   },
m@94:   {
m@94:    "cell_type": "markdown",
m@94:    "metadata": {},
m@94:    "source": [
m@94:     "## Estimate precision at K"
m@94:    ]
m@94:   },
m@94:   {
m@94:    "cell_type": "markdown",
m@94:    "metadata": {},
m@94:    "source": [
m@94:     "First get the ground truth from a majority vote on the top K=10 positions."
m@94:    ]
m@94:   },
m@94:   {
m@94:    "cell_type": "code",
m@94:    "execution_count": 5,
m@94:    "metadata": {
m@94:     "collapsed": true
m@94:    },
m@94:    "outputs": [],
m@94:    "source": [
m@94:     "# majority voting + precision at K\n",
m@94:     "K_vote = 10\n",
m@94:     "country_vote = Counter(ranked_countries_arr[:K_vote, :].ravel())"
m@94:    ]
m@94:   },
m@94:   {
m@94:    "cell_type": "code",
m@94:    "execution_count": 8,
m@94:    "metadata": {},
m@94:    "outputs": [
m@94:     {
m@94:      "data": {
m@94:       "text/html": [
m@94:        "<div>\n",
m@94:        "<table border=\"1\" class=\"dataframe\">\n",
m@94:        "  <thead>\n",
m@94:        "    <tr style=\"text-align: right;\">\n",
m@94:        "      <th></th>\n",
m@94:        "      <th>index</th>\n",
m@94:        "      <th>0</th>\n",
m@94:        "    </tr>\n",
m@94:        "  </thead>\n",
m@94:        "  <tbody>\n",
m@94:        "    <tr>\n",
m@94:        "      <th>0</th>\n",
m@94:        "      <td>Pakistan</td>\n",
m@94:        "      <td>10</td>\n",
m@94:        "    </tr>\n",
m@94:        "    <tr>\n",
m@94:        "      <th>2</th>\n",
m@94:        "      <td>Chad</td>\n",
m@94:        "      <td>10</td>\n",
m@94:        "    </tr>\n",
m@94:        "    <tr>\n",
m@94:        "      <th>5</th>\n",
m@94:        "      <td>Gambia</td>\n",
m@94:        "      <td>10</td>\n",
m@94:        "    </tr>\n",
m@94:        "    <tr>\n",
m@94:        "      <th>10</th>\n",
m@94:        "      <td>Ivory Coast</td>\n",
m@94:        "      <td>10</td>\n",
m@94:        "    </tr>\n",
m@94:        "    <tr>\n",
m@94:        "      <th>12</th>\n",
m@94:        "      <td>Botswana</td>\n",
m@94:        "      <td>10</td>\n",
m@94:        "    </tr>\n",
m@94:        "    <tr>\n",
m@94:        "      <th>6</th>\n",
m@94:        "      <td>Nepal</td>\n",
m@94:        "      <td>9</td>\n",
m@94:        "    </tr>\n",
m@94:        "    <tr>\n",
m@94:        "      <th>13</th>\n",
m@94:        "      <td>Benin</td>\n",
m@94:        "      <td>8</td>\n",
m@94:        "    </tr>\n",
m@94:        "    <tr>\n",
m@94:        "      <th>8</th>\n",
m@94:        "      <td>Senegal</td>\n",
m@94:        "      <td>7</td>\n",
m@94:        "    </tr>\n",
m@94:        "    <tr>\n",
m@94:        "      <th>9</th>\n",
m@94:        "      <td>French Guiana</td>\n",
m@94:        "      <td>7</td>\n",
m@94:        "    </tr>\n",
m@94:        "    <tr>\n",
m@94:        "      <th>4</th>\n",
m@94:        "      <td>El Salvador</td>\n",
m@94:        "      <td>5</td>\n",
m@94:        "    </tr>\n",
m@94:        "    <tr>\n",
m@94:        "      <th>11</th>\n",
m@94:        "      <td>Mozambique</td>\n",
m@94:        "      <td>5</td>\n",
m@94:        "    </tr>\n",
m@94:        "    <tr>\n",
m@94:        "      <th>7</th>\n",
m@94:        "      <td>Uganda</td>\n",
m@94:        "      <td>4</td>\n",
m@94:        "    </tr>\n",
m@94:        "    <tr>\n",
m@94:        "      <th>1</th>\n",
m@94:        "      <td>Bhutan</td>\n",
m@94:        "      <td>3</td>\n",
m@94:        "    </tr>\n",
m@94:        "    <tr>\n",
m@94:        "      <th>3</th>\n",
m@94:        "      <td>Liberia</td>\n",
m@94:        "      <td>2</td>\n",
m@94:        "    </tr>\n",
m@94:        "  </tbody>\n",
m@94:        "</table>\n",
m@94:        "</div>"
m@94:       ],
m@94:       "text/plain": [
m@94:        "            index   0\n",
m@94:        "0        Pakistan  10\n",
m@94:        "2            Chad  10\n",
m@94:        "5          Gambia  10\n",
m@94:        "10    Ivory Coast  10\n",
m@94:        "12       Botswana  10\n",
m@94:        "6           Nepal   9\n",
m@94:        "13          Benin   8\n",
m@94:        "8         Senegal   7\n",
m@94:        "9   French Guiana   7\n",
m@94:        "4     El Salvador   5\n",
m@94:        "11     Mozambique   5\n",
m@94:        "7          Uganda   4\n",
m@94:        "1          Bhutan   3\n",
m@94:        "3         Liberia   2"
m@94:       ]
m@94:      },
m@94:      "execution_count": 8,
m@94:      "metadata": {},
m@94:      "output_type": "execute_result"
m@94:     }
m@94:    ],
m@94:    "source": [
m@94:     "df_country_vote = pd.DataFrame.from_dict(country_vote, orient='index').reset_index()\n",
m@94:     "df_country_vote.sort_values(0, ascending=False)"
m@94:    ]
m@94:   },
m@94:   {
m@94:    "cell_type": "code",
m@94:    "execution_count": 9,
m@94:    "metadata": {
m@94:     "collapsed": true
m@94:    },
m@94:    "outputs": [],
m@94:    "source": [
m@94:     "def precision_at_k(array, gr_truth, k):\n",
m@94:     "    return len(set(array[:k]) & set(gr_truth[:k])) / float(k)\n",
m@94:     "    \n",
m@94:     "k = 10\n",
m@94:     "ground_truth = df_country_vote['index'].get_values()\n",
m@94:     "p_ = []\n",
m@94:     "for j in range(ranked_countries_arr.shape[1]):\n",
m@94:     "    p_.append(precision_at_k(ranked_countries_arr[:, j], ground_truth, k))\n",
m@94:     "p_ = np.array(p_)"
m@94:    ]
m@94:   },
m@94:   {
m@94:    "cell_type": "code",
m@94:    "execution_count": 10,
m@94:    "metadata": {},
m@94:    "outputs": [
m@94:     {
m@94:      "name": "stdout",
m@94:      "output_type": "stream",
m@94:      "text": [
m@94:       "mean 0.67\n",
m@94:       "std 0.0640312423743\n"
m@94:      ]
m@94:     }
m@94:    ],
m@94:    "source": [
m@94:     "print 'mean', np.mean(p_) \n",
m@94:     "print 'std', np.std(p_)"
m@94:    ]
m@94:   },
m@94:   {
m@94:    "cell_type": "code",
m@94:    "execution_count": 11,
m@94:    "metadata": {},
m@94:    "outputs": [
m@94:     {
m@94:      "name": "stdout",
m@94:      "output_type": "stream",
m@94:      "text": [
m@94:       "[ 0.6  0.7  0.7  0.6  0.6  0.7  0.8  0.6  0.7  0.7]\n"
m@94:      ]
m@94:     }
m@94:    ],
m@94:    "source": [
m@94:     "print p_"
m@94:    ]
m@94:   },
m@94:   {
m@94:    "cell_type": "code",
m@94:    "execution_count": null,
m@94:    "metadata": {
m@94:     "collapsed": true
m@94:    },
m@94:    "outputs": [],
m@94:    "source": []
m@94:   }
m@94:  ],
m@94:  "metadata": {
m@94:   "kernelspec": {
m@94:    "display_name": "Python 2",
m@94:    "language": "python",
m@94:    "name": "python2"
m@94:   },
m@94:   "language_info": {
m@94:    "codemirror_mode": {
m@94:     "name": "ipython",
m@94:     "version": 2
m@94:    },
m@94:    "file_extension": ".py",
m@94:    "mimetype": "text/x-python",
m@94:    "name": "python",
m@94:    "nbconvert_exporter": "python",
m@94:    "pygments_lexer": "ipython2",
m@94:    "version": "2.7.12"
m@94:   }
m@94:  },
m@94:  "nbformat": 4,
m@94:  "nbformat_minor": 2
m@94: }