Models_comparison.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def load_labels(path):\n",
    "    with open(path, \"r\") as f:\n",
    "        all_preds = f.readlines()\n",
    "    preds_processed = [[int(e) for e in pred.strip('\\n').split()] for pred in all_preds]\n",
    "    return np.array(preds_processed)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Load label names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(\"test_labels.txt\", \"r\") as f:\n",
    "    label_names = f.readlines()[0].split(\" \")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['Drama',\n",
       " 'Film-Noir',\n",
       " 'Mystery',\n",
       " 'Thriller',\n",
       " 'Adventure',\n",
       " 'Crime',\n",
       " 'Horror',\n",
       " 'Romance',\n",
       " 'Sport',\n",
       " 'Fantasy',\n",
       " 'Biography',\n",
       " 'Action',\n",
       " 'Music',\n",
       " 'War',\n",
       " 'Documentary',\n",
       " 'Short',\n",
       " 'Animation',\n",
       " 'Comedy',\n",
       " 'Family',\n",
       " 'Musical',\n",
       " 'Sci-Fi',\n",
       " 'Western',\n",
       " 'History']"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "label_names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "label2idx = dict([(name, i) for i, name in enumerate(label_names)])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Load test gold predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Preds shape: (7799, 23)\n"
     ]
    }
   ],
   "source": [
    "gold_preds = load_labels(\"model_save_vilbert/vilbert_model_run/test_labels_gold.txt\")\n",
    "print(f\"Preds shape: {gold_preds.shape}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Count total predictions by label"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[(0, 4142),\n",
       " (1, 102),\n",
       " (2, 617),\n",
       " (3, 1567),\n",
       " (4, 821),\n",
       " (5, 1163),\n",
       " (6, 825),\n",
       " (7, 1590),\n",
       " (8, 191),\n",
       " (9, 585),\n",
       " (10, 411),\n",
       " (11, 1044),\n",
       " (12, 311),\n",
       " (13, 401),\n",
       " (14, 629),\n",
       " (15, 142),\n",
       " (16, 306),\n",
       " (17, 2611),\n",
       " (18, 518),\n",
       " (19, 253),\n",
       " (20, 586),\n",
       " (21, 210),\n",
       " (22, 345)]"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import collections\n",
    "\n",
    "preds_unfold = []\n",
    "for pred in gold_preds:\n",
    "    preds_unfold.extend(list(np.where(pred == 1)[0]))\n",
    "    \n",
    "counts = collections.Counter(preds_unfold)\n",
    "sorted(counts.items())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Get indexes of examples wehre prediction should be made by label\n",
    "\n",
    "indexesByLabel[0] -> Indexes of gold_preds where prediction should be made for class 0 <br>\n",
    "indexesByLabel[1] -> Indexes of gold_preds where prediction should be made for class 1 <br>\n",
    "..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "indexesByLabel = []\n",
    "for i in range(23):\n",
    "    exampleIndexes = []\n",
    "    for idx, pred in enumerate(gold_preds):\n",
    "        if pred[i] == 1:\n",
    "            exampleIndexes.append(idx)\n",
    "    indexesByLabel.append(exampleIndexes)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Load test predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ViLBERT preds shape: (7799, 23)\n",
      "MMBT preds shape: (7799, 23)\n",
      "MMTR preds shape: (7799, 23)\n",
      "MMTRBig preds shape: (7799, 23)\n"
     ]
    }
   ],
   "source": [
    "vilbert_preds = load_labels(\"model_save_vilbert/vilbert_model_run/test_labels_pred.txt\")\n",
    "print(f\"ViLBERT preds shape: {vilbert_preds.shape}\")\n",
    "\n",
    "mmbt_preds = load_labels(\"model_save_mmbt/mmbt_model_run/test_labels_pred.txt\")\n",
    "print(f\"MMBT preds shape: {mmbt_preds.shape}\")\n",
    "\n",
    "mmtr_preds = load_labels(\"model_save_mmtr/mmtr_model_run/test_labels_pred.txt\")\n",
    "print(f\"MMTR preds shape: {mmtr_preds.shape}\")\n",
    "\n",
    "mmtrbig_preds = load_labels(\"model_save_mmtrBig/mmtrBig_model_run/test_labels_pred.txt\")\n",
    "print(f\"MMTRBig preds shape: {mmtrbig_preds.shape}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Process counts of correct predictions for each label (0, 1, 2, ..., 22)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "from sklearn.metrics import f1_score\n",
    "\n",
    "def freqs_dataframe(data_preds, model_name, label_names):\n",
    "    countsLabel = []\n",
    "    \n",
    "    scores = f1_score(gold_preds, data_preds, average=None)\n",
    "    '''\n",
    "    for class_id, idxsList in enumerate(indexesByLabel):\n",
    "        count = 0\n",
    "        for idx in idxsList:\n",
    "            if data_preds[idx][class_id] == 1:\n",
    "                count += 1\n",
    "        countsLabel.append(count/len(idxsList))\n",
    "    '''\n",
    "        \n",
    "    df = pd.DataFrame()\n",
    "    df['label'] = label_names\n",
    "    df['freq'] = list(scores)\n",
    "    df['model'] = [model_name]*len(label_names)\n",
    "            \n",
    "    return df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "vilbert_freqs = freqs_dataframe(vilbert_preds, \"vilbert\", label_names)\n",
    "mmbt_freqs = freqs_dataframe(mmbt_preds, \"mmbt\", label_names)\n",
    "mmtr_freqs = freqs_dataframe(mmtr_preds, \"mmtr\", label_names)\n",
    "mmtrbig_freqs = freqs_dataframe(mmtrbig_preds, \"mmtr\", label_names)\n",
    "all_models_freqs = pd.concat([vilbert_freqs, mmbt_freqs, mmtrbig_freqs])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Plot and compare accuracy per label (movie genre) of all models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 176,
   "metadata": {
    "collapsed": true,
    "jupyter": {
     "outputs_hidden": true
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x864 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "sns.set(style=\"whitegrid\")\n",
    "\n",
    "f, ax = plt.subplots(figsize=(6, 12))\n",
    "\n",
    "\n",
    "sns.set_color_codes(\"muted\")\n",
    "\n",
    "sns.barplot(x=\"freq\", y=\"label\", data=vilbert_freqs,\n",
    "            label=\"ViLBERT\", edgecolor=\"r\", facecolor=(1, 1, 1, 0), linewidth=2.5)\n",
    "\n",
    "ax_bar = sns.barplot(x=\"freq\", y=\"label\", data=mmbt3_freqs,\n",
    "            label=\"mmbt3\", edgecolor=\"b\", facecolor=(1, 1, 1, 0), linewidth=2.5)\n",
    "\n",
    "ax.legend(ncol=2, loc=\"lower right\", frameon=True)\n",
    "ax.set(xlim=(0, 1), ylabel=\"\",\n",
    "       xlabel=\"Accuracy by genre\")\n",
    "sns.despine(left=True, bottom=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Ordered classses by training frequency from highest to lowest"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "ordered_classes = ['Drama',\n",
    "                 'Comedy',\n",
    "                 'Romance',\n",
    "                 'Thriller',\n",
    "                 'Crime',\n",
    "                 'Action',\n",
    "                 'Adventure',\n",
    "                 'Horror',\n",
    "                 'Documentary',\n",
    "                 'Mystery',\n",
    "                 'Sci-Fi',\n",
    "                 'Fantasy',\n",
    "                 'Family',\n",
    "                 'War',\n",
    "                 'Biography',\n",
    "                 'History',\n",
    "                 'Music',\n",
    "                 'Animation',\n",
    "                 'Musical',\n",
    "                 'Western',\n",
    "                 'Sport',\n",
    "                 'Short',\n",
    "                 'Film-Noir']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x720 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "sns.set(style=\"whitegrid\")\n",
    "\n",
    "f, ax = plt.subplots(figsize=(6, 10))\n",
    "ax_points = sns.stripplot(x=\"freq\", y=\"label\", data=all_models_freqs, hue=\"model\", jitter=0, order=ordered_classes)\n",
    "ax.legend(ncol=2, loc=\"upper left\", frameon=True, bbox_to_anchor=(0.99, 1.01))\n",
    "ax.set(xlabel=\"F1 score\", ylabel=\"genre\")\n",
    "\n",
    "for i in range(len(label_names)):\n",
    "    ax_points.axhline(i, color='black', alpha=0.1, lw = 2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Plot correct predictions of each example by label (movie genre)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "def pred_class_dataframe(data_preds_list, label_id):\n",
    "    binary_preds = []\n",
    "    ids = []\n",
    "    model_name = []\n",
    "    origin_ids = []\n",
    "\n",
    "    for name, data_preds in data_preds_list.items():\n",
    "        for i, idx in enumerate(indexesByLabel[label_id]):\n",
    "            binary_preds.append(1 if data_preds[idx][label_id] == gold_preds[idx][label_id] else 0)\n",
    "            ids.append(i)\n",
    "            model_name.append(name)\n",
    "            origin_ids.append(idx)\n",
    "    \n",
    "    df = pd.DataFrame()\n",
    "    df['result'] = binary_preds\n",
    "    df['id'] = ids\n",
    "    df['model'] = model_name\n",
    "    df['origin_id'] = origin_ids\n",
    "            \n",
    "    return df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "preds_all_models = {'vilbert': vilbert_preds,\n",
    "                    'mmbt': mmbt_preds,\n",
    "                    'mmtrBig': mmtrbig_preds}\n",
    "\n",
    "results_model = pred_class_dataframe(preds_all_models, label2idx['Film-Noir'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "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>result</th>\n",
       "      <th>id</th>\n",
       "      <th>model</th>\n",
       "      <th>origin_id</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>vilbert</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>vilbert</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "      <td>vilbert</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1</td>\n",
       "      <td>3</td>\n",
       "      <td>vilbert</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0</td>\n",
       "      <td>4</td>\n",
       "      <td>vilbert</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>301</th>\n",
       "      <td>0</td>\n",
       "      <td>97</td>\n",
       "      <td>mmtrBig</td>\n",
       "      <td>97</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>302</th>\n",
       "      <td>0</td>\n",
       "      <td>98</td>\n",
       "      <td>mmtrBig</td>\n",
       "      <td>98</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>303</th>\n",
       "      <td>0</td>\n",
       "      <td>99</td>\n",
       "      <td>mmtrBig</td>\n",
       "      <td>99</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>304</th>\n",
       "      <td>0</td>\n",
       "      <td>100</td>\n",
       "      <td>mmtrBig</td>\n",
       "      <td>100</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>305</th>\n",
       "      <td>0</td>\n",
       "      <td>101</td>\n",
       "      <td>mmtrBig</td>\n",
       "      <td>101</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>306 rows × 4 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "     result   id    model  origin_id\n",
       "0         0    0  vilbert          0\n",
       "1         0    1  vilbert          1\n",
       "2         1    2  vilbert          2\n",
       "3         1    3  vilbert          3\n",
       "4         0    4  vilbert          4\n",
       "..      ...  ...      ...        ...\n",
       "301       0   97  mmtrBig         97\n",
       "302       0   98  mmtrBig         98\n",
       "303       0   99  mmtrBig         99\n",
       "304       0  100  mmtrBig        100\n",
       "305       0  101  mmtrBig        101\n",
       "\n",
       "[306 rows x 4 columns]"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "f, ax = plt.subplots(figsize=(6, 10))\n",
    "sns.stripplot(x=\"id\", y=\"result\", data=results_model, hue=\"model\", jitter=0)\n",
    "ax.legend(ncol=2, loc=\"lower right\", frameon=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 255,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f19eca91978>"
      ]
     },
     "execution_count": 255,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 468x468 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "f, ax = plt.subplots(figsize=(6.5, 6.5))\n",
    "sns.despine(f, left=True, bottom=True)\n",
    "#clarity_ranking = [\"I1\", \"SI2\", \"SI1\", \"VS2\", \"VS1\", \"VVS2\", \"VVS1\", \"IF\"]\n",
    "sns.scatterplot(x=\"id\", y=\"result\",\n",
    "                hue=\"model\",\n",
    "                palette=\"ch:r=-.2,d=.3_r\",\n",
    "                sizes=(1, 1), linewidth=0,\n",
    "                data=results_model, ax=ax)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Get test dataset idexes for a specific label where all models' predictions are incorrect"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "def pred_class_dataframe_ids(data_preds, label_id):\n",
    "    binary_preds = []\n",
    "    origin_ids = []\n",
    "\n",
    "    for i, idx in enumerate(indexesByLabel[label_id]):\n",
    "        binary_preds.append(1 if data_preds[idx][label_id] == gold_preds[idx][label_id] else 0)\n",
    "        origin_ids.append(idx)\n",
    "    \n",
    "    df = pd.DataFrame()\n",
    "    df['result'] = binary_preds\n",
    "    df['origin_id'] = origin_ids\n",
    "            \n",
    "    return df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "target_label = 'Film-Noir'\n",
    "\n",
    "vilbert_results_model_ids = pred_class_dataframe_ids(vilbert_preds, label2idx[target_label])\n",
    "mmbt_results_model_ids = pred_class_dataframe_ids(mmbt_preds, label2idx[target_label])\n",
    "mmtrbig_results_model_ids = pred_class_dataframe_ids(mmtrbig_preds, label2idx[target_label])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [],
   "source": [
    "incorrect_pred = vilbert_results_model_ids['result'] == 0\n",
    "vilbert_incorrect = vilbert_results_model_ids[incorrect_pred]\n",
    "\n",
    "incorrect_pred = mmbt_results_model_ids['result'] == 0\n",
    "mmbt_incorrect = mmbt_results_model_ids[incorrect_pred]\n",
    "\n",
    "incorrect_pred = mmtrbig_results_model_ids['result'] == 0\n",
    "mmtr_incorrect = mmtrbig_results_model_ids[incorrect_pred]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0       1\n",
       "1       5\n",
       "2       7\n",
       "3       8\n",
       "4      12\n",
       "5      14\n",
       "6      21\n",
       "7      24\n",
       "8      26\n",
       "9      33\n",
       "10     34\n",
       "11     38\n",
       "12     39\n",
       "13     41\n",
       "14     47\n",
       "15     51\n",
       "16     52\n",
       "17     57\n",
       "18     65\n",
       "19     72\n",
       "20     74\n",
       "21     75\n",
       "22     82\n",
       "23     92\n",
       "24     93\n",
       "25     95\n",
       "26     97\n",
       "27     99\n",
       "28    100\n",
       "29    101\n",
       "Name: origin_id, dtype: int64"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "temp_merge = pd.merge(vilbert_incorrect, mmbt_incorrect, how='inner', on=['origin_id'])\n",
    "all_incorrect_preds = pd.merge(mmtr_incorrect, temp_merge, how='inner', on=['origin_id'])['origin_id']\n",
    "all_incorrect_preds"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Explore incorrect predictions"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Get all indexes of predicted examples that were incorrect"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "all_incorrect_preds = list(all_incorrect_preds)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Get all test set examples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "\n",
    "data = [json.loads(l) for l in open('/home/est_posgrado_isaac.bribiesca/mmimdb/test.jsonl')]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Filter the incorrect test examples in all models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_filtered = [data[i] for i in all_incorrect_preds]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Keep only the genres"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "genres = [row['label'] for row in data_filtered]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[['Drama', 'Film-Noir'],\n",
       " ['Drama', 'Film-Noir', 'Mystery', 'Romance', 'Thriller'],\n",
       " ['Crime', 'Drama', 'Film-Noir', 'Romance'],\n",
       " ['Crime', 'Drama', 'Film-Noir', 'Thriller'],\n",
       " ['Crime', 'Drama', 'Film-Noir'],\n",
       " ['Drama', 'Film-Noir', 'Thriller'],\n",
       " ['Crime', 'Film-Noir', 'Thriller'],\n",
       " ['Biography', 'Crime', 'Drama', 'Film-Noir'],\n",
       " ['Drama', 'Film-Noir'],\n",
       " ['Drama', 'Film-Noir'],\n",
       " ['Adventure', 'Drama', 'Film-Noir', 'War'],\n",
       " ['Crime', 'Drama', 'Film-Noir', 'Thriller'],\n",
       " ['Crime', 'Film-Noir', 'Thriller'],\n",
       " ['Crime', 'Film-Noir', 'Mystery', 'Thriller'],\n",
       " ['Crime', 'Drama', 'Film-Noir'],\n",
       " ['Crime', 'Drama', 'Film-Noir'],\n",
       " ['Crime', 'Drama', 'Film-Noir', 'Thriller'],\n",
       " ['Thriller', 'Film-Noir', 'Drama'],\n",
       " ['Crime', 'Drama', 'Film-Noir', 'Thriller'],\n",
       " ['Crime', 'Drama', 'Film-Noir', 'Thriller'],\n",
       " ['Drama', 'Film-Noir', 'Sport', 'Thriller'],\n",
       " ['Drama', 'Film-Noir', 'Thriller'],\n",
       " ['Crime', 'Film-Noir', 'Thriller'],\n",
       " ['Drama', 'Film-Noir', 'Thriller'],\n",
       " ['Crime', 'Drama', 'Film-Noir', 'Thriller'],\n",
       " ['Drama', 'Film-Noir', 'Romance'],\n",
       " ['Drama', 'Film-Noir', 'Romance'],\n",
       " ['Crime', 'Drama', 'Film-Noir'],\n",
       " ['Drama', 'Film-Noir', 'Thriller'],\n",
       " ['Crime', 'Film-Noir', 'Thriller']]"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "genres"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Get genre names predicted by model in the incorrect examples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [],
   "source": [
    "mmbt_incorrect_intersection = [mmbt_preds[i] for i in all_incorrect_preds]\n",
    "mmbt_incorrect_label_names = [[label_names[i] for i in np.where(row == 1)[0]] for row in mmbt_incorrect_intersection]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[['Drama', 'Comedy'],\n",
       " ['Drama', 'Comedy'],\n",
       " ['Drama', 'Crime', 'Comedy'],\n",
       " ['Drama', 'Crime', 'Action', 'Comedy'],\n",
       " ['Drama', 'Crime', 'Action'],\n",
       " ['Mystery', 'Thriller', 'Crime', 'Horror'],\n",
       " ['Drama', 'Mystery', 'Thriller', 'Crime'],\n",
       " ['Drama', 'Crime'],\n",
       " ['Drama', 'Romance', 'Comedy'],\n",
       " ['Drama', 'Romance'],\n",
       " ['Drama', 'Adventure', 'War'],\n",
       " ['Drama', 'Thriller', 'Action'],\n",
       " ['Crime', 'Romance', 'Comedy'],\n",
       " ['Thriller', 'Horror', 'Comedy'],\n",
       " ['Drama', 'Romance'],\n",
       " ['Drama', 'War'],\n",
       " ['Drama', 'Thriller', 'Crime', 'Action'],\n",
       " ['Drama', 'Thriller', 'Horror', 'War'],\n",
       " ['Drama', 'Thriller', 'Crime'],\n",
       " ['Drama', 'Romance', 'Biography'],\n",
       " ['Drama', 'Crime', 'Sport', 'Biography'],\n",
       " ['Drama', 'Romance'],\n",
       " ['Drama', 'Thriller', 'Crime', 'Comedy'],\n",
       " ['Drama', 'Mystery', 'Horror'],\n",
       " ['Drama', 'Crime'],\n",
       " ['Drama', 'Mystery', 'Thriller', 'Romance'],\n",
       " ['Drama', 'Romance', 'Comedy'],\n",
       " ['Drama', 'Thriller', 'Crime'],\n",
       " ['Drama', 'Thriller', 'Romance'],\n",
       " ['Drama', 'Crime']]"
      ]
     },
     "execution_count": 54,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mmbt_incorrect_label_names"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Compare true vs pred genres in incorrect predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "true - pred\n",
      "['Drama', 'Film-Noir'] - ['Drama', 'Comedy']\n",
      "['Drama', 'Film-Noir', 'Mystery', 'Romance', 'Thriller'] - ['Drama', 'Comedy']\n",
      "['Crime', 'Drama', 'Film-Noir', 'Romance'] - ['Drama', 'Crime', 'Comedy']\n",
      "['Crime', 'Drama', 'Film-Noir', 'Thriller'] - ['Drama', 'Crime', 'Action', 'Comedy']\n",
      "['Crime', 'Drama', 'Film-Noir'] - ['Drama', 'Crime', 'Action']\n",
      "['Drama', 'Film-Noir', 'Thriller'] - ['Mystery', 'Thriller', 'Crime', 'Horror']\n",
      "['Crime', 'Film-Noir', 'Thriller'] - ['Drama', 'Mystery', 'Thriller', 'Crime']\n",
      "['Biography', 'Crime', 'Drama', 'Film-Noir'] - ['Drama', 'Crime']\n",
      "['Drama', 'Film-Noir'] - ['Drama', 'Romance', 'Comedy']\n",
      "['Drama', 'Film-Noir'] - ['Drama', 'Romance']\n",
      "['Adventure', 'Drama', 'Film-Noir', 'War'] - ['Drama', 'Adventure', 'War']\n",
      "['Crime', 'Drama', 'Film-Noir', 'Thriller'] - ['Drama', 'Thriller', 'Action']\n",
      "['Crime', 'Film-Noir', 'Thriller'] - ['Crime', 'Romance', 'Comedy']\n",
      "['Crime', 'Film-Noir', 'Mystery', 'Thriller'] - ['Thriller', 'Horror', 'Comedy']\n",
      "['Crime', 'Drama', 'Film-Noir'] - ['Drama', 'Romance']\n",
      "['Crime', 'Drama', 'Film-Noir'] - ['Drama', 'War']\n",
      "['Crime', 'Drama', 'Film-Noir', 'Thriller'] - ['Drama', 'Thriller', 'Crime', 'Action']\n",
      "['Thriller', 'Film-Noir', 'Drama'] - ['Drama', 'Thriller', 'Horror', 'War']\n",
      "['Crime', 'Drama', 'Film-Noir', 'Thriller'] - ['Drama', 'Thriller', 'Crime']\n",
      "['Crime', 'Drama', 'Film-Noir', 'Thriller'] - ['Drama', 'Romance', 'Biography']\n",
      "['Drama', 'Film-Noir', 'Sport', 'Thriller'] - ['Drama', 'Crime', 'Sport', 'Biography']\n",
      "['Drama', 'Film-Noir', 'Thriller'] - ['Drama', 'Romance']\n",
      "['Crime', 'Film-Noir', 'Thriller'] - ['Drama', 'Thriller', 'Crime', 'Comedy']\n",
      "['Drama', 'Film-Noir', 'Thriller'] - ['Drama', 'Mystery', 'Horror']\n",
      "['Crime', 'Drama', 'Film-Noir', 'Thriller'] - ['Drama', 'Crime']\n",
      "['Drama', 'Film-Noir', 'Romance'] - ['Drama', 'Mystery', 'Thriller', 'Romance']\n",
      "['Drama', 'Film-Noir', 'Romance'] - ['Drama', 'Romance', 'Comedy']\n",
      "['Crime', 'Drama', 'Film-Noir'] - ['Drama', 'Thriller', 'Crime']\n",
      "['Drama', 'Film-Noir', 'Thriller'] - ['Drama', 'Thriller', 'Romance']\n",
      "['Crime', 'Film-Noir', 'Thriller'] - ['Drama', 'Crime']\n"
     ]
    }
   ],
   "source": [
    "print(\"true - pred\")\n",
    "for gold, pred in zip(genres, mmbt_incorrect_label_names):\n",
    "    print(f'{gold} - {pred}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Missing labels:\n",
      "\n",
      "{'Film-Noir'}\n",
      "{'Romance', 'Film-Noir', 'Thriller', 'Mystery'}\n",
      "{'Romance', 'Film-Noir'}\n",
      "{'Film-Noir', 'Thriller'}\n",
      "{'Film-Noir'}\n",
      "{'Film-Noir', 'Drama'}\n",
      "{'Film-Noir'}\n",
      "{'Biography', 'Film-Noir'}\n",
      "{'Film-Noir'}\n",
      "{'Film-Noir'}\n",
      "{'Film-Noir'}\n",
      "{'Film-Noir', 'Crime'}\n",
      "{'Film-Noir', 'Thriller'}\n",
      "{'Film-Noir', 'Crime', 'Mystery'}\n",
      "{'Film-Noir', 'Crime'}\n",
      "{'Film-Noir', 'Crime'}\n",
      "{'Film-Noir'}\n",
      "{'Film-Noir'}\n",
      "{'Film-Noir'}\n",
      "{'Film-Noir', 'Crime', 'Thriller'}\n",
      "{'Film-Noir', 'Thriller'}\n",
      "{'Film-Noir', 'Thriller'}\n",
      "{'Film-Noir'}\n",
      "{'Film-Noir', 'Thriller'}\n",
      "{'Film-Noir', 'Thriller'}\n",
      "{'Film-Noir'}\n",
      "{'Film-Noir'}\n",
      "{'Film-Noir'}\n",
      "{'Film-Noir'}\n",
      "{'Film-Noir', 'Thriller'}\n"
     ]
    }
   ],
   "source": [
    "print(\"Missing labels:\\n\")\n",
    "for gold, pred in zip(genres, mmbt_incorrect_label_names):\n",
    "    missing = set(gold).difference(set(pred))\n",
    "    #missing.remove(target_label)\n",
    "    print(missing)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Incorrect labels:\n",
      "\n",
      "{'Comedy'}\n",
      "{'Comedy'}\n",
      "{'Comedy'}\n",
      "{'Action', 'Comedy'}\n",
      "{'Action'}\n",
      "{'Horror', 'Crime', 'Mystery'}\n",
      "{'Drama', 'Mystery'}\n",
      "set()\n",
      "{'Romance', 'Comedy'}\n",
      "{'Romance'}\n",
      "set()\n",
      "{'Action'}\n",
      "{'Romance', 'Comedy'}\n",
      "{'Horror', 'Comedy'}\n",
      "{'Romance'}\n",
      "{'War'}\n",
      "{'Action'}\n",
      "{'Horror', 'War'}\n",
      "set()\n",
      "{'Romance', 'Biography'}\n",
      "{'Biography', 'Crime'}\n",
      "{'Romance'}\n",
      "{'Comedy', 'Drama'}\n",
      "{'Horror', 'Mystery'}\n",
      "set()\n",
      "{'Thriller', 'Mystery'}\n",
      "{'Comedy'}\n",
      "{'Thriller'}\n",
      "{'Romance'}\n",
      "{'Drama'}\n"
     ]
    }
   ],
   "source": [
    "print(\"Incorrect labels:\\n\")\n",
    "for gold, pred in zip(genres, mmbt_incorrect_label_names):\n",
    "    missing = set(pred).difference(set(gold))\n",
    "    #missing.remove(target_label)\n",
    "    print(missing)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### GMU gates visualization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 300,
   "metadata": {},
   "outputs": [],
   "source": [
    "gates = np.load('model_save_mmtrgmu/mmtrgmu_model_run/gates.npy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 301,
   "metadata": {},
   "outputs": [],
   "source": [
    "gates_rounded = gates.round()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 306,
   "metadata": {},
   "outputs": [],
   "source": [
    "gates_freqs = np.sum(gates_rounded, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 309,
   "metadata": {},
   "outputs": [],
   "source": [
    "gates_freqs = gates_freqs/30"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 310,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0.6333333 , 0.56666666, 0.53333336, ..., 0.6666667 , 0.6333333 ,\n",
       "       0.53333336], dtype=float32)"
      ]
     },
     "execution_count": 310,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gates_freqs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 312,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0.5582661 , 0.56419015, 0.49996036, ..., 0.5745738 , 0.5874417 ,\n",
       "       0.53913426], dtype=float32)"
      ]
     },
     "execution_count": 312,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.mean(gates, axis=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Training and Validation loss"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_loss(path):\n",
    "    \n",
    "    with open(path, \"r\") as f:\n",
    "        log = f.readlines()\n",
    "        \n",
    "    start_index = [i for i, s in enumerate(log) if \"Training..\" in s][0]\n",
    "    log_metrics = log[start_index+1:]\n",
    "    \n",
    "    train_loss = [s.split(\"-\")[-1].split(\":\")[-1].strip().strip(\"\\n\") for s in log_metrics if \"Train\" in s]\n",
    "    train_loss = np.array([float(v) for v in train_loss])\n",
    "    \n",
    "    val_loss = [s.split(\"-\")[-1].split(\"|\")[0].split(\":\")[-1].strip().strip(\"\\n\") for s in log_metrics if \"Val\" in s]\n",
    "    val_loss = np.array([float(v) for v in val_loss])\n",
    "    \n",
    "    return train_loss, val_loss"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Graph loss vs epoch MMBT"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "train_loss, val_loss = get_loss(\"model_save_mmbt/mmbt_model_run/logfile.log\")\n",
    "\n",
    "plt.subplot(111)\n",
    "plt.plot(train_loss, label='train')\n",
    "plt.plot(val_loss, label='val')\n",
    "plt.xlabel('epoch')\n",
    "plt.ylabel('loss')\n",
    "plt.legend(loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Graph loss vs epoch MMTRBig"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 116,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYsAAAEMCAYAAAA1VZrrAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3deXhU9dn/8fesyWTfw4SwRtn3IMuDthK2WFCwCiKK1RasotJaFxBRDGBrKNU+IsoPvFoXBH2oCwYQEBBFVJSISMJaloTsCdm3Wc/vj8BIgDAhy0xmcr+ui4tZvuec+/bIfGbOzPkelaIoCkIIIcRVqN1dgBBCiLZPwkIIIYRTEhZCCCGckrAQQgjhlISFEEIIp7TuLqCl2O12qqqq0Ol0qFQqd5cjhBAeQVEULBYL/v7+qNUNf37wmrCoqqri+PHj7i5DCCE8Uo8ePQgMDGzwea8JC51OB9Q1rNfrr3n5tLQ0+vXr19JluZW39eRt/YD39eRt/YD39XRpP2azmePHjzteQxviNWFx4dCTXq/Hx8enSeto6nJtmbf15G39gPf15G39gPf1dKV+nB2+ly+4hRBCOCVhIYQQwimvOQx1NXa7naysLKqqqhoco9VqOXLkiAuran0XevL39yc2Nvaqv3QQQoiraRdhUVRUhEqlomfPng2+YFZVVeHv7+/iylpXVVUVBoOB7OxsioqKiIqKcndJQggP1S7eapaWlhIdHd0u31mr1Wqio6MpKytzdylCCA/WLl49bTab05+FeTOdTofVanV3GUIID9YuwgKc/yzMm7Xn3oXwduZz2WS+/gi12a17UnK7CYu2ZMWKFZjN5mte7tChQzzxxBOtUJEQwhPZLSYKPlqO3VSNNiiyVbclYeEGr732GhaL5bLHnR0q6t+/P//4xz9aqywhhIc5t+1NzAVniZr8J7SBoa26rXbxa6i2JCkpCYDp06ejVqvp2LEjRqORM2fOUFJSwkcffcQTTzzB6dOnsVgsdO7cmb/+9a8EBwezb98+kpOT+eijj8jKyuKOO+5g+vTpfPnll9TU1PDiiy8ydOhQN3cohHCFip+/oOLgLkJG3Ylf90Gtvj2XhcWcOXPIyspCrVbj5+fHc889R+/eveuNsdlsLF26lD179qBSqXjwwQeZOnVqi9axa38mn3+fednjNpsNjUbTrHWPG9aZhKGdrzpm0aJFrFu3jvfffx9/f3/mz5/PgQMHWLt2LX5+fgA8++yzhIWFAfDKK6+wZs0annzyycvWVVpayqBBg3j88cf59NNPWb58Oe+//36zehBCtH3mgkyKPluNb5e+hP5qmku26bKwSE5OdsxouGPHDhYsWMDHH39cb0xKSgqZmZls376d0tJSpkyZwsiRI4mNjXVVmW6RmJjoCAqAjRs3kpKSgsViobq6mq5du15xOT8/P0aPHg3AoEGDSE5OdkW5QohLmM/lUJn2JYrVjGK11P2xWS66bUaxWlGsZlAUgodPIqDvTU3alt1cQ/5Hy1H7+BE1+XFU6ua9yW0sl4XFxVPfVlZWXvEXOlu2bGHq1Kmo1WrCwsIYO3YsW7duZdasWS1WR8LQK7/7d+dJeRcHxf79+1m/fj3vv/8+YWFhpKSk8H//939XXO7i2XXVarX8PFYIN7DVVpG3Lglr+TlUWh0qrb7ub43ul/saLSqtHrWvH9aKEgo++SfmgkxCb74blarxXx0rikLRZ6uxFOdinPF8q39PcTGXfmfx7LPPsnfvXhRF4c0337zs+dzcXGJiYhz3jUYjeXl5rizRJfz9/amsrLxiOJWXlxMQEEBISAhms5kPP/zQDRUKIRpDURSKtqzCWllCzP1/w7fj9c6XsVko2vompd98hLmo7stptd7QqO1V/LSDyrSvCP3VdAxd+ze3/Gvi0rB48cUXAfjkk09YtmwZa9asafFtpKWlXfaYVqu96rxQFzRmTEu49957mTlzJj4+PsTExGAymRzbjo+PJyYmhgkTJhAVFUWfPn1IT0+nqqqK2tpaxxUBa2pq6tV86f1LezKbzaSmprqkv9bkDT1cytt68rZ+oOGe9Fk/43/kG2quv5n0vHLIa2Tv0Tfg00uFcnQH/131F6qGTMVuCL7qIpryfAK/ewtreDdOGbpCM/47N2kfKW7Sv39/pbi4uN5js2fPVj777DPH/aSkJGXNmjWNWl9tba2yf/9+pba29rLnDh8+7HT5ysrKRm3Hk1zcU2P+G7R1+/fvd3cJLc7bevK2fhSl4Z5MRdnKqeQZSva7zyl2m7VJ6646eUA5/fd7ldMv36/UZDb8b9RWW6VkrJyjnPnnLMVaWdqkbV1waT9Xe+28mEvOs6iqqiI3N9dxf9euXQQHBxMSElJvXGJiIhs2bMBut1NcXMyOHTuYMGGCK0oUQrQDiqJQ/uN2Crf8P+wWU9PXY7NQ8Mk/UWm1RN32pyZ/yezXfRAxD7yExtefnLUvUHFw1xVrLtz8OtbSAqJufxyN/9U/gbQWlxyGqqmp4U9/+hM1NTWo1WqCg4NZtWoVKpWK2bNnM3fuXPr378/kyZM5ePAg48ePB+CRRx6hU6dOrihRCOHl7BYTRZ+tpvLQbgAsRWfpMO0Z1L7X/sOW4t3rMeedJPrOp9EGhTerLn14R2Luf4mCj/9B4aaVmAszCUuY6Qig8tStVB35lrDR92Do3KdZ22oOl4RFREREg7/oufh7C41G4zhpTQghWoq1rJC8/yzDnHeK0JvuQhceQ8Gnr5KzdhHGu5+7pnfr1acPUvbdRgKHjMe/5/AWqU9jCKDD9IWc+/wtyvalYC7KInrK41iKczm34y0McUMIHjmlRbbVVHIGtxDCq9WcOUT+xy+j2KxET52Pf48bAFD7+JH/4d/JeXchxhmL0AZFOF2XraqMwo2voouIJXzs/S1ap0qtIWLCH9BHdqJo25tkv70AxWpG4x9C1G1zr+kntq1B5oYSQnglRVEo3ZdC7rrFaPyC6PjAS46gAPC7bggd7n4Oa2UpOW8/i6U4x+n6CjetxFZbSdSUx1HrfFql7qAh4zHOeB5bVSnW8nNE3/4XNH6BzhdsZRIWQgivY7eYKNz4vxTveAu/HjfQ8f6X0Id3vGycoXMfYu5Jwm41k/POQkz5ZxpcZ3nqVqr/m0r4mPvwie7aesUDhi79iJ31MjG/+yu+sT1bdVuNJWEhhPAqltJ8ct5aQGX614TePIPoO55E7dPwSW8+xu7EzFwCai25a5+nNuvoZWPMBRkU73gbQ9xggob+pjXLd9AGheMbc51LttUYEhYeYObMmXzxxRfuLkOINq/61EGy//U01vJCOty1gNBRdzTqWL8+IpaY3y1F4xdE7rrFVJ/66ZcnbRbyP3kFta8/Ubc+1m4vJiZhIYTwaHZTNbXZxyn+8n3y3l+KJiCMjg8k43fdkGtajy44CuPMpehCjeR98Dcqj34LgOHYTiyFZ4m89VG3nePQFrS7X0NV/Lz7iie+2Gw2ypo5RXngwAQCB9x81TErV66krKyMBQsWAFBSUkJiYiLJycm88cYbmEwmbDYbDz30EBMnTmxWPUJ4E1tNBZaibMxFZzEXZWEpOou5MAtbxTnHGP/e/0PkpDmNnmvpUtqAEIz3JpH3wV8p+OhlagaNwTfzR4KH34pf3OCWasUjtbuwcLfbb7+dadOm8fTTT6PVatm0aRMJCQkMHjyYdevWodFoKCoq4re//S033ngjwcHt952MEDVnDlHy9X+wFGVhqyp1PK7S+aALj8XQtR/6iFh0EZ3QR8SiDe3Q7MNEGkMAxhnPk//hMioOfI41KJqwm+9pbiser92FReCAm6/47t9VU5THxMQQFxfHl19+yZgxY/j4449ZsGABxcXFLFiwgIyMDDQaDWVlZZw+fZpBg1r/ClhCtEXWyhLyP/oHKp0Phrgh6CNjHcGgDY5o1fMO1HpfOkx9hrLUzzhtCUCl1bXatjxFuwuLtuD222/nk08+oVOnTlRUVDB06FB+97vfkZCQwGuvvYZKpWLChAmYTE2fu0YIT6YoCkWb30CxmIi5byn6CNdfAE2l1REy/DbsXjiLblPIF9xuMGHCBH744Qf+9a9/cfvttwNQUVFBx44dUalU7N27l4yMDDdXKYT7VBzcSfV/UwkbfY9bgkJcTj5ZuIHBYGDMmDF89NFH7Ny5E4AnnniCpKQk1qxZQ8+ePenZs22ciCOEq1lKCzj3+b/x7dKXoBtcc06DcE7Cwk1efPFFx8WgAEaNGsX27duvOPbdd991VVlCuJWi2ClMeQ1QETnpUbfPhyR+IXtCCNFmlP+whdrMdMLHPYAuJMrd5YiLSFgIIdoEc1EWxbvW4nddPIEDE9xdjrhEuwkLRVHcXYLbtOfehWdQbFYKP30Vld6XiIkPt9spNdqydhEWGo0Gi8Xi7jLcxmKxoNXK11Oi7Sr95iNMuSeJuOWPaANC3V2OuIJ2ERYhISHk5+djt9vdXYrL2e128vPz5Uxw0WaZck9S8vV/COh7EwG9R7q7HNGAdvF2MyIigqysLI4dO9bgGLPZjF6vd2FVre9CT/7+/kREOL8KmBCuZreaKfj0VTR+wYRPmOXucsRVtIuwUKvVdO7c+apjUlNTGThwoIsqcg1v7El4l5Ld67EUZdFh+kI0hgB3lyOuol0chhJCtD01memU7UshcMj4dj+jqyeQsBBCuJzdVEPhp6+hDYkifMx97i5HNEK7OAwlhHA/RVGwVZVhLSuk/IfNWMsKiblvSZOvPSFcyyVhUVJSwtNPP01mZiZ6vZ4uXbqwePFiwsLC6o1bsWIF69atIyqq7szNIUOGsGjRIleUKES7Yso/Q/Wx7/GJicOnY88W+b5AUeyoasupPXsUS1kB1rLC838u3C5CsZod40P+57f4durd7O0K13BJWKhUKmbNmsXw4cMBSE5OZvny5fz1r3+9bOyUKVOYN2+eK8oSol1SrBYKPv4HlnM5jsd0EbH4xvbCN7Ynvp16oQ01XvXEOLupGnNhJub8M5gKMjAXZGAuyCTEXEPORePUfkHogiPRR3bG77qhaIMj0QZHoguJRhfZqRW7FC3NJWEREhLiCAqAQYMGsX79eldsWghxidLvNmI5l0P0HU+hNgRQe/YotVnHqDr6LRU/7QDqXuR9O9YFh0/MddhqKjDnXwiFM1hLCxzrU/v6o4/qQuCA0eRW24kbcD4UgiJQ633d1aZoYS7/zsJut7N+/XoSEq4898vmzZv5+uuviYyM5LHHHmPwYPmVhBAtxVKSR+neD/HvNRL/XiMAMHTpB9QdRrIUZVObVRcepqyjVJ/44ZeFVWp0YUZ8jNcROGgs+qgu+ER3RRMY7vgUkpGaKr9s8lIqxcUTByUlJZGfn89rr72GWl3/x1iFhYWEhISg0+nYu3cvTz75JFu2bCE01Pnp/yaTibS0tNYqWwjPpygEpH6AtiSLspv+iOIb6HQRlbkKTVkeit4PW0AEaOTyot6qX79++Pj4NPi8Sz9ZJCcnk5GRwapVqy4LCoDIyEjH7VGjRmE0Gjlx4gTDhg1r9DacNdyQ1NRU4uPjr3m5tszbevK2fsC1PVUe+ZaColOEj3uAuGE3t8o2ZB+1fZf209g32i47z+KVV14hLS2NlStXNjitRn5+vuP2kSNHyM7Oplu3bq4qUQivZTfVcO7zf6GP7kbQ0FvcXY7wQC75ZHHixAlWrVpF165dmT59OgCxsbGsXLmS2bNnM3fuXPr378/LL79Meno6arUanU7HsmXL6n3aEEI0TclX72OrKCH6jqdQqTXuLkd4IJeExfXXX9/gJH5r1qxx3E5OTnZFOUK0K6a805T9sIXAwePw7djD3eUIDyXTfQjhxRTFTtHW1agNAYSNvsfd5QgPJmEhhBerOLADU/Zxwsf+TmZ1Fc0iYSGEl7JVlVH8xVp8u/QloN+v3V2O8HASFkJ4qXM738FuNhGR+KBc01o0m4SFEF6oJiONykO7CRk5GX1ErLvLEV5AwkIIL6PYLBRtXYM2JIqQUXe4uxzhJSQshPAypd+lYCnKImLCLNS6a5/NQIgrkbAQwotYSvMp/XoD/r1G4Hed90xRIdxPrpQnhBvYLSaqjnyD/uxJqvytaPyDUfsFofELQu3r7/QLaUVRUCy12KrKsFWX1/1dVUblod2gVhM+7veuaUS0GxIWQriQ3Wqm4sAOSr/5CFtlCf5Afvpn9QepNWgMgWj8g+vCwy8IlVaP/UIoVNcFw8VXnXNQqYlInI02KNwl/Yj2Q8JCCBdQrBbKf9pJ6TcfYqsoxrdzH6Km/JkjWYX0jevyyyeE6vK6UHAEQzmW3JMoVgsavyA0/sHoImLR+Aeh8Qs+Hyjn/z7/6UStvfJEnUI0h4SFEK1IsVmoOPgFJXs/xFZehE9sL6Jum4tvl36oVCqUolR8OnR3d5lCOCVhIUQrUGxWKg7tpvTr/2AtK8SnYw8iJ87B0G2AnCAnPJKEhRAtSLHbqDz0JSVf/wdraT4+xuuIuOWPGLoPkpAQHk3CQogWVPTZaip+2oG+Q3eipz2D33XxEhLCK0hYCNFCqo7to+KnHQSPmExYwkwJCeFV5KQ8IVqAtbKUwi2r0HfoTtjNd0tQCK8jYSFEMymKQtHm11HMtUTdNheVRufukoRocRIWQjRTxU87qP5vKmEJ96KP7OTucoRoFRIWQjSDpTiXc5+/haHbAIKG3uLucoRoNRIWQjSRYrdR8OmrqDQaIic9ikol/5yE95L/u4VootJvPsaUfZyIxAdlLibh9SQshGgCU85/Kdnzf/j3vZGAvje6uxwhWp1LzrMoKSnh6aefJjMzE71eT5cuXVi8eDFhYWH1xtlsNpYuXcqePXtQqVQ8+OCDTJ061RUlCtFodouJgk//F41/CBETZru7HCFcwiWfLFQqFbNmzWLbtm2kpKTQqVMnli9fftm4lJQUMjMz2b59Ox988AErVqwgKyvLFSUK0WjFu97Fci6HqFsfRWMIcHc5QriES8IiJCSE4cOHO+4PGjSInJycy8Zt2bKFqVOnolarCQsLY+zYsWzdutUVJYp2SFEUFLvtmpapPnmA8v2fETRsEoZuA1qpMiHaHpdP92G321m/fj0JCQmXPZebm0tMTIzjvtFoJC8v75rWn5aW1uTaUlNTm7xsW+VtPbVEPypzNfqcNHyyfkJdVYI1rBOWiDgskXHY/cOhgbOvVeZqgva+iRIQQUZwbzJa6L+t7KO2z9t6ako/Lg+LJUuW4Ofnx7333tsq6+/Xrx8+Ptd+kfrU1FTi473rmsXe1lNz+lEUOzWnf6bip51UHfse7FZ8Yq7Hp9cwas78jOXYTji2E21wJIa4wfjFDcHQtR9qveH88goFH/+DKkstHe9d1GLXoJB91PZ5W0+X9mMymRr1JtulYZGcnExGRgarVq1Crb78CJjRaCQnJ4cBA+o+3l/6SUOIa2UtL6Li4C4qDu7CWlaI2hBA0NBEggaOQR/V2THOUlZAzcmfqD75I5WHvqLix+2g0WLo3AdD3GBQoOrIt4TefI9crEi0Sy4Li1deeYW0tDRWr16NXn/lyz4mJiayYcMGxo8fT2lpKTt27OC9995zVYnCSyg2C1Un9lNxYCc1p34CFAzdBhKWMBP/HsNQaS+fu0kXHIVuyHiChoxHsVqoPXuE6pMHqD75I8U73gbAJ7YXISMnu7gbIdoGl4TFiRMnWLVqFV27dmX69OkAxMbGsnLlSmbPns3cuXPp378/kydP5uDBg4wfPx6ARx55hE6dZK4d0Xi22iqy//U01pI8NIFhhNx4B4EDE9CFRDd6HSqtDkO3ARi6DSB87O+wlBVQm3EYQ/eBqNSaVqxeiLbLJWFx/fXXc+zYsSs+t2bNGsdtjUZDUlKSK0oSXqrky/VYSwuI+u2T+Pcc1iIv7rrgKHQDolqgOiE8l1z8SHgNU+4pylO3ERQ/gYDeI91djhBeRab7EF5BUewUbV2Nxi+I0F/f7e5yhPA6EhbCK1T8tBNTzgnCxsxE4+vv7nKE8DoSFsLj2arLKf5iLb6d+xDQ79fuLkcIryRhITxe8RfvYTfVEJE4W659LUQrkbAQHq02+zgVP+0geNhE9JGdnS8ghGgSCQvhsRS7jaLPVqMJDCP0pmnuLkcIryZhITxWeeo2zPmnCR/3e8ccTkKI1iFhITyStbKE4i/XY+g+EP9eI9xdjhBeT8JCeKTine+gWM1ETJglX2oL4QISFsLj1GSkU5n2FSEjpqALk1mJhXAFCQvhURSblaKtq9GGRBEy6rfuLkeIdkPCQniUsu83YSnKInz8H1Drrv0iV0KIppGJBIXHUNWWU7J3A37X34D/9UPdXY4Q7Yp8shAew+/IDlDshI//vbtLEaLdaXRYfPfdd5w9exaAgoIC5s2bxzPPPENhYWGrFScEgN1qpnRfCvr8o4TceCe6ELm2hBCu1uiwSEpKQqOpu5BMcnIyVqsVlUrFc88912rFifbNVl1ByZ4NnH3tIYp3vIUltBMhw29zd1lCtEuN/s4iPz+fmJgYrFYrX3/9Nbt27UKn03HTTTe1Zn2iHbKU5FH2/SYqDu5CsZgwxA0mZMRkDheZrnj9bCFE62t0WAQEBFBUVMSJEyeIi4vD398fs9mM1WptzfpEO1KbfYKyfRupOroPVGoC+v2KkOG3oo86P0HguVT3FihEO9bosLj33nu58847sVgsLFiwAIAff/yR7t27t1pxwvspip3qE6mU7fuU2szDqH38CB5xG8E3TEQbGObu8oQQ5zU6LB588EHGjRuHRqOhc+e6d3rR0dEsXbq01YoT3sturqHy0JeU/bAFy7lstEERhI97gMCBY1D7yKSAQrQ113SeRbdu3Ry3v/vuOzQaDTfccEOLFyW8l6U0n/L9W6n4aQd2UzX6DnFETf4z/r1HotLIaT9CtFXXdBjq8ccfJz4+ntWrV/PWW2+h0Wi45557eOihh1qzRuHhFEWhNjOdsu83U31iPwD+vUYQPGwiPh17ykSAQniARofFiRMnGDRoEAAbNmzg3Xffxc/Pj7vvvttpWCQnJ7Nt2zays7NJSUmhR48el41ZsWIF69atIyqq7jf0Q4YMYdGiRdfSi2hj7BYTlel7KP9hC+aCDNSGQEJGTiEoPhFtULi7yxNCXINGh4XdbkelUpGZmYmiKMTFxQFQVlbmdNkxY8Zw3333cc8991x13JQpU5g3b15jSxJtlK22irJvP6H8wOfYayrQR3UmYuLDBPS9SeZzEsJDNTos4uPjWbx4MYWFhYwbNw6AzMxMQkNDnS47dKjM49NeKIpCwccvU3PqIH49biB42ER8O/eVQ01CeDiVoihKYwaWlJTw73//G61Wyx/+8Af8/f3ZvXs3Z86c4f7772/UxhISEli1alWDh6E2bNhAcHAwkZGRPPbYYwwePLjRjZhMJtLS0ho9XrQO/dkD+Kd/RnXv8Zi6yJsEITxFv3798PG5yid/xYVGjx6tHDt27IrPFRQUKGazWVEURfn666+VESNGKMXFxY1ed21trbJ//36ltra2SbXt37+/Scu1Za7uyVySp5xaNkPJXrtIsdttLb5+2Udtn7f1oyje19Ol/TT2tbPRc0NZLBZeffVVxowZQ//+/RkzZgyvvvoqZrO5JUKNyMhIdLq6qRxGjRqF0WjkxIkTLbJu0foUxU5hykpAReSkOahUMqGxEN6k0d9Z/P3vf+fnn38mKSmJmJgYcnJyeP3116msrHSc0d0c+fn5REdHA3DkyBGys7Prndch2rby/Z9Rm5lOxMSH0QXLrLBCeJtGh8XWrVvZuHGj4wvt7t2706dPHyZPnuw0LJYuXcr27dspKirigQceICQkhM2bNzN79mzmzp1L//79efnll0lPT0etVqPT6Vi2bBmRkZHN6064hPlcDsW71mKIG0LgwDHuLkcI0QoaHRZKA9+DN/T4xRYuXMjChQsve3zNmjWO28nJyY0tRbQhit1GYcprqLR6Iic+LL96EsJLNfrAcmJiIg8//DB79uzh5MmTfPXVVzzyyCMkJia2Zn2ijSvbl4Ip+xjhE/4gE/8J4cUa/cniqaee4o033mDx4sUUFBQQHR3Nb37zG+bMmdOa9Yk2zFyQSfGX6/HrOZyAvnJdEyG82VXD4ttvv613f9iwYQwbNqzeY6mpqYwcObLlKxNtmmKzUpCyArWPH5G3/FEOPwnh5a4aFs8+++wVH7/wwqAoCiqVip07d7Z8ZaJNK937Eea8U0Tf8RQa/2B3lyOEaGVXDYtdu3a5qg7hQUy5pyjZ+x8C+t6Ef68R7i5HCOECcuaUuCaK1UJBygo0fkGET/iDu8sRQriIhIW4JiV7PsBSmEnkxIfRGALdXY4QwkUkLESj1WYfp/TbjQQOHIPfdfHuLkcI4UJyHUvhlGK1UJm+h+KvPkAbGEb4uPvdXZIQwsUkLESDbFVllP+4jfLUbdiqStFFdiZy4sOoffzcXZoQwsUkLMRlzIWZlH2/mcpDX6LYLBjiBhM8/FYMXQfI+RRCtFMSFgKoO2em5tRPlH2fQs2pg6i0egIGjCZ42ET0EbHuLk8I4WYSFu2cYrVQcWg3Zd9vwlKUhSYglNBf303QkPFo/ILcXZ4Qoo2QsGjHbDWV5G94idqzR9BHdyPytscI6DMKlUbn7tKEEG2MhEU7ZS0vInf9EiwleURO/hMBfW+S7yOEEA2SsGiHzAWZ5L6/BLu5FuP0hRi69nd3SUKINk7Cop2pyUwnf0MyKq2emJlL8Inu6u6ShBAeQMKiHak6+h0Fn/wTbUgkHe5+Tq6VLYRoNAmLdqJs/1bObXsTn47X02HaAjR+Mq+TEKLxJCy8nKIolHy5ntK9H+J3/VCibv8Lap2Pu8sSQngYCQtvZrdRtPl1Kg7uInDQWCJueRCVWuPuqoQQHkjCwkvZzbUEHPgPFYUnCblpGqE3TZOfxgohmswlU5QnJyeTkJBAz549OX78+BXH2Gw2kpKSGDt2LOPGjWPDhg2uKM0r2WoqyH3vBbSFp4i45Y+E/eouCQohRLO4JCzGjBnDe++9R8eOHRsck5KSQmZmJtu3b+eDDz5gxYoVZGVluaI8r2KtLCV37SJM+aepGjbTWV4AABJRSURBVPxbgoaMd3dJQggv4JKwGDp0KEaj8apjtmzZwtSpU1Gr1YSFhTF27Fi2bt3qivK8hrX8HLnvPoelJI8Ody3AEt3T3SUJIbxEm7lSXm5uLjExMY77RqORvLw8N1bkWSwleeS8sxBrVSnGu5/Dr9tAd5ckhPAiXvcFd1paWpOXTU1NbcFKXEddWUTgD+vAbqNy6HSKC6qhoK4XT+2pId7WD3hfT97WD3hfT03pp82EhdFoJCcnhwEDBgCXf9JorH79+uHjc+3nEaSmphIf73nXlTblnSZ3/WuotFqMM5agj+rseM5Te2qIt/UD3teTt/UD3tfTpf2YTKZGvcluM4ehEhMT2bBhA3a7neLiYnbs2MGECRPcXVabVpt9nNz3FtXN83Rf/aAQQoiW5JKwWLp0Kb/61a/Iy8vjgQceYOLEiQDMnj2bQ4cOATB58mRiY2MZP34806ZN45FHHqFTp06uKM8j1WSkkfteEmpDIDH3LUEXdu2fwoQQorFcchhq4cKFLFy48LLH16xZ47it0WhISkpyRTker/q/qeR/uBxtSBTGGS+gDQx1d0lCCC/XZr6zEI1TefRbCj7+J/qozhjvfk4ufSqEcAkJCw+gKArm/NNUHNxFeeo2fDpej/GuZ1H7+ru7NCFEOyFh0YbZqsqoSPuKyp+/wFyQgUqjI6Dfr4hInIVab3B3eUKIdkTCoo1RbBaqT/xIxc9fUH3yR7Db8Im5nojEB/HvMwqNIcDdJQoh2iEJizbClHeaip93UZm2B3tNBZqAUIKH30pg/5vRR8qvwoQQ7iVh4WamnP9StO1NTDknQKPFv8cwAgeMxtB9oFx7QgjRZkhYuIndXEvJVx9Q9v0mNP7BhE+YRUDfG9EY5HKnQoi2R8LCDapPHaTos1VYSwsIHDKe8NH3yi+bhBBtmoSFC9mqKzi34y0qD+1GFxaDceYSDJ37uLssIYRwSsLCBRRFoerw1xRt/xf22ipCRt1JyI13oNbq3V2aEEI0ioRFK7OWFVL42WpqTv6IT8z1RE58GH1UF3eXJYQQ10TCAjAXZOL/08eUmjLx7dofnw7dmv1LJMVqofzA5xTvfg8UCB/3AEFDb5FfOAkhPJKEBQAKmspCir9YC4DKxw9D5z74dumHoWt/9FGdUakanqDXbjVjzs/AnHcSU+4pTHmnMBdmgt2GofsgIm75I7qQKFc1I4QQLU7CAtBHdaH8xgcZ2LM7tRnp1GSkUXPmENUn9gOgNgRi6NIX3y79MXTpg91ciym3LhjMeScxF54FxX5+bAA+HeIIGXEbvp36YIgbjEqlcmd7QgjRbBIWF9EGhBLQ90YC+t4IgLW8iJozadRkpFF75hBVR7+rN17tF4RPh+6EXBePjzEOvbE72qBICQchhNeRsLgKbVAEgQNuJnDAzSiKgrU0n9rMw6h9/fExxqEJDJdgEEK0CxIWjaRSqdCFdkAX2sHdpQghhMu1mWtwCyGEaLskLIQQQjglYSGEEMIpCQshhBBOSVgIIYRwSsJCCCGEUy776ezp06eZP38+paWlhISEkJycTNeuXeuNWbFiBevWrSMqqm5qjCFDhrBo0SJXlSiEEKIBLguLRYsWMWPGDCZPnszGjRt5/vnneeeddy4bN2XKFObNm+eqsoQQQjSCSw5DnTt3jsOHDzNp0iQAJk2axOHDhykuLnbF5oUQQjSTS8IiNzeX6OhoNJq66bk1Gg1RUVHk5uZeNnbz5s3ceuut/P73v+fAgQOuKE8IIYQTbWq6j+nTp/PQQw+h0+nYu3cvc+bMYcuWLYSGhjZ6HWlpaU3efmpqapOXbau8rSdv6we8rydv6we8r6em9OOSsDAajeTn52Oz2dBoNNhsNgoKCjAajfXGRUZGOm6PGjUKo9HIiRMnGDZsWKO31a9fP3x8fK65xtTUVOLj4695ubbM23rytn7A+3rytn7A+3q6tB+TydSoN9kuOQwVHh5O79692bRpEwCbNm2id+/ehIWF1RuXn5/vuH3kyBGys7Pp1q2bK0oUQghxFS47DPXCCy8wf/58Xn/9dYKCgkhOTgZg9uzZzJ07l/79+/Pyyy+Tnp6OWq1Gp9OxbNmyep82hBBCuIfLwiIuLo4NGzZc9viaNWscty8EiBBCiLZFzuAWQgjhlISFEEIIpyQshBBCOCVhIYQQwikJCyGEEE5JWAghhHBKwkIIIYRTEhZCCCGckrAQQgjhlISFEEIIpyQshBBCOCVhIYQQwikJCyGEEE5JWAghhHBKwkIIIYRTEhZCCCGckrAQQgjhlISFEEIIpyQshBBCOCVhIYQQwikJCyGEEE5JWAghhHBKwkIIIYRTLguL06dPc9dddzFhwgTuuusuzpw5c9kYm81GUlISY8eOZdy4cWzYsMFV5QkhhLgKras2tGjRImbMmMHkyZPZuHEjzz//PO+88069MSkpKWRmZrJ9+3ZKS0uZMmUKI0eOJDY2tlVrO5ZRzMrNeQTu3o3BV4vBR4ufj9Zx2+Cjxe+i2zqtBp1WjUatQqtVo9Oo0WhUaDXqS/6oUKtVaDRqtGoVGo0KtVqNWgUqlapVexJCiJbkkrA4d+4chw8f5t///jcAkyZNYsmSJRQXFxMWFuYYt2XLFqZOnYparSYsLIyxY8eydetWZs2a1ar1BQf4EBOmR2/wpcZkpbi8lmyTlZrzf0xmW4tvU6NW1f1xBEjdfdX5IFGrVahVoFar6u6rVKjVoFap6j3/y23V+bF1Y9QqFRUV5Xya+q1jnReeu/i+SqVCxS9/owIVKlRqHI8DdWOpWwbq6uKS5dQXr+PS9dXbFr8se1FmOpa5aLlftg05ueUcP3fMcZ9L1n+l9aguFHFhPRct6Kjj4h1zSYarLn3gIgpKg89dWNZRq+qS2+frycioosiSccVaLq61/v36ty9d8tLn6venavi5a3Sl9zunzlRTrc5uxkobvy1XOZVZTY22GT25gF6rIb5XFBpN6x0scklY5ObmEh0djUajAUCj0RAVFUVubm69sMjNzSUmJsZx32g0kpeXd03bSktLa1KNt4+8UIfusufsdgWzVcFktWOyKNhsCjY72OwKdkXBZgObUveY3a5gs1/0vB3sioJdqXvusvsK2OygKAqKAnal7kVIUXD8sSv2i27XvUQpCijnl7M6lqu/HhTILyqpG3t+GS4se/F6qBv7y9/KL+Mufs6xDuWy9Z0fhnJ+ofrj66/70scu+qveMld0qPwqT3qo70vcXUHL+qbY3RW0vK/bfk8zR0cQZ/Rt1NjU1NRrXr/LDkO5Sr9+/fDx8bnm5VJTU4mPj2+FitzHW3pSzqfH/tRUhgyJd6TJL4FVP9yUi9JIuWQdF49z3MEx/JLt1q/hSocOG3rHeyGQf9neLyF68eM/H/qZ/v0GnN/+xRusX5NylQStV+elXVy+yno9NVVDi6anp9O3b98mrvPKK216lS3jcPph+vTt4+Yqrk6v1WCM8G/U2EtfF0wmU6PeZLskLIxGI/n5+dhsNjQaDTabjYKCAoxG42XjcnJyGDCg7h/PpZ80RPt04UX6wqG65h08aVuC/bREhhrcXUaLKcjS0Sk60N1ltKiibB1dOgS5uwy3c8mvocLDw+nduzebNm0CYNOmTfTu3bveISiAxMRENmzYgN1up7i4mB07djBhwgRXlCiEEOIqXPbT2RdeeIG1a9cyYcIE1q5dS1JSEgCzZ8/m0KFDAEyePJnY2FjGjx/PtGnTeOSRR+jUqZOrShRCCNEAl31nERcXd8XzJtasWeO4rdFoHCEihBCi7ZAzuIUQQjglYSGEEMIpCQshhBBOec15Fhd+o202m5u8DpPJ1FLltBne1pO39QPe15O39QPe19PF/Vx4zXR23o1Kac6ZOW1IRUUFx48fd3cZQgjhkXr06EFgYMPnyHhNWNjtdqqqqtDpdDJJnxBCNJKiKFgsFvz9/VGrG/5mwmvCQgghROuRL7iFEEI4JWEhhBDCKQkLIYQQTklYCCGEcErCQgghhFMSFkIIIZySsBBCCOGU10z30RynT59m/vz5lJaWEhISQnJyMl27dnV3WU2WkJCAXq93XF72ySef5KabbnJzVY2XnJzMtm3byM7OJiUlhR49egCevZ8a6slT91VJSQlPP/00mZmZ6PV6unTpwuLFiwkLC/PI/XS1fjx1HwHMmTOHrKws1Go1fn5+PPfcc/Tu3btp+0gRysyZM5VPPvlEURRF+eSTT5SZM2e6uaLmGT16tHLs2DF3l9FkP/zwg5KTk3NZH568nxrqyVP3VUlJifLdd9857r/00kvKM888oyiKZ+6nq/XjqftIURSlvLzccfvzzz9XpkyZoihK0/ZRuz8Mde7cOQ4fPsykSZMAmDRpEocPH6a4uNjNlbVfQ4cOvez67J6+n67UkycLCQlh+PDhjvuDBg0iJyfHY/dTQ/14uovneqqsrESlUjV5H7X7w1C5ublER0ej0WiAuqv1RUVFkZube9k1wj3Jk08+iaIoxMfH85e//IWgIM++4Ly37ifw/H1lt9tZv349CQkJXrGfLu7nAk/eR88++yx79+5FURTefPPNJu+jdv/Jwhu99957fPrpp3z44YcoisLixYvdXZJogDfsqyVLluDn58e9997r7lJaxKX9ePo+evHFF9m9ezePP/44y5Yta/J62n1YGI1G8vPzsdlsANhsNgoKCjz6kMGF2vV6PTNmzODHH390c0XN5437CTx/XyUnJ5ORkcE///lP1Gq1x++nS/sBz99HF0yZMoV9+/bRoUOHJu2jdh8W4eHh9O7dm02bNgGwadMmevfu7TEfmS9VXV1NRUUFUDf18JYtW+jdu7ebq2o+b9tP4Pn76pVXXiEtLY2VK1ei1+sBz95PV+rHk/dRVVUVubm5jvu7du0iODi4yftIpigHTp48yfz58ykvLycoKIjk5GS6d+/u7rKa5OzZszz22GPYbDbsdjtxcXEsXLiQqKgod5fWaEuXLmX79u0UFRURGhpKSEgImzdv9uj9dKWeVq1a5bH76sSJE0yaNImuXbvi6+sLQGxsLCtXrvTI/dRQP/Pnz/fYfVRUVMScOXOoqalBrVYTHBzMvHnz6Nu3b5P2kYSFEEIIp9r9YSghhBDOSVgIIYRwSsJCCCGEUxIWQgghnJKwEEII4ZSEhRBtTFZWFj179sRqtbq7FCEcJCyEEEI4JWEhhBDCKQkLIRohPz+fxx57jBEjRpCQkMA777wDwIoVK5g7dy5//vOfGTx4MLfffjtHjx51LHfy5ElmzpzJ0KFDmThxIjt37nQ8V1tby0svvcTo0aOJj4/n7rvvpra21vF8SkoKN998M8OHD+eNN95wXbNCXIGEhRBO2O12Hn74YXr27MlXX33F22+/zdtvv82ePXsA2LlzJ4mJiXz//fdMmjSJOXPmYLFYsFgsPPTQQ4waNYpvvvmGhQsX8uSTT3Lq1CmgbtK69PR03n//fb7//nueeuopx+R1AKmpqWzdupW3337bMY2GEO4iYSGEE4cOHaK4uJhHH30UvV5Pp06dmDZtGlu2bAGgb9++JCYmotPpeOCBBzCbzRw8eJCDBw9SXV3Ngw8+iF6vZ+TIkYwePZrNmzdjt9v58MMPefbZZx3XFhgyZIhjAjuARx99FF9fX3r16kWvXr3qfWIRwtXa/cWPhHAmOzubgoIChg4d6njMZrMxdOhQYmJi6NChg+NxtVpNdHQ0BQUFAHTo0KHep4WYmBjy8/MpKSnBZDLRqVOnBrcbERHhuG0wGKiurm7JtoS4JhIWQjhhNBqJjY1l+/btlz23YsUK8vLyHPftdjv5+fmOWUnz8vKw2+2OwMjNzaVr166Ehobi4+PD2bNn6dWrl2saEaIZ5DCUEE4MGDCAgIAAVq9eTW1tLTabjePHj/Pzzz8DkJ6ezvbt27Farbz99tvo9XoGDhzIgAEDMBgMvPnmm1gsFvbt28euXbv4zW9+g1qt5o477uBvf/ub40I0Bw4cwGw2u7lbIa5MwkIIJzQaDW+88QZHjx5lzJgxjBgxgoULF1JZWQnAmDFj2LJlCzfccAMbN25kxYoV6HQ69Ho9b7zxBl999RUjRowgKSmJZcuWERcXB8C8efPo0aMHd955J8OGDWP58uXY7XZ3tipEg+R6FkI0w4oVK8jIyGD58uXuLkWIViWfLIQQQjglYSGEEMIpOQwlhBDCKflkIYQQwikJCyGEEE5JWAghhHBKwkIIIYRTEhZCCCGckrAQQgjh1P8HnKgT9j3jC5wAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "train_loss, val_loss = get_loss(\"model_save_mmtrBig/mmtrBig_model_run/logfile.log\")\n",
    "\n",
    "plt.subplot(111)\n",
    "plt.plot(train_loss, label='train')\n",
    "plt.plot(val_loss, label='val')\n",
    "plt.xlabel('epoch')\n",
    "plt.ylabel('loss')\n",
    "plt.legend(loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Graph loss vs epoch ViLBERT"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "train_loss, val_loss = get_loss(\"model_save_vilbert/vilbert_model_run/logfile.txt\")\n",
    "\n",
    "plt.subplot(111)\n",
    "plt.plot(train_loss, label='train')\n",
    "plt.plot(val_loss, label='val')\n",
    "plt.xlabel('epoch')\n",
    "plt.ylabel('loss')\n",
    "plt.legend(loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Validation accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 136,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_f1(path):\n",
    "    \n",
    "    with open(path, \"r\") as f:\n",
    "        log = f.readlines()\n",
    "        \n",
    "    start_index = [i for i, s in enumerate(log) if \"Training..\" in s][0]\n",
    "    log_metrics = log[start_index+1:]\n",
    "    \n",
    "    micro = [s.split(\"-\")[-1].split(\"|\")[2][11:].strip().strip(\"\\n\") for s in log_metrics if \"Val\" in s]\n",
    "    micro = np.array([float(v) for v in micro])\n",
    "    \n",
    "    macro = [s.split(\"-\")[-1].split(\"|\")[1][9:].strip().strip(\"\\n\") for s in log_metrics if \"Val\" in s]\n",
    "    macro = np.array([float(v) for v in macro])\n",
    "    \n",
    "    return micro, macro"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Validation F1 scores MMBT"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 153,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "micro_test, macro_test = get_f1(\"model_save_mmbt/mmbt_model_run/logfile.log\")\n",
    "\n",
    "plt.subplot(111)\n",
    "plt.plot(micro_test, label='micro f1')\n",
    "plt.plot(macro_test, label='macro f1')\n",
    "plt.xlabel('epoch')\n",
    "plt.ylabel('F1 score')\n",
    "plt.legend(loc='upper left')\n",
    "plt.title(\"Validation F1 score\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Validation F1 scores MMTRBig"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 154,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "micro_test, macro_test = get_f1(\"model_save_mmtrBig/mmtrBig_model_run/logfile.log\")\n",
    "\n",
    "plt.subplot(111)\n",
    "plt.plot(micro_test, label='micro f1')\n",
    "plt.plot(macro_test, label='macro f1')\n",
    "plt.xlabel('epoch')\n",
    "plt.ylabel('F1 score')\n",
    "plt.legend(loc='upper left')\n",
    "plt.title(\"Validation F1 score\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Validation F1 scores ViLBERT"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 152,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "micro_test, macro_test = get_f1(\"model_save_vilbert/vilbert_model_run/logfile.txt\")\n",
    "\n",
    "plt.subplot(111)\n",
    "plt.plot(micro_test, label='micro f1')\n",
    "plt.plot(macro_test, label='macro f1')\n",
    "plt.xlabel('epoch')\n",
    "plt.ylabel('F1 score')\n",
    "plt.legend(loc='upper left')\n",
    "plt.title(\"Validation F1 score\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Visualize raw predicitons"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "raw_preds = np.load('model_save_mmbt/mmbt_model_run/preds_raw.npy')\n",
    "max_pred_indexes = np.argmax(raw_preds, axis=1)\n",
    "max_pred_category = [label_names[i] for i in max_pred_indexes]\n",
    "gold_pred_max = gold_preds[:, :]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "pred_max = np.zeros(raw_preds.shape)\n",
    "\n",
    "for i in range(pred_max.shape[0]):\n",
    "    pred_max[i][max_pred_indexes[i]] = 1\n",
    "    \n",
    "    for j in range(pred_max.shape[1]):\n",
    "        if j != max_pred_indexes[i]:\n",
    "            gold_pred_max[i][j] = 0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "from sklearn.metrics import f1_score\n",
    "\n",
    "def freqs_dataframe_max(data_preds, model_name, label_names):\n",
    "    countsLabel = []\n",
    "    \n",
    "    scores = f1_score(gold_pred_max, data_preds, average=None)\n",
    "        \n",
    "    df = pd.DataFrame()\n",
    "    df['label'] = label_names\n",
    "    df['freq'] = list(scores)\n",
    "    df['model'] = [model_name]*len(label_names)\n",
    "            \n",
    "    return df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "mmbt_freqs_max = freqs_dataframe_max(pred_max, \"mmbt\", label_names)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x720 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "sns.set(style=\"whitegrid\")\n",
    "\n",
    "f, ax = plt.subplots(figsize=(6, 10))\n",
    "ax_points = sns.stripplot(x=\"freq\", y=\"label\", data=mmbt_freqs_max, hue=\"model\", jitter=0, order=ordered_classes)\n",
    "ax.legend(ncol=2, loc=\"upper left\", frameon=True, bbox_to_anchor=(0.99, 1.01))\n",
    "ax.set(xlabel=\"F1 score\", ylabel=\"genre\")\n",
    "\n",
    "for i in range(len(label_names)):\n",
    "    ax_points.axhline(i, color='black', alpha=0.1, lw = 2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Complementarity"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Compare argmax classes between models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "raw_preds_img = np.load('model_save_mmtrBig/mmtrImg_model_run/preds_raw.npy')\n",
    "raw_preds_txt = np.load('model_save_mmtrBig/mmtrTxt_model_run/preds_raw.npy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "max_pred_indexes_img = np.argmax(raw_preds_img, axis=1)\n",
    "max_pred_indexes_txt = np.argmax(raw_preds_txt, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "diff = np.sum(max_pred_indexes_img != max_pred_indexes_txt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.2687524041543788\n"
     ]
    }
   ],
   "source": [
    "print(diff/max_pred_indexes_img.shape[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Maximum Possible Accuracy"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Combining MMTr Text and Image models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "mpa_pred = np.maximum(raw_preds_img, raw_preds_txt)\n",
    "mpa_pred = np.around(mpa_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.65456729737847"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "f1_score(gold_preds, mpa_pred, average='micro')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.6038054838795545"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "f1_score(gold_preds, mpa_pred, average='macro')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Original model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "mmtr_preds = load_labels(\"model_save_mmtrBig/mmtrBig_model_run/test_labels_pred.txt\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.6642833309499977"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "f1_score(gold_preds, mmtr_preds, average='micro')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.6115393540081885"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "f1_score(gold_preds, mmtr_preds, average='macro')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Coincident Falure Diversity"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "txt_preds = load_labels(\"model_save_mmtrBig/mmtrTxt_model_run/test_labels_pred.txt\")\n",
    "img_preds = load_labels(\"model_save_mmtrBig/mmtrImg_model_run/test_labels_pred.txt\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1296"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "txt_gold = load_labels(\"model_save_mmtrBig/mmtrTxt_model_run/test_labels_gold.txt\")\n",
    "np.sum(np.all(np.equal(txt_preds, txt_gold), axis=1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1291"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "img_gold = load_labels(\"model_save_mmtrBig/mmtrImg_model_run/test_labels_gold.txt\")\n",
    "np.sum(np.all(np.equal(img_preds, img_gold), axis=1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "txt_bool = np.all(np.equal(txt_preds, txt_gold), axis=1)\n",
    "img_bool = np.all(np.equal(img_preds, img_gold), axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [],
   "source": [
    "p0 = np.logical_and(txt_bool, img_bool)\n",
    "p1 = np.logical_xor(txt_bool, img_bool)\n",
    "p2 = ~np.logical_or(p0, p1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {},
   "outputs": [],
   "source": [
    "p0 = np.sum(p0)/p0.shape[0]\n",
    "p1 = np.sum(p1)/p1.shape[0]\n",
    "p2 = np.sum(p2)/p2.shape[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {},
   "outputs": [],
   "source": [
    "cdf = (1./1.-p0) * (p1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 90,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.1322353836250685"
      ]
     },
     "execution_count": 90,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "cdf"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.metrics import f1_score\n",
    "from sklearn.metrics import average_precision_score, roc_auc_score, precision_recall_curve"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def load_labels(path):\n",
    "    with open(path, \"r\") as f:\n",
    "        all_preds = f.readlines()\n",
    "    preds_processed = [[int(e) for e in pred.strip('\\n').split()] for pred in all_preds]\n",
    "    return np.array(preds_processed)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 256,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Gold shape: (987, 13)\n"
     ]
    }
   ],
   "source": [
    "gold_preds = load_labels(\"model_save_mmbtAdapter/HoulsbyMoviescopeMetaSeed4_model_run/test_labels_gold.txt\")\n",
    "print(f\"Gold shape: {gold_preds.shape}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 257,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model shape: (987, 13)\n"
     ]
    }
   ],
   "source": [
    "model_preds = load_labels(\"model_save_mmbtAdapter/HoulsbyMoviescopeMetaSeed4_model_run/test_labels_pred.txt\")\n",
    "print(f\"Model shape: {model_preds.shape}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 258,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Raw preds shape: (987, 13)\n"
     ]
    }
   ],
   "source": [
    "model_preds_raw = np.load('model_save_mmbtAdapter/HoulsbyMoviescopeMetaSeed4_model_run/preds_raw.npy')\n",
    "print(f\"Raw preds shape: {model_preds_raw.shape}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 259,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AUC PR micro: 0.74663\n"
     ]
    }
   ],
   "source": [
    "auc_pr_score = average_precision_score(gold_preds, model_preds_raw, average='micro')\n",
    "print(f'AUC PR micro: {auc_pr_score:.5f}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 260,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AUC PR macro: 0.72010\n"
     ]
    }
   ],
   "source": [
    "auc_pr_score = average_precision_score(gold_preds, model_preds_raw, average='macro')\n",
    "print(f'AUC PR macro: {auc_pr_score:.5f}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 261,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "AUC PR samples: 0.82801\n"
     ]
    }
   ],
   "source": [
    "auc_pr_score = average_precision_score(gold_preds, model_preds_raw, average='samples')\n",
    "print(f'AUC PR samples: {auc_pr_score:.5f}')"
   ]
  }
 ],
 "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": "3.6.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}