notebooks/evaluate_distribution_of_similarities.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "\n",
    "os.chdir('..')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[nltk_data] Downloading package wordnet to /Users/alishri/nltk_data...\n",
      "[nltk_data]   Package wordnet is already up-to-date!\n"
     ]
    }
   ],
   "source": [
    "import pickle\n",
    "import itertools\n",
    "import glob\n",
    "import json\n",
    "from PIL import Image\n",
    "\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "from tqdm import tqdm\n",
    "from matplotlib import pyplot as plt\n",
    "\n",
    "import configs\n",
    "from utils import laion_utils as laionu\n",
    "from utils.ilsvrc_utils import load_lemmas_and_wnids\n",
    "from core.queries import QueryType"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Settings"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Path"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# LAION\n",
    "laion_path = 'laion400m'\n",
    "laion_until_part = 31\n",
    "\n",
    "sm_file_name = configs.LAIONConfig.SUBSET_SM_FILTERED_PREFIX \\\n",
    "             + configs.LAIONConfig.PREDICTED_PREFIX \\\n",
    "             + laionu.get_laion_subset_file_name(0, laion_until_part)\n",
    "\n",
    "ic_ms_file_name = configs.LAIONConfig.SUBSET_IC_MOST_SIMILAR_TXT_TXT_PREFIX \\\n",
    "                + laionu.get_laion_subset_file_name(0, laion_until_part)\n",
    "\n",
    "# ILSVRC\n",
    "ilsvrc_path = 'ilsvrc2012'\n",
    "\n",
    "ic_file_name = 'imagenet_captions.parquet'\n",
    "ival_file_name = 'ILSVRC2012_val.parquet'\n",
    "\n",
    "# Labels (maps)\n",
    "laion_labels_path = os.path.join('laion400m', 'processed', 'ilsvrc_labels')\n",
    "\n",
    "sm_wnid2laionindices_file_name = 'wnid2laionindices(substring_matched).pkl'\n",
    "\n",
    "ilsvrc_labels_path = os.path.join('ilsvrc2012', 'processed', 'labels')\n",
    "\n",
    "ic_imagename2wnid_file_name = 'icimagename2wnid.pkl'\n",
    "ival_imagename2wnid_file_name = 'imagename2wnid.pkl'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Column names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "chosen_q_type = QueryType.NAME_DEF\n",
    "\n",
    "# Image-query similarity\n",
    "image_query_sim_col = f'image_to_{chosen_q_type}_wnid_similarity'\n",
    "\n",
    "# Text-query similarity\n",
    "text_query_sim_col = f'text_to_{chosen_q_type}_wnid_similarity'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Loading"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load SM LAION subset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(879221, 88)\n"
     ]
    },
    {
     "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>SAMPLE_ID</th>\n",
       "      <th>URL</th>\n",
       "      <th>TEXT</th>\n",
       "      <th>HEIGHT</th>\n",
       "      <th>WIDTH</th>\n",
       "      <th>LICENSE</th>\n",
       "      <th>NSFW</th>\n",
       "      <th>similarity</th>\n",
       "      <th>a_photo_of_name_def_wnid</th>\n",
       "      <th>text_to_a_photo_of_name_def_wnid_similarity</th>\n",
       "      <th>...</th>\n",
       "      <th>top_5_is_correct_beit-large-patch16-224-pt22k-ft22k</th>\n",
       "      <th>top_1_is_correct_beit-base-patch16-224</th>\n",
       "      <th>top_5_is_correct_beit-base-patch16-224</th>\n",
       "      <th>top_1_is_correct_beit-base-patch16-384</th>\n",
       "      <th>top_5_is_correct_beit-base-patch16-384</th>\n",
       "      <th>top_1_is_correct_beit-large-patch16-224</th>\n",
       "      <th>top_5_is_correct_beit-large-patch16-224</th>\n",
       "      <th>top_1_is_correct_beit-large-patch16-384</th>\n",
       "      <th>top_5_is_correct_beit-large-patch16-384</th>\n",
       "      <th>image_to_text_similarity</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>1200056513</th>\n",
       "      <td>4.745710e+11</td>\n",
       "      <td>https://t1.ftcdn.net/jpg/00/75/08/12/240_F_750...</td>\n",
       "      <td>Colorful patchwork quilt pattern</td>\n",
       "      <td>373.0</td>\n",
       "      <td>240.0</td>\n",
       "      <td>?</td>\n",
       "      <td>UNLIKELY</td>\n",
       "      <td>0.328658</td>\n",
       "      <td>a photo of quilt which is bedding made of two ...</td>\n",
       "      <td>0.805882</td>\n",
       "      <td>...</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>0.325056</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1200090089</th>\n",
       "      <td>1.129810e+11</td>\n",
       "      <td>https://farm6.staticflickr.com/5132/5411062185...</td>\n",
       "      <td>basket quilt top</td>\n",
       "      <td>320.0</td>\n",
       "      <td>285.0</td>\n",
       "      <td>?</td>\n",
       "      <td>UNLIKELY</td>\n",
       "      <td>0.310124</td>\n",
       "      <td>a photo of quilt which is bedding made of two ...</td>\n",
       "      <td>0.823052</td>\n",
       "      <td>...</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>True</td>\n",
       "      <td>0.314476</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1200091068</th>\n",
       "      <td>4.145784e+12</td>\n",
       "      <td>https://theplaidporticodotcom4.files.wordpress...</td>\n",
       "      <td>QuiltCon charity quilt</td>\n",
       "      <td>584.0</td>\n",
       "      <td>820.0</td>\n",
       "      <td>?</td>\n",
       "      <td>UNLIKELY</td>\n",
       "      <td>0.307640</td>\n",
       "      <td>a photo of quilt which is bedding made of two ...</td>\n",
       "      <td>0.846488</td>\n",
       "      <td>...</td>\n",
       "      <td>None</td>\n",
       "      <td>None</td>\n",
       "      <td>None</td>\n",
       "      <td>None</td>\n",
       "      <td>None</td>\n",
       "      <td>None</td>\n",
       "      <td>None</td>\n",
       "      <td>None</td>\n",
       "      <td>None</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>3 rows × 88 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "               SAMPLE_ID                                                URL  \\\n",
       "1200056513  4.745710e+11  https://t1.ftcdn.net/jpg/00/75/08/12/240_F_750...   \n",
       "1200090089  1.129810e+11  https://farm6.staticflickr.com/5132/5411062185...   \n",
       "1200091068  4.145784e+12  https://theplaidporticodotcom4.files.wordpress...   \n",
       "\n",
       "                                        TEXT  HEIGHT  WIDTH LICENSE      NSFW  \\\n",
       "1200056513  Colorful patchwork quilt pattern   373.0  240.0       ?  UNLIKELY   \n",
       "1200090089                  basket quilt top   320.0  285.0       ?  UNLIKELY   \n",
       "1200091068            QuiltCon charity quilt   584.0  820.0       ?  UNLIKELY   \n",
       "\n",
       "            similarity                           a_photo_of_name_def_wnid  \\\n",
       "1200056513    0.328658  a photo of quilt which is bedding made of two ...   \n",
       "1200090089    0.310124  a photo of quilt which is bedding made of two ...   \n",
       "1200091068    0.307640  a photo of quilt which is bedding made of two ...   \n",
       "\n",
       "            text_to_a_photo_of_name_def_wnid_similarity  ...  \\\n",
       "1200056513                                     0.805882  ...   \n",
       "1200090089                                     0.823052  ...   \n",
       "1200091068                                     0.846488  ...   \n",
       "\n",
       "           top_5_is_correct_beit-large-patch16-224-pt22k-ft22k  \\\n",
       "1200056513                                               True    \n",
       "1200090089                                               True    \n",
       "1200091068                                               None    \n",
       "\n",
       "            top_1_is_correct_beit-base-patch16-224  \\\n",
       "1200056513                                    True   \n",
       "1200090089                                    True   \n",
       "1200091068                                    None   \n",
       "\n",
       "           top_5_is_correct_beit-base-patch16-224  \\\n",
       "1200056513                                   True   \n",
       "1200090089                                   True   \n",
       "1200091068                                   None   \n",
       "\n",
       "            top_1_is_correct_beit-base-patch16-384  \\\n",
       "1200056513                                    True   \n",
       "1200090089                                    True   \n",
       "1200091068                                    None   \n",
       "\n",
       "            top_5_is_correct_beit-base-patch16-384  \\\n",
       "1200056513                                    True   \n",
       "1200090089                                    True   \n",
       "1200091068                                    None   \n",
       "\n",
       "           top_1_is_correct_beit-large-patch16-224  \\\n",
       "1200056513                                    True   \n",
       "1200090089                                    True   \n",
       "1200091068                                    None   \n",
       "\n",
       "           top_5_is_correct_beit-large-patch16-224  \\\n",
       "1200056513                                    True   \n",
       "1200090089                                    True   \n",
       "1200091068                                    None   \n",
       "\n",
       "           top_1_is_correct_beit-large-patch16-384  \\\n",
       "1200056513                                    True   \n",
       "1200090089                                    True   \n",
       "1200091068                                    None   \n",
       "\n",
       "           top_5_is_correct_beit-large-patch16-384 image_to_text_similarity  \n",
       "1200056513                                    True                 0.325056  \n",
       "1200090089                                    True                 0.314476  \n",
       "1200091068                                    None                      NaN  \n",
       "\n",
       "[3 rows x 88 columns]"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sm_df = pd.read_parquet(\n",
    "    os.path.join(laion_path, sm_file_name)\n",
    ")\n",
    "\n",
    "print(sm_df.shape)\n",
    "\n",
    "sm_df.head(3)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load IC most similar subset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(161903, 11)\n"
     ]
    },
    {
     "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>SAMPLE_ID</th>\n",
       "      <th>URL</th>\n",
       "      <th>TEXT</th>\n",
       "      <th>HEIGHT</th>\n",
       "      <th>WIDTH</th>\n",
       "      <th>LICENSE</th>\n",
       "      <th>NSFW</th>\n",
       "      <th>similarity</th>\n",
       "      <th>image_to_text_similarity</th>\n",
       "      <th>name_def_wnid</th>\n",
       "      <th>text_to_name_def_wnid_similarity</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>1271</th>\n",
       "      <td>1.883650e+11</td>\n",
       "      <td>https://thumbs.dreamstime.com/t/calton-cemeter...</td>\n",
       "      <td>Calton cemetery and obelisk Edinburgh Stock Ph...</td>\n",
       "      <td>241.0</td>\n",
       "      <td>160.0</td>\n",
       "      <td>?</td>\n",
       "      <td>UNLIKELY</td>\n",
       "      <td>0.322448</td>\n",
       "      <td>0.310787</td>\n",
       "      <td>obelisk which is a stone pillar having a recta...</td>\n",
       "      <td>0.665240</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2891</th>\n",
       "      <td>3.478983e+12</td>\n",
       "      <td>https://ssl.c.photoshelter.com/img-get2/I0000P...</td>\n",
       "      <td>LHS basketball Laconia versus Kennett at Lacon...</td>\n",
       "      <td>200.0</td>\n",
       "      <td>135.0</td>\n",
       "      <td>?</td>\n",
       "      <td>UNLIKELY</td>\n",
       "      <td>0.309529</td>\n",
       "      <td>0.318361</td>\n",
       "      <td>basketball which is an inflated ball used in p...</td>\n",
       "      <td>0.379896</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2901</th>\n",
       "      <td>9.773084e+09</td>\n",
       "      <td>http://cache-graphicslib.viator.com/graphicsli...</td>\n",
       "      <td>Photo of Sydney Sydney Harbour Highlights Crui...</td>\n",
       "      <td>260.0</td>\n",
       "      <td>195.0</td>\n",
       "      <td>?</td>\n",
       "      <td>UNLIKELY</td>\n",
       "      <td>0.334991</td>\n",
       "      <td>0.334656</td>\n",
       "      <td>None</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         SAMPLE_ID                                                URL  \\\n",
       "1271  1.883650e+11  https://thumbs.dreamstime.com/t/calton-cemeter...   \n",
       "2891  3.478983e+12  https://ssl.c.photoshelter.com/img-get2/I0000P...   \n",
       "2901  9.773084e+09  http://cache-graphicslib.viator.com/graphicsli...   \n",
       "\n",
       "                                                   TEXT  HEIGHT  WIDTH  \\\n",
       "1271  Calton cemetery and obelisk Edinburgh Stock Ph...   241.0  160.0   \n",
       "2891  LHS basketball Laconia versus Kennett at Lacon...   200.0  135.0   \n",
       "2901  Photo of Sydney Sydney Harbour Highlights Crui...   260.0  195.0   \n",
       "\n",
       "     LICENSE      NSFW  similarity  image_to_text_similarity  \\\n",
       "1271       ?  UNLIKELY    0.322448                  0.310787   \n",
       "2891       ?  UNLIKELY    0.309529                  0.318361   \n",
       "2901       ?  UNLIKELY    0.334991                  0.334656   \n",
       "\n",
       "                                          name_def_wnid  \\\n",
       "1271  obelisk which is a stone pillar having a recta...   \n",
       "2891  basketball which is an inflated ball used in p...   \n",
       "2901                                               None   \n",
       "\n",
       "      text_to_name_def_wnid_similarity  \n",
       "1271                          0.665240  \n",
       "2891                          0.379896  \n",
       "2901                               NaN  "
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ic_ms_df = pd.read_parquet(\n",
    "    os.path.join(laion_path, ic_ms_file_name)\n",
    ")\n",
    "\n",
    "print(ic_ms_df.shape)\n",
    "\n",
    "ic_ms_df.head(3)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load ILSVRC val set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(50000, 2)\n"
     ]
    },
    {
     "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>image_to_name_def_wnid_similarity</th>\n",
       "      <th>name_def_wnid</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>ILSVRC2012_val_00049998.JPEG</th>\n",
       "      <td>0.305767</td>\n",
       "      <td>border collie which is developed in the area b...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ILSVRC2012_val_00049999.JPEG</th>\n",
       "      <td>0.251934</td>\n",
       "      <td>groom which is a man participant in his own ma...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ILSVRC2012_val_00050000.JPEG</th>\n",
       "      <td>0.303315</td>\n",
       "      <td>llama which is wild or domesticated South Amer...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                              image_to_name_def_wnid_similarity  \\\n",
       "ILSVRC2012_val_00049998.JPEG                           0.305767   \n",
       "ILSVRC2012_val_00049999.JPEG                           0.251934   \n",
       "ILSVRC2012_val_00050000.JPEG                           0.303315   \n",
       "\n",
       "                                                                  name_def_wnid  \n",
       "ILSVRC2012_val_00049998.JPEG  border collie which is developed in the area b...  \n",
       "ILSVRC2012_val_00049999.JPEG  groom which is a man participant in his own ma...  \n",
       "ILSVRC2012_val_00050000.JPEG  llama which is wild or domesticated South Amer...  "
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ival_df = pd.read_parquet(os.path.join(ilsvrc_path, ival_file_name))\n",
    "\n",
    "print(ival_df.shape)\n",
    "\n",
    "ival_df.tail(3)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load ImageNet-Captions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(443798, 5)\n"
     ]
    },
    {
     "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>TEXT</th>\n",
       "      <th>image_to_name_def_wnid_similarity</th>\n",
       "      <th>name_def_wnid</th>\n",
       "      <th>image_to_text_similarity</th>\n",
       "      <th>text_to_name_def_wnid_similarity</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ic_index</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>n15075141_9819.JPEG</th>\n",
       "      <td>Margaret Thatcher toilet paper Margaret Thatch...</td>\n",
       "      <td>0.255074</td>\n",
       "      <td>toilet tissue which is a soft thin absorbent p...</td>\n",
       "      <td>0.321070</td>\n",
       "      <td>0.672603</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>n15075141_9942.JPEG</th>\n",
       "      <td>Toilet Roll II toilet roll II toilet roll duan...</td>\n",
       "      <td>0.322143</td>\n",
       "      <td>toilet tissue which is a soft thin absorbent p...</td>\n",
       "      <td>0.284109</td>\n",
       "      <td>0.721321</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>n15075141_9993.JPEG</th>\n",
       "      <td>Toilet Paper toilet paper flush light green wa...</td>\n",
       "      <td>0.271322</td>\n",
       "      <td>toilet tissue which is a soft thin absorbent p...</td>\n",
       "      <td>0.294822</td>\n",
       "      <td>0.380635</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                                  TEXT  \\\n",
       "ic_index                                                                 \n",
       "n15075141_9819.JPEG  Margaret Thatcher toilet paper Margaret Thatch...   \n",
       "n15075141_9942.JPEG  Toilet Roll II toilet roll II toilet roll duan...   \n",
       "n15075141_9993.JPEG  Toilet Paper toilet paper flush light green wa...   \n",
       "\n",
       "                     image_to_name_def_wnid_similarity  \\\n",
       "ic_index                                                 \n",
       "n15075141_9819.JPEG                           0.255074   \n",
       "n15075141_9942.JPEG                           0.322143   \n",
       "n15075141_9993.JPEG                           0.271322   \n",
       "\n",
       "                                                         name_def_wnid  \\\n",
       "ic_index                                                                 \n",
       "n15075141_9819.JPEG  toilet tissue which is a soft thin absorbent p...   \n",
       "n15075141_9942.JPEG  toilet tissue which is a soft thin absorbent p...   \n",
       "n15075141_9993.JPEG  toilet tissue which is a soft thin absorbent p...   \n",
       "\n",
       "                     image_to_text_similarity  \\\n",
       "ic_index                                        \n",
       "n15075141_9819.JPEG                  0.321070   \n",
       "n15075141_9942.JPEG                  0.284109   \n",
       "n15075141_9993.JPEG                  0.294822   \n",
       "\n",
       "                     text_to_name_def_wnid_similarity  \n",
       "ic_index                                               \n",
       "n15075141_9819.JPEG                          0.672603  \n",
       "n15075141_9942.JPEG                          0.721321  \n",
       "n15075141_9993.JPEG                          0.380635  "
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ic_df = pd.read_parquet(os.path.join(ilsvrc_path, ic_file_name))\n",
    "\n",
    "print(ic_df.shape)\n",
    "\n",
    "ic_df.tail(3)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load maps"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(os.path.join(laion_labels_path, sm_wnid2laionindices_file_name), 'rb') as f:\n",
    "    sm_wnid2laionindices = pickle.load(f)\n",
    "    \n",
    "# Drop empty indices\n",
    "sm_wnid2laionindices = {wnid: indices for wnid, indices in sm_wnid2laionindices.items() if len(indices) > 0}\n",
    "\n",
    "# Find the inverse map\n",
    "sm_laionindex2wnids = {}\n",
    "\n",
    "for wnid, laion_indices in sm_wnid2laionindices.items():\n",
    "    for laion_idx in laion_indices:\n",
    "        if laion_idx not in sm_laionindex2wnids:\n",
    "            sm_laionindex2wnids[laion_idx] = []\n",
    "        sm_laionindex2wnids[laion_idx].append(wnid)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(os.path.join(ilsvrc_labels_path, ival_imagename2wnid_file_name), 'rb') as f:\n",
    "    ival_imagename2wnid = pickle.load(f)\n",
    "    \n",
    "# Find the inverse map\n",
    "ival_wnid2imagenames = {}\n",
    "for image_name, wnid in ival_imagename2wnid.items():    \n",
    "    if wnid not in ival_wnid2imagenames:\n",
    "        ival_wnid2imagenames[wnid] = []\n",
    "    ival_wnid2imagenames[wnid].append(image_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(os.path.join(ilsvrc_labels_path, ic_imagename2wnid_file_name), 'rb') as f:\n",
    "    ic_imagename2wnid = pickle.load(f)\n",
    "    \n",
    "# Find the inverse map\n",
    "ic_wnid2imagenames = {}\n",
    "for image_name, wnid in ic_imagename2wnid.items():    \n",
    "    if wnid not in ic_wnid2imagenames:\n",
    "        ic_wnid2imagenames[wnid] = []\n",
    "    ic_wnid2imagenames[wnid].append(image_name)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Compare distribution of image-query sim."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Resample"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "45900 45900\n"
     ]
    }
   ],
   "source": [
    "all_resampled_sm_laion_indices = []\n",
    "all_resampled_ival_image_names = []\n",
    "\n",
    "for wnid, laion_indices in sm_wnid2laionindices.items():\n",
    "    if wnid not in ival_wnid2imagenames:\n",
    "        continue\n",
    "        \n",
    "    resampled_ival_image_names = np.array(ival_wnid2imagenames[wnid])\n",
    "    \n",
    "    resampled_sm_laion_indices = np.random.choice(\n",
    "        laion_indices, \n",
    "        size=(len(resampled_ival_image_names),), \n",
    "        replace=True\n",
    "    )\n",
    "    \n",
    "    all_resampled_sm_laion_indices.extend(resampled_sm_laion_indices)\n",
    "    all_resampled_ival_image_names.extend(resampled_ival_image_names)\n",
    "    \n",
    "print(len(all_resampled_sm_laion_indices), len(all_resampled_ival_image_names))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "sm_iqsims = sm_df.loc[all_resampled_sm_laion_indices, image_query_sim_col].tolist()\n",
    "ival_iqsims = ival_df.loc[all_resampled_ival_image_names, image_query_sim_col].tolist()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Plot distributions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 340x250 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(3.4, 2.5))\n",
    "\n",
    "plt.hist(sm_iqsims, bins=60, color='b', alpha=0.3, density=True)\n",
    "plt.hist(ival_iqsims, bins=60, color='r', alpha=0.3, density=True)\n",
    "\n",
    "plt.xlabel('image-to-[name: def] CLIP similarity')\n",
    "plt.ylabel('\\npdf')\n",
    "\n",
    "plt.legend(['LAIONet', 'ImageNet'])\n",
    "\n",
    "plt.xlim([0, 1])\n",
    "\n",
    "plt.tight_layout()\n",
    "\n",
    "plt.savefig(os.path.join('figs', 'distribution_image-query_sim.pdf'), bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Test per WNID"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|███████████████████████████████████████████████████████████████████████████████████████████████████| 918/918 [00:00<00:00, 929.18it/s]\n"
     ]
    }
   ],
   "source": [
    "sm_avg_iqsims = []\n",
    "sm_err_iqsims = []\n",
    "ival_avg_iqsims = []\n",
    "ival_err_iqsims = []\n",
    "\n",
    "for wnid, laion_indices in tqdm(sm_wnid2laionindices.items()):\n",
    "    if wnid not in ival_wnid2imagenames:\n",
    "        continue\n",
    "        \n",
    "    sm_iqsims = np.array(sm_df.loc[laion_indices, image_query_sim_col].tolist())\n",
    "    sm_iqsims = sm_iqsims[~np.isnan(sm_iqsims)]\n",
    "    \n",
    "    ival_iqsims = np.array(ival_df.loc[ival_wnid2imagenames[wnid], image_query_sim_col].tolist())\n",
    "    ival_iqsims = ival_iqsims[~np.isnan(ival_iqsims)]\n",
    "    \n",
    "    if len(sm_iqsims) < 7 or len(ival_iqsims) < 7:\n",
    "        continue\n",
    "        \n",
    "    sm_avg_iqsims.append(np.mean(sm_iqsims))\n",
    "    sm_err_iqsims.append(1.96*np.std(sm_iqsims)/np.sqrt(len(sm_iqsims) - 1))\n",
    "    \n",
    "    ival_avg_iqsims.append(np.mean(ival_iqsims))\n",
    "    ival_err_iqsims.append(1.96*np.std(ival_iqsims)/np.sqrt(len(ival_iqsims) - 1))\n",
    "    \n",
    "sm_avg_iqsims = np.array(sm_avg_iqsims)\n",
    "sm_err_iqsims = np.array(sm_err_iqsims)\n",
    "ival_avg_iqsims = np.array(ival_avg_iqsims)\n",
    "ival_err_iqsims = np.array(ival_err_iqsims)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "avg > 0: 0.682983682983683\n",
      "UCB < 0: 118 out of 858 (0.13752913752913754)\n",
      "LCB > 0: 373 out of 858 (0.4347319347319347)\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 340x250 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(3.4, 2.5))\n",
    "\n",
    "diff_avg_iqsims = sm_avg_iqsims - ival_avg_iqsims\n",
    "diff_err_iqsims = np.sqrt(sm_err_iqsims**2 + ival_err_iqsims**2)\n",
    "\n",
    "sorted_locs = np.argsort(diff_avg_iqsims)\n",
    "\n",
    "ucb_iqsims = np.sort(diff_avg_iqsims + diff_err_iqsims)\n",
    "lcb_iqsims = np.sort(diff_avg_iqsims - diff_err_iqsims)\n",
    "\n",
    "x = range(len(sorted_locs))\n",
    "\n",
    "plt.plot(x, diff_avg_iqsims[sorted_locs], color='k')\n",
    "plt.plot(x, ucb_iqsims, 'r')\n",
    "plt.plot(x, lcb_iqsims, 'b')\n",
    "\n",
    "plt.plot(x, [0]*len(x), 'k--', alpha=0.5)\n",
    "\n",
    "plt.legend(['LAIONet - ImageNet', 'UCB', 'LCB'])\n",
    "\n",
    "plt.xlabel(' ')\n",
    "plt.ylabel(r'$\\Delta$ image-to-[name: def]' + '\\n' + 'CLIP similarity')\n",
    "\n",
    "plt.tight_layout()\n",
    "\n",
    "plt.savefig(os.path.join('figs', 'image-query_sim_diff.pdf'), bbox_inches='tight')\n",
    "\n",
    "print(f'avg > 0: {np.mean(diff_avg_iqsims > 0)}')\n",
    "print(f'UCB < 0: {np.sum(ucb_iqsims < 0)} out of {len(ucb_iqsims)} ({np.mean(ucb_iqsims < 0)})')\n",
    "print(f'LCB > 0: {np.sum(lcb_iqsims > 0)} out of {len(lcb_iqsims)} ({np.mean(lcb_iqsims > 0)})')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Compare distribution of text-query sim."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Settings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "ic_all_file_name = 'imagenet_captions_with_sims_to_all_queries.parquet'\n",
    "\n",
    "synsets_file_name = 'ILSVRC2012_synsets.txt'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "text_to_query_col_func = lambda w: f'text_to_name_def_{w}_similarity'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "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>TEXT</th>\n",
       "      <th>image_to_name_def_wnid_similarity</th>\n",
       "      <th>name_def_wnid</th>\n",
       "      <th>image_to_text_similarity</th>\n",
       "      <th>text_to_name_def_wnid_similarity</th>\n",
       "      <th>text_to_name_def_n02119789_similarity</th>\n",
       "      <th>text_to_name_def_n02100735_similarity</th>\n",
       "      <th>text_to_name_def_n02096294_similarity</th>\n",
       "      <th>text_to_name_def_n02066245_similarity</th>\n",
       "      <th>text_to_name_def_n02509815_similarity</th>\n",
       "      <th>...</th>\n",
       "      <th>text_to_name_def_n02091032_similarity</th>\n",
       "      <th>text_to_name_def_n02102318_similarity</th>\n",
       "      <th>text_to_name_def_n02102480_similarity</th>\n",
       "      <th>text_to_name_def_n02113023_similarity</th>\n",
       "      <th>text_to_name_def_n02086646_similarity</th>\n",
       "      <th>text_to_name_def_n02091244_similarity</th>\n",
       "      <th>text_to_name_def_n02089973_similarity</th>\n",
       "      <th>text_to_name_def_n02105251_similarity</th>\n",
       "      <th>text_to_name_def_n02093754_similarity</th>\n",
       "      <th>text_to_name_def_n02112837_similarity</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ic_index</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>n15075141_9819.JPEG</th>\n",
       "      <td>Margaret Thatcher toilet paper Margaret Thatch...</td>\n",
       "      <td>0.255074</td>\n",
       "      <td>toilet tissue which is a soft thin absorbent p...</td>\n",
       "      <td>0.321070</td>\n",
       "      <td>0.672603</td>\n",
       "      <td>0.314751</td>\n",
       "      <td>0.302764</td>\n",
       "      <td>0.376714</td>\n",
       "      <td>0.353923</td>\n",
       "      <td>0.375146</td>\n",
       "      <td>...</td>\n",
       "      <td>0.385962</td>\n",
       "      <td>0.425324</td>\n",
       "      <td>0.274066</td>\n",
       "      <td>0.418079</td>\n",
       "      <td>0.320774</td>\n",
       "      <td>0.332809</td>\n",
       "      <td>0.360865</td>\n",
       "      <td>0.330163</td>\n",
       "      <td>0.374264</td>\n",
       "      <td>0.359121</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>n15075141_9942.JPEG</th>\n",
       "      <td>Toilet Roll II toilet roll II toilet roll duan...</td>\n",
       "      <td>0.322143</td>\n",
       "      <td>toilet tissue which is a soft thin absorbent p...</td>\n",
       "      <td>0.284109</td>\n",
       "      <td>0.721321</td>\n",
       "      <td>0.392613</td>\n",
       "      <td>0.283349</td>\n",
       "      <td>0.407590</td>\n",
       "      <td>0.410512</td>\n",
       "      <td>0.376387</td>\n",
       "      <td>...</td>\n",
       "      <td>0.434423</td>\n",
       "      <td>0.416335</td>\n",
       "      <td>0.250755</td>\n",
       "      <td>0.428895</td>\n",
       "      <td>0.288285</td>\n",
       "      <td>0.323935</td>\n",
       "      <td>0.369291</td>\n",
       "      <td>0.318132</td>\n",
       "      <td>0.382490</td>\n",
       "      <td>0.441221</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>n15075141_9993.JPEG</th>\n",
       "      <td>Toilet Paper toilet paper flush light green wa...</td>\n",
       "      <td>0.271322</td>\n",
       "      <td>toilet tissue which is a soft thin absorbent p...</td>\n",
       "      <td>0.294822</td>\n",
       "      <td>0.380635</td>\n",
       "      <td>0.107943</td>\n",
       "      <td>0.129566</td>\n",
       "      <td>0.120912</td>\n",
       "      <td>0.165459</td>\n",
       "      <td>0.166710</td>\n",
       "      <td>...</td>\n",
       "      <td>0.156739</td>\n",
       "      <td>0.161972</td>\n",
       "      <td>0.089361</td>\n",
       "      <td>0.150558</td>\n",
       "      <td>0.098518</td>\n",
       "      <td>0.148388</td>\n",
       "      <td>0.154622</td>\n",
       "      <td>0.099526</td>\n",
       "      <td>0.142871</td>\n",
       "      <td>0.176329</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>3 rows × 1004 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                                  TEXT  \\\n",
       "ic_index                                                                 \n",
       "n15075141_9819.JPEG  Margaret Thatcher toilet paper Margaret Thatch...   \n",
       "n15075141_9942.JPEG  Toilet Roll II toilet roll II toilet roll duan...   \n",
       "n15075141_9993.JPEG  Toilet Paper toilet paper flush light green wa...   \n",
       "\n",
       "                     image_to_name_def_wnid_similarity  \\\n",
       "ic_index                                                 \n",
       "n15075141_9819.JPEG                           0.255074   \n",
       "n15075141_9942.JPEG                           0.322143   \n",
       "n15075141_9993.JPEG                           0.271322   \n",
       "\n",
       "                                                         name_def_wnid  \\\n",
       "ic_index                                                                 \n",
       "n15075141_9819.JPEG  toilet tissue which is a soft thin absorbent p...   \n",
       "n15075141_9942.JPEG  toilet tissue which is a soft thin absorbent p...   \n",
       "n15075141_9993.JPEG  toilet tissue which is a soft thin absorbent p...   \n",
       "\n",
       "                     image_to_text_similarity  \\\n",
       "ic_index                                        \n",
       "n15075141_9819.JPEG                  0.321070   \n",
       "n15075141_9942.JPEG                  0.284109   \n",
       "n15075141_9993.JPEG                  0.294822   \n",
       "\n",
       "                     text_to_name_def_wnid_similarity  \\\n",
       "ic_index                                                \n",
       "n15075141_9819.JPEG                          0.672603   \n",
       "n15075141_9942.JPEG                          0.721321   \n",
       "n15075141_9993.JPEG                          0.380635   \n",
       "\n",
       "                     text_to_name_def_n02119789_similarity  \\\n",
       "ic_index                                                     \n",
       "n15075141_9819.JPEG                               0.314751   \n",
       "n15075141_9942.JPEG                               0.392613   \n",
       "n15075141_9993.JPEG                               0.107943   \n",
       "\n",
       "                     text_to_name_def_n02100735_similarity  \\\n",
       "ic_index                                                     \n",
       "n15075141_9819.JPEG                               0.302764   \n",
       "n15075141_9942.JPEG                               0.283349   \n",
       "n15075141_9993.JPEG                               0.129566   \n",
       "\n",
       "                     text_to_name_def_n02096294_similarity  \\\n",
       "ic_index                                                     \n",
       "n15075141_9819.JPEG                               0.376714   \n",
       "n15075141_9942.JPEG                               0.407590   \n",
       "n15075141_9993.JPEG                               0.120912   \n",
       "\n",
       "                     text_to_name_def_n02066245_similarity  \\\n",
       "ic_index                                                     \n",
       "n15075141_9819.JPEG                               0.353923   \n",
       "n15075141_9942.JPEG                               0.410512   \n",
       "n15075141_9993.JPEG                               0.165459   \n",
       "\n",
       "                     text_to_name_def_n02509815_similarity  ...  \\\n",
       "ic_index                                                    ...   \n",
       "n15075141_9819.JPEG                               0.375146  ...   \n",
       "n15075141_9942.JPEG                               0.376387  ...   \n",
       "n15075141_9993.JPEG                               0.166710  ...   \n",
       "\n",
       "                     text_to_name_def_n02091032_similarity  \\\n",
       "ic_index                                                     \n",
       "n15075141_9819.JPEG                               0.385962   \n",
       "n15075141_9942.JPEG                               0.434423   \n",
       "n15075141_9993.JPEG                               0.156739   \n",
       "\n",
       "                     text_to_name_def_n02102318_similarity  \\\n",
       "ic_index                                                     \n",
       "n15075141_9819.JPEG                               0.425324   \n",
       "n15075141_9942.JPEG                               0.416335   \n",
       "n15075141_9993.JPEG                               0.161972   \n",
       "\n",
       "                     text_to_name_def_n02102480_similarity  \\\n",
       "ic_index                                                     \n",
       "n15075141_9819.JPEG                               0.274066   \n",
       "n15075141_9942.JPEG                               0.250755   \n",
       "n15075141_9993.JPEG                               0.089361   \n",
       "\n",
       "                     text_to_name_def_n02113023_similarity  \\\n",
       "ic_index                                                     \n",
       "n15075141_9819.JPEG                               0.418079   \n",
       "n15075141_9942.JPEG                               0.428895   \n",
       "n15075141_9993.JPEG                               0.150558   \n",
       "\n",
       "                     text_to_name_def_n02086646_similarity  \\\n",
       "ic_index                                                     \n",
       "n15075141_9819.JPEG                               0.320774   \n",
       "n15075141_9942.JPEG                               0.288285   \n",
       "n15075141_9993.JPEG                               0.098518   \n",
       "\n",
       "                     text_to_name_def_n02091244_similarity  \\\n",
       "ic_index                                                     \n",
       "n15075141_9819.JPEG                               0.332809   \n",
       "n15075141_9942.JPEG                               0.323935   \n",
       "n15075141_9993.JPEG                               0.148388   \n",
       "\n",
       "                     text_to_name_def_n02089973_similarity  \\\n",
       "ic_index                                                     \n",
       "n15075141_9819.JPEG                               0.360865   \n",
       "n15075141_9942.JPEG                               0.369291   \n",
       "n15075141_9993.JPEG                               0.154622   \n",
       "\n",
       "                     text_to_name_def_n02105251_similarity  \\\n",
       "ic_index                                                     \n",
       "n15075141_9819.JPEG                               0.330163   \n",
       "n15075141_9942.JPEG                               0.318132   \n",
       "n15075141_9993.JPEG                               0.099526   \n",
       "\n",
       "                     text_to_name_def_n02093754_similarity  \\\n",
       "ic_index                                                     \n",
       "n15075141_9819.JPEG                               0.374264   \n",
       "n15075141_9942.JPEG                               0.382490   \n",
       "n15075141_9993.JPEG                               0.142871   \n",
       "\n",
       "                     text_to_name_def_n02112837_similarity  \n",
       "ic_index                                                    \n",
       "n15075141_9819.JPEG                               0.359121  \n",
       "n15075141_9942.JPEG                               0.441221  \n",
       "n15075141_9993.JPEG                               0.176329  \n",
       "\n",
       "[3 rows x 1004 columns]"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ic_all_df = pd.read_parquet(os.path.join(ilsvrc_path, ic_all_file_name))\n",
    "\n",
    "ic_all_df.tail(3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "id_lemmas_df = load_lemmas_and_wnids(os.path.join(ilsvrc_path, 'ILSVRC2012_synsets.txt'))\n",
    "all_wnids = id_lemmas_df[configs.ILSVRCConfigs.WNID_COL].tolist()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Plot distribution"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAUAAAADwCAYAAABxLb1rAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjcuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/bCgiHAAAACXBIWXMAAA9hAAAPYQGoP6dpAAA0s0lEQVR4nO3deVRT1/o38G8IJIAQKAoBFAFFBBTFkcaxVSq9UlveOuBEAdFWBa3igNQqDlUsTjiPFdp7sahtrQM4oqhFql5wVlARxFYGrQwCMu/3Dy/nZwAVQkICeT5rZS2zz84+T0LyuM/Z5+zNY4wxEEKIGtJQdgCEEKIslAAJIWqLEiAhRG1RAiSEqC1KgIQQtUUJkBCitigBEkLUFiVAQoja0lR2AC1FVVUVnjx5An19ffB4PGWHQ0iLwhjDixcvYG5uDg0N+fXbKAHKyZMnT2BhYaHsMAhp0R4/fox27drJrT1KgHKir68P4NUfSCQSKTkaQlqWgoICWFhYcL8zeaEEKCfVh70ikYgSICEKIu/TSzQIQghRW5QACSFqixIgIURt0TnAZqqyshLl5eXKDoMQudDS0gKfz2/y/VICbGYYY8jKykJeXp6yQyFErgwNDWFqatqk19FSAmxmqpOfiYkJdHV16aJrIlcFBUBTX8TAGENxcTFycnIAAGZmZk22b0qAzUhlZSWX/Fq3bq3scEgLVFICaGs3/X51dHQAADk5OTAxMWmyw2EaBGlGqs/56erqKjkSQuSv+nvdlOe2KQE2Q3TYS1oiZXyvKQESQtQWJUBCiNqiQZAW4siRpt3fiBENq+/t7Y28vDz8/vvvb6zz119/oUOHDrC1tcWtW7dqbefxeDh48CDc3d25sqNHj2L16tVISkpCZWUlunTpAj8/P3h7e3N10tPTYW1tDWNjY6SmpkrdUO/k5AR3d3csWbKkXu8jIiICs2bNosuQWgjqARKVERERgTFjxqCgoACXLl16Z/1Nmzbhs88+Q//+/XHp0iXcuHEDY8eOxdSpUzF37txa9V+8eIE1a9YoInTSTFECJCqBMYbw8HB4enpi/Pjx+OGHH95a//Hjx5gzZw5mzZqFlStXwsHBATY2NpgzZw5Wr16NtWvX1kqiM2bMwLp167jrzepSWlqKuXPnom3btmjVqhWcnZ0RFxcHAIiLi4OPjw/y8/PB4/HA4/Hq3XMkqokSIFEJZ8+eRXFxMVxcXDBx4kRERUWhqKjojfV/+eUXlJeX19nT++qrr6Cnp4eff/5ZqnzcuHGwsbHBsmXL3tiuv78/EhISEBUVhRs3bmD06NH4+OOPcf/+ffTr1w9hYWEQiUTIzMxEZmZmnftv7tTp6J4SIFEJP/zwA8aOHQs+n4+uXbuiQ4cOOHDgwBvr37t3DwYGBnXeNSAQCNChQwfcu3dPqpzH42HVqlXYuXMnUlNTa70uIyMD4eHhOHDgAAYOHIiOHTti7ty5GDBgAMLDwyEQCGBgYAAejwdTU1OYmppCT0+v8W+eKA0lQKJ0eXl5+O233zBx4kSubOLEie88DJaFq6srBgwYgEWLFtXadvPmTVRWVsLW1hZ6enrc49y5c3UmTNL80SgwUbq9e/eipKQEzs7OXBljDFVVVbh37x5sbW1rvcbW1hb5+fl48uQJzM3NpbaVlZUhNTUVH374YZ37W7VqFSQSCebNmydVXlhYCD6fj8TExFq3YlFPr2WiHiBRuh9++AFz5szBtWvXuMf169cxcOBA7Nmzp87XjBw5ElpaWli7dm2tbdu3b0dRURHGjRtX52v79u2Lzz//HAsWLJAq79GjByorK5GTkwMbGxuph6mpKYBXh9eVlZWNfMdEVVAPkDSZ/Px8XLt2TarsxYsXSEpKQmRkJOzs7KS2jRs3DsuWLcN3330HTU3pr2r79u0RGhqKOXPmQFtbG56entDS0sKhQ4fwzTffYM6cOVI9yppWrFiBLl26SLVra2uLCRMm4IsvvsDatWvRo0cPPH36FLGxsejWrRvc3NxgZWWFwsJCxMbGonv37tDV1aV7s5sx6gGSJhMXF4cePXpIPfbs2QMHB4dayQ8A/t//+3/IyclBTExMne3NmjULBw8exIULF9C7d2907doVe/fuxbZt2955vZ+trS0mTZqEkpISqfLw8HB88cUXmDNnDjp37gx3d3dcuXIF7du3BwD069cPU6dOhYeHB4yNjREaGirjp0FUAY8xxpQdREtQUFAAAwMD5OfnK2xVuJKSEqSlpcHa2hraypiziLR41ZfAGBo2/b7f9v1W1O+LeoCEELWl9AT4999/Y+LEiWjdujV0dHTg6OiI//73v9x2xhgWL14MMzMz6OjowMXFBffv35dq4/nz55gwYQJEIhEMDQ3h6+uLwsJCqTo3btzAwIEDoa2tDQsLizoPXQ4cOAA7Oztoa2vD0dHxjYdehJCWQakJMDc3F/3794eWlhaOHTuGO3fuYO3atXjvvfe4OqGhodi4cSO2b9+OS5cuoVWrVnB1dZU6dzNhwgTcvn0bp06dwtGjR3H+/Hl8+eWX3PaCggIMGzYMlpaWSExMxOrVq7FkyRLs3LmTq3Px4kWMGzcOvr6+uHr1Ktzd3eHu7l7nTfmEkBaCKVFgYCAbMGDAG7dXVVUxU1NTtnr1aq4sLy+PCYVC9vPPPzPGGLtz5w4DwK5cucLVOXbsGOPxeOzvv/9mjDG2detW9t5777HS0lKpfXfu3Jl7PmbMGObm5ia1f2dnZ/bVV1/V673k5+czACw/P79e9WXx8uVLdufOHfby5UuF7YOot9zcVw9leNv3W1G/L6X2AA8fPozevXtj9OjRMDExQY8ePbBr1y5ue1paGrKysuDi4sKVGRgYwNnZGQkJCQCAhIQEGBoaonfv3lwdFxcXaGhocDfDJyQkYNCgQRAIBFwdV1dXpKSkIDc3l6vz+n6q61TvhxDS8ig1AT58+BDbtm1Dp06dcOLECUybNg0zZ87Ejz/+CODVCmgAIBaLpV4nFou5bVlZWTAxMZHarqmpCSMjI6k6dbXx+j7eVKd6e02lpaUoKCiQehBCmhelXghdVVWF3r17Y+XKlQBeXYl/69YtbN++HV5eXsoM7Z1CQkKwdOlSZYdBCGkEpfYAzczM4ODgIFVmb2+PjIwMAOBuP8rOzpaqk52dzW0zNTWtNb9bRUUFnj9/LlWnrjZe38eb6lRvrykoKAj5+fnc4/Hjx/V704QQlaHUBNi/f3+kpKRIld27dw+WlpYAAGtra5iamiI2NpbbXj1bsEQiAQBIJBLk5eUhMTGRq3PmzBlUVVVxt0JJJBKcP39earm9U6dOoXPnztyIs0QikdpPdZ3q/dQkFAohEomkHoQ0V3FxceDxeMjPz1N2KE1LrkMqDXT58mWmqanJVqxYwe7fv88iIyOZrq4u+89//sPVWbVqFTM0NGSHDh1iN27cYJ999hmztraWGin6+OOPWY8ePdilS5fYH3/8wTp16sTGjRvHbc/Ly2NisZh5enqyW7dusaioKKarq8t27NjB1YmPj2eamppszZo17O7duyw4OJhpaWmxmzdv1uu90Cjw23l5ebHPPvtM2WG8U1paGgPAjI2NWUFBgdS27t27s+Dg4Hq3FR4ezgwMDOpdPykpiY0aNYqZmJgwoVDIbGxs2OTJk1lKSkq926iPwYMHs6+//lqqrLS0lGVmZrLnz6vUahRYqecA+/Tpg4MHDyIoKAjLli2DtbU1wsLCMGHCBK7O/PnzUVRUhC+//BJ5eXkYMGAAjh8/LnWrTGRkJPz9/TF06FBoaGhg5MiR2LhxI7fdwMAAJ0+ehJ+fH3r16oU2bdpg8eLFUtcK9uvXD3v37sW3336Lb775Bp06dcLvv/+Orl27Ns2H0ViqvipSM1O9fkhTnec9evQoRo4cCVdXV0RGRqJjx47IycnBgQMHsGjRIuzbt0+h+xcIBDA1NVWr2aABFbgT5JNPPsHNmzdRUlKCu3fvYsqUKVLbeTweli1bhqysLJSUlOD06dO15oczMjLC3r178eLFC+Tn52PPnj215m/r1q0bLly4gJKSEvz1118IDAysFcvo0aORkpKC0tJS3Lp1C8OHD5f/Gyb44IMPMGPGDMyaNQvvvfcexGIxdu3ahaKiIvj4+EBfXx82NjY4duwY95rKykr4+vrC2toaOjo66Ny5MzZs2CDVbkVFBWbOnAlDQ0O0bt0agYGB8PLyklpFrqqqCiEhIVw73bt3xy+//FIrxqZcP6S4uBg+Pj4YPnw4Dh8+DBcXF1hbW8PZ2Rlr1qzBjh076v0ZeHt7w93dHUuXLoWxsTFEIhGmTp2KsrIybvu5c+ewYcMGLq709PQ6D4F//fVXdOnSBUKhEFZWVrWmHrOyssLKlSsxadIk6Ovro3379lI3F5SVlcHf3x9mZmbQ1taGpaUlQkJC3vh5KoPSEyBRTz/++CPatGmDy5cvY8aMGZg2bRpGjx6Nfv36ISkpCcOGDYOnpyeKi4sBvEpc7dq1w4EDB3Dnzh0sXrwY33zzDfbv38+1+f333yMyMhLh4eGIj49HQUFBrWU4Q0JC8NNPP2H79u24ffs2Zs+ejYkTJ+LcuXNS9Zpy/ZATJ07g2bNnmD9/fp3bDf83M0F9PgMAiI2Nxd27dxEXF4eff/4Zv/32G9eT3bBhAyQSCaZMmcLFZWFhUWufcXGJGDNmDMaOHYubN29iyZIlWLRoESIiIqTqrV27Fr1798bVq1cxffp0TJs2jTuvv3HjRhw+fBj79+9HSkoKIiMjYWVl9cbPUynkekCtxpR+DvDw4aZ9NNDr5wAHDx4sdQdQRUUFa9WqFfP09OTKMjMzGQCWkJDwxjb9/PzYyJEjuedisVjqrqGKigrWvn17br8lJSVMV1eXXbx4UaodX19f7pxx9TnAq1evsuPHjzMtLS324MEDxpj0OcBHjx4xPp/P3W1UbejQoSwoKIgxVv9zgN9//z0DwJ4/f/7OujXV/Ay8vLyYkZERKyoq4sq2bdvG9PT0WGVlJWOs7nOAZ8+eZQBYenouy81lbNSo8eyjjz6SqjNv3jzm4ODAPbe0tGQTJ07knldVVTETExO2bds2xhhjM2bMYEOGDGFVVVX1ei9qdycIUV/dunXj/s3n89G6dWs4OjpyZdUXpb9+CLplyxb06tULxsbG0NPTw86dO7lLpvLz85GdnY2+fftKtdurVy/u+YMHD1BcXIyPPvpIas2Pn376qc41PxSxfsjKlSul6mdkZIA1YEa6t30G1aonaq0mkUhQWFjYoEu17t27i/79+0uV9e/fH/fv35eaEfv1v2P1YlHVfzNvb29cu3YNnTt3xsyZM3Hy5Ml677+p0IzQRCm0tLSknvN4PKkyHo8H4NVhHwBERUVh7ty5WLt2LSQSCfT19bF69ep6LaBerXqGoOjoaLRt21Zqm1AorPM18l4/ZOrUqRgzZgz33NzcnDunnZyc/MbLrgD5fAbyVtffsfpv1rNnT6SlpeHYsWM4ffo0xowZAxcXlzrPuSoLJUDSLMTHx6Nfv36YPn06V/Z6T8vAwABisRhXrlzBoEGDALwaNEhKSoKTkxMAwMHBAUKhEBkZGRg8eHC99luf9UMGDhxY52vrWj/EyMgIRkZGUmXDhg1DmzZtEBoaioMHD9ZqJy8vD4aGhu/8DKpdv34dL1++hI6ODgDgzz//hJ6eHneurz7rmtja2iM+Pl6qLD4+Hra2trUS/tuIRCJ4eHjAw8MDo0aNwscff4znz5/X+gyUhRIgaRY6deqEn376CSdOnIC1tTX+/e9/48qVK7C2tubqzJgxAyEhIbCxsYGdnR02bdqE3Nxcrjepr6+PuXPnYvbs2aiqqsKAAQOQn5+P+Ph4iESiN95+qej1Q1q1aoXdu3dj9OjR+PTTTzFz5kzY2Njg2bNn2L9/PzIyMhAVFVWvzwB4Nfrq6+uLb7/9Funp6QgODoa/vz80NF6d8bKyssKlS5eQnp4OPT29OpORv/8cDBnSB8uXL4eHhwcSEhKwefNmbN26td5/s3Xr1sHMzAw9evSAhoYGDhw4AFNTU25QRxXQOUDSLHz11Vf4/PPP4eHhAWdnZ/zzzz9SPSEACAwMxLhx4/DFF19AIpFAT08Prq6uUteMLl++HIsWLUJISAjs7e3x8ccfIzo6ulYSeV1TrB/y2Wef4eLFi9DS0sL48eNhZ2eHcePGIT8/H9999129PwMAGDp0KDp16oRBgwbBw8MDn376qdQlOHPnzgWfz4eDgwOMjY1rnUMEgO7deyI8fD+ioqLQtWtXLF68GMuWLYO3t/cb30NN+vr6CA0NRe/evdGnTx+kp6cjJiaGS8SqgNYEkRNaE0T1VFVVwd7eHmPGjMHy5cuVHU6T8Pb2Rl5eXq3Lf+qr5oXQTdlZU8aaIHQITFqMR48e4eTJkxg8eDBKS0uxefNmpKWlYfz48coOjago1emLEtJIGhoaiIiIQJ8+fdC/f3/cvHkTp0+fhr29vbJDIyqKeoCkxbCwsKg1cqluat6pQd6OeoCEELVFCZAQorYoATZDNHBPWiJlfK8pATYj1bcdVc+QQoiiNeX8gNXf65q31ykSDYI0I3w+H4aGhtzN5rq6utxdDoTIw/+mDZRS4/pvuWOMobi4GDk5OTA0NGzQrXaNRQmwmalepOltE3USIqu6Di7quHtPIQwNDd+4CJmiUAJsZng8HszMzGBiYiK1yBMh8nD2bO2yDz9U/H61tLSatOdXjRJgM8Xn85XyhSEtW0VF7bKWfNclDYIQQtQWJUBCiNqiBEgIUVuUAAkhaosSICFEbVECJISoLUqAhBC1RQmQEKK2KAESQtQWJUBCiNqiBEgIeasjR5QdgeJQAiSEqC1KgIQQtUUJkBCitigBEkLUlkwJcMiQIcirY7GAgoICDBkypLExEUJIk5ApAcbFxaGsjsUDSkpKcOHChUYHRQghTaFBM0LfuHGD+/edO3eQlZXFPa+srMTx48fRtm1b+UVHCFEJR44AI0YoOwr5a1ACdHJyAo/HA4/Hq/NQV0dHB5s2bZJbcIQQokgNOgROS0tDamoqGGO4fPky0tLSuMfff/+NgoICTJo0SaZAVq1aBR6Ph1mzZnFlJSUl8PPzQ+vWraGnp4eRI0ciOztb6nUZGRlwc3ODrq4uTExMMG/ePFTUWNggLi4OPXv2hFAohI2NDSIiImrtf8uWLbCysoK2tjacnZ1x+fJlmd4HIaT5aFAP0NLSEgBQVVUl1yCuXLmCHTt2oFu3blLls2fPRnR0NA4cOAADAwP4+/vj888/R3x8PIBXh91ubm4wNTXFxYsXkZmZiS+++AJaWlpYuXIlgFdJ283NDVOnTkVkZCRiY2MxefJkmJmZwdXVFQCwb98+BAQEYPv27XB2dkZYWBhcXV2RkpICExMTub5XQojq4DHGmCwvvH//Ps6ePYucnJxaCXHx4sX1bqewsBA9e/bE1q1b8d1338HJyQlhYWHIz8+HsbEx9u7di1GjRgEAkpOTYW9vj4SEBLz//vs4duwYPvnkEzx58gRisRgAsH37dgQGBuLp06cQCAQIDAxEdHQ0bt26xe1z7NixyMvLw/HjxwEAzs7O6NOnDzZv3gzgVYK3sLDAjBkzsGDBgnq9j4KCAhgYGCA/Px8ikaje758QVfK2296UeQ5QUb8vmUaBd+3aBXt7eyxevBi//PILDh48yD1+//33BrXl5+cHNzc3uLi4SJUnJiaivLxcqtzOzg7t27dHQkICACAhIQGOjo5c8gMAV1dXFBQU4Pbt21ydmm27urpybZSVlSExMVGqjoaGBlxcXLg6hJCWSaZ1gb/77jusWLECgYGBjdp5VFQUkpKScOXKlVrbsrKyIBAIYGhoKFUuFou50eesrCyp5Fe9vXrb2+oUFBTg5cuXyM3NRWVlZZ11kpOT3xh7aWkpSktLuecFBQXveLeEEFUjUw8wNzcXo0ePbtSOHz9+jK+//hqRkZHQboYrL4eEhMDAwIB7WFhYKDskQkgDyZQAR48ejZMnTzZqx4mJicjJyUHPnj2hqakJTU1NnDt3Dhs3boSmpibEYjHKyspq3XGSnZ0NU1NTAICpqWmtUeHq5++qIxKJoKOjgzZt2oDP59dZp7qNugQFBSE/P597PH78WKbPgRCiPDIdAtvY2GDRokX4888/4ejoCC0tLantM2fOfGcbQ4cOxc2bN6XKfHx8YGdnh8DAQFhYWEBLSwuxsbEYOXIkACAlJQUZGRmQSCQAAIlEghUrViAnJ4cbrT116hREIhEcHBy4OjExMVL7OXXqFNeGQCBAr169EBsbC3d3dwCvBkFiY2Ph7+//xviFQiGEQuE73ychRHXJNApsbW395gZ5PDx8+FCmYD744ANuFBgApk2bhpiYGEREREAkEmHGjBkAgIsXLwJ4dRmMk5MTzM3NERoaiqysLHh6emLy5MlSl8F07doVfn5+mDRpEs6cOYOZM2ciOjpa6jIYLy8v7NixA3379kVYWBj279+P5OTkWucG34RGgUlLoG6jwDL1ANPS0uQWwNusX78eGhoaGDlyJEpLS+Hq6oqtW7dy2/l8Po4ePYpp06ZBIpGgVatW8PLywrJly7g61tbWiI6OxuzZs7Fhwwa0a9cOu3fv5pIfAHh4eODp06dYvHgxsrKy4OTkhOPHj9c7+RGiDlri7XAyXwdIpFEPkLQE75r+XlkJUKV6gO+63W3Pnj0yBUMIIU1JpgSYm5sr9by8vBy3bt1CXl4ezQdICGk2ZEqABw8erFVWVVWFadOmoWPHjo0OihBCmoLcpsTX0NBAQEAA1q9fL68mCSFEoeS6JkhqamqtqagIIURVyXQIHBAQIPWcMYbMzExER0fDy8tLLoERQoiiyZQAr169KvVcQ0MDxsbGWLt2rcwTohJCSFOTKQGePXtW3nEQQkiTkykBVnv69ClSUlIAAJ07d4axsbFcgiKEkKYg0yBIUVERJk2aBDMzMwwaNAiDBg2Cubk5fH19UVxcLO8YCSEq4l13ijQ3MiXAgIAAnDt3DkeOHEFeXh7y8vJw6NAhnDt3DnPmzJF3jIQQohAyHQL/+uuv+OWXX/DBBx9wZcOHD4eOjg7GjBmDbdu2ySs+QghRGJl6gMXFxXXOlGJiYkKHwISQZkOmBCiRSBAcHIySkhKu7OXLl1i6dCk30SghhKg6mQ6Bw8LC8PHHH6Ndu3bo3r07AOD69esQCoWNniqfEEKaikwJ0NHREffv30dkZCS3ctq4ceMwYcIE6OjoyDVAQghRFJkSYEhICMRiMaZMmSJVvmfPHjx9+rTRy2USQkhTkOkc4I4dO2BnZ1ervEuXLti+fXujgyKEqK6WdC2gTAkwKysLZmZmtcqNjY2RmZnZ6KAIIaQpyJQALSwsEB8fX6s8Pj4e5ubmjQ6KEEKagkznAKdMmYJZs2ahvLycmwI/NjYW8+fPpztBCCHNhkwJcN68efjnn38wffp0lJWVAQC0tbURGBiIoKAguQZICFE9LWWJzEYti1lYWIi7d+9CR0cHnTp1glAolGdszQoti0lagoYMcDRlAlSpZTGr6enpoU+fPvKKhRBCmpRc1wQhhJDmhBIgIUQmLeF6QEqAhBC1RQmQEKK2KAESQtQWJUBCiNpq1GUwhBAFqmuUoSVcfaxCKAESogpawpBqM0QJkBBlkDXh1Xwd9Qgbhc4BEkLUFvUACVE0RR7e0nnCRqEeICFEbVECJISoLToEJqSloYGSelNqDzAkJAR9+vSBvr4+TExM4O7ujpSUFKk6JSUl8PPzQ+vWraGnp4eRI0ciOztbqk5GRgbc3Nygq6sLExMTzJs3DxUVFVJ14uLi0LNnTwiFQtjY2CAiIqJWPFu2bIGVlRW0tbXh7OyMy5cvy/09kxbuyJHaD6KylJoAz507Bz8/P/z55584deoUysvLMWzYMBQVFXF1Zs+ejSNHjuDAgQM4d+4cnjx5gs8//5zbXllZCTc3N5SVleHixYv48ccfERERgcWLF3N10tLS4Obmhg8//BDXrl3DrFmzMHnyZJw4cYKrs2/fPgQEBCA4OBhJSUno3r07XF1dkZOT0zQfBiHNUHPP742aEVrenj59ChMTE5w7dw6DBg1Cfn4+jI2NsXfvXowaNQoAkJycDHt7eyQkJOD999/HsWPH8Mknn+DJkycQi8UAgO3btyMwMBBPnz6FQCBAYGAgoqOjcevWLW5fY8eORV5eHo4fPw4AcHZ2Rp8+fbB582YAQFVVFSwsLDBjxgwsWLDgnbHTjNBqqjlkgAYcAsvydpriCFtRvy+VGgTJz88HABgZGQEAEhMTUV5eDhcXF66OnZ0d2rdvj4SEBABAQkICHB0dueQHAK6urigoKMDt27e5Oq+3UV2nuo2ysjIkJiZK1dHQ0ICLiwtXp6bS0lIUFBRIPQghzYvKDIJUVVVh1qxZ6N+/P7p27Qrg1frDAoEAhoaGUnXFYjGysrK4Oq8nv+rt1dveVqegoAAvX75Ebm4uKisr66yTnJxcZ7whISFYunSpbG+WkKZE1wq+kcr0AP38/HDr1i1ERUUpO5R6CQoKQn5+Pvd4/PixskMihDSQSvQA/f39cfToUZw/fx7t2rXjyk1NTVFWVoa8vDypXmB2djZMTU25OjVHa6tHiV+vU3PkODs7GyKRCDo6OuDz+eDz+XXWqW6jJqFQqNar4Kml5nC+jzSIUnuAjDH4+/vj4MGDOHPmDKytraW29+rVC1paWoiNjeXKUlJSkJGRAYlEAgCQSCS4efOm1GjtqVOnIBKJ4ODgwNV5vY3qOtVtCAQC9OrVS6pOVVUVYmNjuTqEkJZHqT1APz8/7N27F4cOHYK+vj53zs7AwAA6OjowMDCAr68vAgICYGRkBJFIhBkzZkAikeD9998HAAwbNgwODg7w9PREaGgosrKy8O2338LPz4/roU2dOhWbN2/G/PnzMWnSJJw5cwb79+9HdHQ0F0tAQAC8vLzQu3dv9O3bF2FhYSgqKoKPj0/TfzCEkCah1MtgeDxeneXh4eHw9vYG8OpC6Dlz5uDnn39GaWkpXF1dsXXrVqlD00ePHmHatGmIi4tDq1at4OXlhVWrVkFT8//ye1xcHGbPno07d+6gXbt2WLRoEbePaps3b8bq1auRlZUFJycnbNy4Ec7OzvV6L3QZjBpo6YfAI0bI/BYVPaaiqN+XSl0H2JxRAlQDlADf9lKFUtTvSyUGQQhRSS094RHVuQyGEEKaGiVAQojaogRICFFbdA6QEIDO9wHAkSMQv3ZPQXbfln+7HPUACSFqixIgIURtUQIkhDRacz2DQAmQEKK2aBCEqJ/m2l0hckc9QEKIXDTH/1eoB0gIqZP4cu2M1tIujaEESFq+5tg1aaaOHGles+3TITAhRG1RAiSEqC1KgIQQtUUJkBCitmgQhLQsNOChUC1tZJh6gIQQtUU9QNJ8UW+PNBL1AAkhaosSICFEbdEhMGk+6JC3Waj+MzWHO0IoARJCGqXmyHBzGhWmQ2BCiNqiBEgIUYjmcMaCDoGJamoOvx7S7FECJKqBEh5RAjoEJoSoLeoBEkLkqq77hVX1mhhKgKTp0eEuURF0CEwIUZjLl5UdwdtRD5AoFvX21N7ly4CqXhtNPUBCiMKp6v+D1AMk8qWq33RC6kAJkBCicLVGhlVkVJgSIKkf6tmRRrp8GejbV9lRSKNzgDVs2bIFVlZW0NbWhrOzMy6r+jAWIc2Iqv2cqAf4mn379iEgIADbt2+Hs7MzwsLC4OrqipSUFJiYmCg7PMWh3h1pQpcvA9lQjaNgHmOMKTsIVeHs7Iw+ffpg8+bNAICqqipYWFhgxowZWLBgwVtfW1BQAAMDA+Tn50MkEjVFuLKjhEfeoKl7aFKHxG/JiIr6fVEP8H/KysqQmJiIoKAgrkxDQwMuLi5ISEhQYmQNQImNNDPKPi9ICfB/nj17hsrKSojFYqlysViM5OTkWvVLS0tRWlrKPc/Pzwfw6n+qBjt2rOGvIUQBCkvfXUfezlz43z8u7AMA9O712sZ//QvA//2u5H3ASglQRiEhIVi6dGmtcgsLCyVEQ4h6+Oeff2BgYCC39igB/k+bNm3A5/ORnZ0tVZ6dnQ1TU9Na9YOCghAQEMA9z8vLg6WlJTIyMuT6B1KEgoICWFhY4PHjxyp/vrI5xQo0r3ibU6z5+flo3749jIyM5NouJcD/EQgE6NWrF2JjY+Hu7g7g1SBIbGws/P39a9UXCoUQCoW1yg0MDFT+y1RNJBJRrArSnOJtTrFqaMj3yj1KgK8JCAiAl5cXevfujb59+yIsLAxFRUXw8fFRdmiEEAWgBPgaDw8PPH36FIsXL0ZWVhacnJxw/PjxWgMjhJCWgRJgDf7+/nUe8r6LUChEcHBwnYfFqoZiVZzmFC/FShdCE0LUGN0LTAhRW5QACSFqixIgIURtUQJsgIZOlXXgwAHY2dlBW1sbjo6OiImJaaJIGxbrrl27MHDgQLz33nt477334OLi0qTTgMk6BVlUVBR4PB533WZTaWi8eXl58PPzg5mZGYRCIWxtbZvsu9DQWMPCwtC5c2fo6OjAwsICs2fPRklJicLjPH/+PEaMGAFzc3PweDz8/vvv73xNXFwcevbsCaFQCBsbG0RERDR8x4zUS1RUFBMIBGzPnj3s9u3bbMqUKczQ0JBlZ2fXWT8+Pp7x+XwWGhrK7ty5w7799lumpaXFbt68qXKxjh8/nm3ZsoVdvXqV3b17l3l7ezMDAwP2119/qVys1dLS0ljbtm3ZwIED2WeffabwOKs1NN7S0lLWu3dvNnz4cPbHH3+wtLQ0FhcXx65du6ZysUZGRjKhUMgiIyNZWloaO3HiBDMzM2OzZ89WeKwxMTFs4cKF7LfffmMA2MGDB99a/+HDh0xXV5cFBASwO3fusE2bNjE+n8+OHz/eoP1SAqynvn37Mj8/P+55ZWUlMzc3ZyEhIXXWHzNmDHNzc5Mqc3Z2Zl999ZVC42Ss4bHWVFFRwfT19dmPP/6oqBA5ssRaUVHB+vXrx3bv3s28vLyaNAE2NN5t27axDh06sLKysqYKkdPQWP38/NiQIUOkygICAlj//v0VGmdN9UmA8+fPZ126dJEq8/DwYK6urg3aFx0C10P1VFkuLi5c2bumykpISJCqDwCurq4Kn1pLllhrKi4uRnl5udzvu6xJ1liXLVsGExMT+Pr6KjS+mmSJ9/Dhw5BIJPDz84NYLEbXrl2xcuVKVFZWqlys/fr1Q2JiIneY/PDhQ8TExGD48OEKjVUW8vp90YXQ9dDQqbIAICsrq876WVlZCosTkC3WmgIDA2Fubl7rCyZvssT6xx9/4IcffsC1a9cUGltdZIn34cOHOHPmDCZMmICYmBg8ePAA06dPR3l5OYKDg1Uq1vHjx+PZs2cYMGAAGGOoqKjA1KlT8c033ygsTlm96fdVUFCAly9fQkdHp17tUA+QSFm1ahWioqJw8OBBaGtrKzscKS9evICnpyd27dqFNm3aKDuceqmqqoKJiQl27tyJXr16wcPDAwsXLsT27duVHVotcXFxWLlyJbZu3YqkpCT89ttviI6OxvLly5UdmsJQD7AeGjpVFgCYmpo2qL68yBJrtTVr1mDVqlU4ffo0unXrpsgwATQ81tTUVKSnp2PEa1OnV1VVAQA0NTWRkpKCjh07qky8AGBmZgYtLS3w+XyuzN7eHllZWSgrK4NAIFCZWBctWgRPT09MnjwZAODo6IiioiJ8+eWXWLhwodxnYmmMN/2+RCJRvXt/APUA6+X1qbKqVU+VJZFI6nyNRCKRqg8Ap06demN9ZcYKAKGhoVi+fDmOHz+O3r17KzTGag2N1c7ODjdv3sS1a9e4x6effooPP/wQ165dU/hktLJ8tv3798eDBw+4RA0A9+7dg5mZmcKSn6yxFhcX10py1Ymbqdgds3L7fTVsfEZ9RUVFMaFQyCIiItidO3fYl19+yQwNDVlWVhZjjDFPT0+2YMECrn58fDzT1NRka9asYXfv3mXBwcFNehlMQ2JdtWoVEwgE7JdffmGZmZnc48WLFyoXa01NPQrc0HgzMjKYvr4+8/f3ZykpKezo0aPMxMSEfffddyoXa3BwMNPX12c///wze/jwITt58iTr2LEjGzNmjMJjffHiBbt69Sq7evUqA8DWrVvHrl69yh49esQYY2zBggXM09OTq199Gcy8efPY3bt32ZYtW+gyGEXbtGkTa9++PRMIBKxv377szz//5LYNHjyYeXl5SdXfv38/s7W1ZQKBgHXp0oVFR0erZKyWlpYMQK1HcHCwysVaU1MnQMYaHu/FixeZs7MzEwqFrEOHDmzFihWsoqJC5WItLy9nS5YsYR07dmTa2trMwsKCTZ8+neXm5io8zrNnz9b5HayOz8vLiw0ePLjWa5ycnJhAIGAdOnRg4eHhDd4vzQZDCFFbdA6QEKK2KAESQtQWJUBCiNqiBEgIUVuUAAkhaosSICFEbVECJISoLUqAhBC1RQmQ1BIREQEejwcej4dZs2YpOxyFWrJkCZycnBr0muTkZLz//vvQ1tbmXlv9eRkaGsoUR3p6Ong8XqOn+frggw+k/mZWVlYICwtrVJsA4O3t3eRLDzQFmg1GBX3wwQdwcnKSyxdX1nZFIhFSUlLQqlUrucbQEgQHB6NVq1ZISUmBnp4eACAzMxP79u2TeY4/CwsLZGZmNnqar99++w1aWlqNaqMuGzZskJoQQVHf0aZGCZDUicfjKXzqruYqNTUVbm5usLS05MpMTU1hYGAgc5t8Pl8un7e8Z/GurKwEj8dr1HtTZXQIrGK8vb1x7tw5bNiwgTusSk9PBwDcunUL//rXv6CnpwexWAxPT088e/YMwKvJLAUCAS5cuMC1FRoaChMTE2RnZ7+13fqysrLCypUrMWnSJOjr66N9+/bYuXOnVJ3AwEDY2tpCV1cXHTp0wKJFi1BeXs5trz7k3LNnD9q3bw89PT1Mnz4dlZWVCA0NhampKUxMTLBixQqpdvPy8jB58mQYGxtDJBJhyJAhuH79eoPiB15N+CoWi6Gvrw9fX986VzzbvXs37O3toa2tDTs7O2zdupXbxuPxkJiYiGXLloHH42HJkiX13ndubi4mTJgAY2Nj6OjooFOnTggPDwdQ+xA4Li4OPB4PJ06cQI8ePaCjo4MhQ4YgJycHx44dg729PUQiEcaPH4/i4mJuHzUPgWtat24dHB0d0apVK1hYWGD69OkoLCzktkdERMDQ0BCHDx+Gg4MDhEIhMjIypA6B6/oupaWlwcbGBmvWrJHa37Vr18Dj8fDgwYN6f05NqpGTOBA5y8vLYxKJhE2ZMoWblqqiooLl5uYyY2NjFhQUxO7evcuSkpLYRx99xD788EPutfPmzWOWlpYsLy+PJSUlMYFAwA4dOvTWdusSHh7ODAwMapVbWloyIyMjtmXLFnb//n0WEhLCNDQ0WHJyMldn+fLlLD4+nqWlpbHDhw8zsVjMvv/+e257cHAw09PTY6NGjWK3b99mhw8fZgKBgLm6urIZM2aw5ORktmfPHgZAauYSFxcXNmLECHblyhV27949NmfOHNa6dWv2zz//MMZerRIHgJ09e/aNn+2+ffuYUChku3fvZsnJyWzhwoVMX1+fde/enavzn//8h5mZmbFff/2VPXz4kP3666/MyMiIRUREMMYYy8zMZF26dGFz5sypNWXYmz63an5+fszJyYlduXKFpaWlsVOnTrHDhw9LxX/16lXG2P/NjvL++++zP/74gyUlJTEbGxs2ePBgNmzYMJaUlMTOnz/PWrduzVatWsXtY/Dgwezrr7+W+putX7+ee75+/Xp25swZlpaWxmJjY1nnzp3ZtGnTpN6DlpYW69evH4uPj2fJycmsqKhIatadN32XVqxYwRwcHKTe88yZM9mgQYPe+JkoGyVAFVTzS8zYq8QybNgwqbLHjx8zACwlJYUx9moJRicnJzZmzBjm4ODApkyZ8s526/K2BDhx4kTueVVVFTMxMWHbtm17Y1urV69mvXr14p4HBwczXV1dVlBQwJW5uroyKysrVllZyZV17tyZW73swoULTCQSsZKSEqm2O3bsyHbs2MEYY+yvv/5inTt3ZpcuXXpjLBKJhE2fPl2qzNnZWSoBduzYke3du1eqzvLly5lEIuGed+/evc6pwt6VAEeMGMF8fHzq3PamBHj69GmuTkhICAPAUlNTubKvvvpKaiW0dyXAmg4cOMBat24t9R4A1Fq2s+a0Y3V9l/7++2/G5/O5v0FZWRlr06YN95+HKqJzgM3E9evXcfbsWe6k++tSU1Nha2sLgUCAyMhIdOvWDZaWlli/fv072+3SpQsePXoEABg4cCCOHTv21vqvT5VffZ4wJyeHK9u3bx82btyI1NRUFBYWoqKiAiKRSKoNKysr6Ovrc8/FYjH4fL7UbMRisZhr9/r16ygsLETr1q2l2nn58iVSU1MBAG3btn3nok93797F1KlTpcokEgnOnj0LACgqKkJqaip8fX0xZcoUrk5FRYVczoFNmzYNI0eORFJSEoYNGwZ3d3f069fvra95/fMWi8XcqYXXyxqyiP3p06cREhKC5ORkFBQUoKKiAiUlJSguLoauri6AV7NJy7Ikgrm5Odzc3LBnzx707dsXR44cQWlpKUaPHt3gtpoKJcBmorCwECNGjMD3339fa5uZmRn374sXLwIAnj9/jufPn79zFDcmJoY7R1eftRRqjjDyeDxuuveEhARMmDABS5cuhaurKwwMDBAVFYW1a9e+s423tVtYWAgzMzPExcXVikfWy07qUn0ubNeuXXB2dpba9vqaHrL617/+hUePHiEmJganTp3C0KFD4efnV+u82ete/1ze9Tm9S3p6Oj755BNMmzYNK1asgJGREf744w/4+vqirKyMS4A6Ojrg8XgyvENg8uTJ8PT0xPr16xEeHg4PDw+uXVVECVAFCQSCWuvG9uzZE7/++iusrKygqVn3ny01NRWzZ8/Grl27sG/fPnh5eeH06dNcz6qudl8fyWysixcvwtLSEgsXLuTKqnuXjdGzZ09kZWVBU1MTVlZWMrdjb2+PS5cu4YsvvuDK/vzzT+7fYrEY5ubmePjwISZMmNCYkN/I2NgYXl5e8PLywsCBAzFv3ry3JkB5SkxMRFVVFdauXct9J/bv3y9TW3V9lwBg+PDhaNWqFbZt24bjx4/j/PnzjYpZ0WgUWAVZWVnh0qVLSE9Px7Nnz1BVVQU/Pz88f/4c48aNw5UrV5CamooTJ07Ax8cHlZWVqKysxMSJE+Hq6gofHx+Eh4fjxo0bUr2vutqVp06dOiEjIwNRUVFITU3Fxo0bcfDgwUa36+LiAolEAnd3d5w8eRLp6em4ePEiFi5ciP/+978AgL///ht2dnZvPRz8+uuvsWfPHoSHh+PevXsIDg7G7du3peosXboUISEh2LhxI+7du4ebN28iPDwc69ata/T7WLx4MQ4dOoQHDx7g9u3bOHr0KOzt7Rvdbn3Z2NigvLwcmzZtwsOHD/Hvf/9b5uU53/Rd4vP58Pb2RlBQEDp16qTwRcAaixKgCpo7dy74fD4cHBxgbGyMjIwMmJubIz4+HpWVlRg2bBgcHR0xa9YsGBoaQkNDAytWrMCjR4+wY8cOAK8Oi3fu3Ilvv/2Wu1ykrnbl6dNPP8Xs2bPh7+8PJycnXLx4EYsWLWp0uzweDzExMRg0aBB8fHxga2uLsWPH4tGjR9zi2OXl5UhJSZG6JKQmDw8PLFq0CPPnz0evXr3w6NEjTJs2TarO5MmTsXv3boSHh8PR0RGDBw9GREQErK2tG/0+BAIBgoKC0K1bNwwaNAh8Ph9RUVGNbre+unfvjnXr1uH7779H165dERkZiZCQEJnaett3qfqQ2sfHR16hKwytCUJqiYiIwKxZs5CXl6fsUJoV+txeuXDhAoYOHYrHjx9z/0GpKuoBkjrl5+dDT08PgYGByg6lWdDT06s1wqxuSktL8ddff2HJkiUYPXq0yic/gHqApA4vXrxAdnY2gFejrI29P1UdVN/pwOfz5XK43BxFRETA19cXTk5OOHz4MNq2bavskN6JEiAhRG3RITAhRG1RAiSEqC1KgIQQtUUJkBCitigBEkLUFiVAQojaogRICFFblAAJIWqLEiAhRG39fzrawWrZOt0nAAAAAElFTkSuQmCC",
      "text/plain": [
       "<Figure size 330x250 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(3.3, 2.5))\n",
    "\n",
    "_ = plt.hist(sm_df[text_query_sim_col].tolist(), bins=50, color='b', alpha=0.3)\n",
    "ibins = plt.hist(ic_df[text_query_sim_col].tolist(), bins=50, color='r', alpha=0.3)\n",
    "\n",
    "plt.legend(['LAIONet', 'ImageNet-Captions'])\n",
    "\n",
    "plt.xlabel('text-[name: def] similarity')\n",
    "plt.ylabel('count')\n",
    "\n",
    "plt.xlim([0, 1])\n",
    "\n",
    "plt.tight_layout()\n",
    "\n",
    "plt.savefig(os.path.join('figs', 'text-query_sim_distribution.pdf'), bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Extract text to queries sim. matrix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(443798, 999)\n"
     ]
    }
   ],
   "source": [
    "sim_mat = ic_all_df.loc[\n",
    "    [image_name for image_name in ic_imagename2wnid], \n",
    "    [text_to_query_col_func(wnid) for wnid in all_wnids]\n",
    "].to_numpy()\n",
    "\n",
    "print(sim_mat.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Calc. proportion of more similar false classes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|███████████████████████████████████████████████████████████████████████████████████████████| 443798/443798 [00:13<00:00, 31973.89it/s]\n"
     ]
    }
   ],
   "source": [
    "p_vals = np.zeros((len(ic_imagename2wnid),))\n",
    "\n",
    "for i_i, image_name in enumerate(tqdm(ic_imagename2wnid)):\n",
    "    wnid = ic_imagename2wnid[image_name]\n",
    "    \n",
    "    if wnid not in all_wnids:\n",
    "        continue\n",
    "    \n",
    "    p_vals[i_i] = np.mean(sim_mat[i_i, all_wnids.index(wnid)] < sim_mat[i_i])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAASMAAAD5CAYAAABs8lPQAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjcuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/bCgiHAAAACXBIWXMAAA9hAAAPYQGoP6dpAAAy7klEQVR4nO3deVyU5fo/8M8w7OsosggiiIosgrihgB5cUDwayTllpqnIL6ncSjEzzaLyq1SakaZZplKmuSQtppFIagqiiKIo+6J4DBBcQBZZZq7fHx6e48gM68CMer1fr3m9Zp65n/u57tG5eLb7GhERERhjTM201B0AY4wBnIwYYxqCkxFjTCNwMmKMaQRORowxjcDJiDGmETgZMcY0AicjxphG4GTEGNMInIwYYxpBrcnor7/+QmBgIGxsbCASifDzzz83u87x48cxaNAg6OnpoU+fPoiKiurwOBljHU+tyaiyshIDBgzApk2bWtQ+Pz8fkyZNwujRo5GSkoJFixZhzpw5+OOPPzo4UsZYRxNpykRZkUiEn376CUFBQUrbLFu2DIcOHcLly5eFZS+++CLu3r2LmJiYToiSMdZRtNUdQGucPn0a/v7+cssCAgKwaNEipevU1NSgpqZGeC2TyXD79m2Ym5tDJBJ1VKiMPbGICPfu3YONjQ20tFR3cPVYJaOioiJYWVnJLbOyskJ5eTmqq6thYGDQaJ2IiAh88MEHnRUiY0+N69evo0ePHirr77FKRm2xfPlyhIWFCa/LysrQs2dPXL9+Haampp0WR21tLT799FMAwJIlS6Crq9tp22ZMlcrLy2FnZwcTExOV9vtYJSNra2sUFxfLLSsuLoapqanCvSIA0NPTg56eXqPlpqamnZ6MGuIwNTXlZMQee6o+zfFYJSNvb28cPnxYbllsbCy8vb3VFFHL6ejoCOe2dHR01BsMYxpIrZf2KyoqkJKSgpSUFAAPLt2npKSgoKAAwINDrFmzZgntX3vtNeTl5eGtt95CRkYGNm/ejH379mHx4sXqCL9VRCIRJBIJJBIJnzhnTAG17hmdO3cOo0ePFl43nNsJDg5GVFQUCgsLhcQEAL169cKhQ4ewePFifP755+jRowe++eYbBAQEdHrsjD2upFIpqqqqcO3aNYjFYlRWVqKsrAxVVVWorKxERUUFqqurcffuXVRXV6O6uhr37t1DVVUVbt++jfLy8g6JS2PuM+os5eXlMDMzQ1lZWaeeM5JKpYiLiwMAjB07FmKxuNO2zR5vRITa2lrcuHEDAFBVVYWysjJUVFTg7Nmz6NGjB/766y/06tULR48ehaOjI6qrq3Hnzh1UVVXh/v37SE5OhpWVVaNzrsro6OjA0NAQEokEBgYG0NfXh6mpKQwNDWFkZIQDBw6o/Dv0WJ0zepxJpVIkJCQAAEaNGsXJ6ClQX1+P6upq3Lp1C9XV1cjOzoaBgQEuXrwIiUSCuro6HDt2DH379kVMTAxcXV1x6NAh9O/fH+fOnYOenh7u3r3bqm06Ojri5MmTGDt2LEQiEcrKyuDj44MRI0ZAIpHAxMQEubm58PPzg76+PsrKyuDh4YHa2lrY2trC2NgYJiYmTd4/VF5ejgMHDrTz02mMkxFjStTX16OyshJXr17FnTt3cOfOHaSkpMDW1hZxcXHo3bs3YmJi4OLigt9//x3Ozs44ffo0LC0tcfPmzWb7F4vFICLIZDI4OTlh3759GDNmDLKysjBr1iwUFBTA3d0dVVVV6NevH7p3747i4mIMGjQI9fX1sLS0RNeuXWFmZvZEXBThZMSeaDKZDH///Tfu3buH9PR0iMViJCYmwtLSEsePH0fv3r1x8OBBuLm54eDBg+jfvz8uXboEY2NjVFRUKO1XLBZDKpWif//+2Lt3LyZMmIC0tDQsWLAAWVlZGD9+PPLz8zFy5EgUFhZi6NChMDAwgImJCUxMTGBgYAAzM7NO/CQ0Hycj9liRSqUoKipCUVERkpOTUVZWhrKyMpw5cwZaWlrCc0dHR+Tl5TXZl7a2Nurr69G/f38kJiYiMDAQN27cwOLFi3Hjxg34+vqipKQEw4cPx71792Bvbw9HR0dIJBKF966x9uFkxNSKiFBWVob8/HzU1tYiISEB//nPf2Bubo5ffvkFJSUl6NatG5KSkprsR09PDzU1NejVqxfGjBmD+/fvIyQkBAUFBRg7dixu3boFNzc3WFlZwcbGBsbGxiqdV8Xaj5MR6zClpaVIS0vD3bt3kZubi5SUFOTn50NfXx9JSUnNnpzt3r07KioqIJVKMXfuXBQVFcHa2hqjRo1CVVUVhg4dColEgu7du3NieQJwMmKtJpPJUFBQgJycHJw7dw4ymQyxsbHIz8/HtWvXWtSHt7c3zM3NYWRkhBEjRsDAwAASiQQeHh6wsrLq1NsumGbgZNRJdHR0MG/ePOG5Jquvr8eVK1eQlJSEixcv4tKlS8jNzRXuc1Gka9euuH37NmxtbWFvb49+/fph0KBBqKurg7OzMzw8PGBtbc23NDClOBl1EpFIBEtLS3WHISgrK0NycjISExNx7tw5ZGRk4Nq1a6iqqlK6jp2dHfr16wdra2sEBAQIl6Td3d1haGjI01xYu3AyekJJpVLk5+fjzJkzSE5ORlpaGvLy8pCXlwepVKpwnW7dusHNzQ29evWCq6srhgwZgqFDh8LCwoITDetwnIw6iVQqxcmTJwEAI0eOVNnhyt27dxEfH4/ExESkpKQgIyMDOTk5SttbWlrCyckJffv2haenJ7y9veHh4cGXqpnacTLqJFKpFMePHwcA+Pj4tDoZSaVSJCcn4+jRozhx4gQuXbqEoqIihW179OgBDw8PuLu7w8vLC0OHDoWtrS1fcWIajZORhqqsrMSuXbtw5MgRSKVShT/j1KdPHwwcOBC+vr7w8/ODu7s7nyBmjy1ORhri9u3biImJwaZNm4QJtQ+ztLTExx9/jB49esDLy4svfbMnDiej/6qtrVX6npaWFrS1tVvUViQSyV26b2hbW1srnDhueH727FlERUUhOjparkbMsGHDMGjQILi5ueH5559Hly5dGvVbV1cHZdVfWtMWgFwJ3Na0ra+vh0wmU0lbHR0d4SR5R7WVSqVKT963tq22trZw2KsJbWUyGerr65W2FYvFwl5ze9s29f+/PTgZ/deaNWuUvte3b1+89NJLwuu1a9eirq5OYVsHBwfMnj1beB0ZGYmqqipIpVKcOHECpaWl+O677+SKxjk4OGDRokWoqamBnp6e8IUoKSnBl19+CQCwsLDA/PnzhXW+/vprlJSUKIxBIpHI/XzTjh078Pfffytsa2hoiLfeekt4vWvXLly9elVhWx0dHbzzzjvC67179yI7O1thWwB4//33hefR0dFIS0tT2nbFihVC8vrtt9+E6p+KLF26FEZGRgCAP/74o8mpIosWLYJEIgEAxMXFKdzrbDBv3jzh9ouTJ08K5/gUCQ0Nha2tLQAgMTERsbGxStvOnj0bDg4OAIDk5ORGpZMfNn36dDg5OQEAUlNTm/yV5SlTpsDNzQ0AkJ6ejv379yttGxQUBE9PTwBATk4Odu/erbTtxIkT4eXlBQAoKCho9KvND//0lypxMupgMpkM6enpiI+Pl7tp0N7eHhMmTMCGDRuEL2FkZGSr69cw9qTgSo//perDtBs3biA0NBS///678F6vXr0wYsQIbN68Gbq6uh126MWHaYrb8mGaag7TysvLYWFhwZUeO0prfjqoqbYZGRmYPHkysrKyAACDBw/Gxx9/jNGjR6OwsBDAg0MjRZfZWzNNRBPaPpygH4e2D3/JnrS2WlpaLf4/3N62HfUzW5yMVOT+/ft45plnhDrXU6ZMwVdffYUuXboIbRrOMTDGGuO74FRgw4YNMDAwQFxcHIYNG4by8nLs27dPLhExxprGe0btUFxcDBcXF9y5cwfAgyslgwYNUthWKpUiMTERADB8+HC+OZGxR/CeURtFRUXB2toad+7cwZIlS0BEShMR8CAZxcbGIjY2tsmTkow9rXjPqA0WL16MyMhIAEBeXh569eql3oAYewJwMmqlWbNmYefOnTA3N0dRUVGrruYwxpTjw7RWeP3117Fz5044OTmhpKSEExFjKsTJqIWOHDmCjRs3wtzcHBkZGVxsjDEV42TUAnfv3kVAQACAB+eIOBExpnqcjFrg+eefB/Bg74hLdzDWMfikRzMuX74s3Mw4bty4Nvejra0tzObnc02MNcbfima8+uqrANBkeYaW0NLSEspIMMYa48O0Jly9ehUJCQl49tlnYWdnp+5wGHui8Z5RE9auXQsAWL16dbv7aiioDzyYyc/TQRiTx8lICalUis2bN6NHjx7o37+/SvprqPDn6enJyYixR/BhmhINk1rffvttNUfC2NOBk5ESGzZsAAAEBgaqORLGng6cjJQ4ffo0PDw80LNnT3WHwthTgZORAqmpqbh+/TrGjBmj7lAYe2pwMlKg4SeA5s6dq+ZIGHt6cDJS4LPPPgMAmJiYqDkSxp4efGlfgbKyMgwfPhzdu3dXWZ/a2tqYPn268JwxJo/3jB5RU1ODkydPok+fPirtV0tLC05OTnByclL4M0WMPe34W/GIU6dOAQAmTZqk5kgYe7rw8YISw4cPV2l/UqkUqampAAB3d3e+A5uxR6h9z2jTpk1wcHCAvr4+hg0bhrNnzyptW1dXhw8//BC9e/eGvr4+BgwYgJiYGJXGU1xcrNL+GkilUvz888/4+eef+ddBGFNArclo7969CAsLQ3h4OM6fP48BAwYgICAAN2/eVNh+5cqV+Oqrr7Bx40akpaXhtddew7/+9S9cuHBBZTF98cUXAICuXbuqrE/GWPPUmozWr1+P0NBQhISEwNXVFVu2bIGhoSG2b9+usP3OnTuxYsUKTJw4EY6Ojpg7dy4mTpyITz/9VGUxERFmzJjBFR0Z62RqS0a1tbVITk6Gv7///4LR0oK/vz9Onz6tcJ2amhro6+vLLTMwMBBOOitbp7y8XO7RFJFIBB0dnVaMhDGmCmpLRqWlpZBKpbCyspJbbmVlhaKiIoXrBAQEYP369cjOzoZMJkNsbCyio6NRWFiodDsREREwMzMTHs0VSbt161brB8MYaze1n8Bujc8//xx9+/aFs7MzdHV1sWDBAoSEhDR5387y5ctRVlYmPK5fv6607ZUrV5CVlcWHaIypgdqSUbdu3SAWixtdvSouLoa1tbXCdSwsLPDzzz+jsrIS165dQ0ZGBoyNjeHo6Kh0O3p6ejA1NZV7KNOQqBYtWtT6ATHG2kVtyUhXVxeDBw9GXFycsEwmkyEuLg7e3t5Nrquvrw9bW1vU19fjwIEDmDx5skpj64jpGtra2pgyZQqmTJnC00EYU0Ct34qwsDAEBwdjyJAh8PLyQmRkJCorKxESEgLgwe/a29raIiIiAgBw5swZ3LhxA56enrhx4wbef/99yGQyvPXWW+ocRotoaWnBzc1N3WEwprHUmoymTp2KkpISvPfeeygqKoKnpydiYmKEk9oFBQVy54Pu37+PlStXIi8vD8bGxpg4cSJ27twJiUSiphEwxlRFRESk7iA6U3l5OczMzFBWVtbo/NG+ffswdepUXL9+HT169FDpdmUyGdLT0wEALi4uPFmWPbaa+g61B38jHtKRdYzq6+uxf/9+7N+/H/X19Srvn7HHHSejh9TV1eH555+HmZmZukNh7KnDyeghWlpaPCeNMTXhZMQY0wicjBhjGoGTEWNMI3AyYoxpBJ6X0EnEYjGCgoKE54wxeZyMHtJcraP2EIvF8PT07LD+GXvc8WHaf12+fBmZmZn8w42MqQnvGf1Xfn4+gI4rHyKTyZCTkwMA6NOnD08HYewR/I14REeVnK2vr8fu3buxe/dung7CmAItTkb//ve/hXMq3333HWpqajosKMbY06fFyei3335DZWUlACAkJARlZWUdFhRj7OnT4nNGzs7OWL58OUaPHg0iwr59+5SWD5g1a5bKAmSMPR1anIy2bNmCsLAwHDp0CCKRCCtXroRIJGrUTiQScTJijLVai5ORj48PEhMTATyY3Z6VlQVLS8sOC4wx9nRp09W0/Px8WFhYqDoWxthTrMV7RpcuXZJ7nZqaqrSth4dH2yN6QonFYkycOFF4zhiT1+Jk5OnpCZFIBCJSeK7oYVKptN2BPWnEYjG8vLzUHQZjGqvFh2n5+fnIy8tDfn4+Dhw4gF69emHz5s24cOECLly4gM2bN6N37944cOBAR8bLGHtCtXjPyN7eXng+ZcoUbNiwQTjsAB4cmtnZ2eHdd98VZqez/5HJZCgoKAAA9OzZk6eDMPaINn0jUlNT0atXr0bLe/XqhbS0tHYH9SSqr69HVFQUoqKieDoIYwq0KRm5uLggIiICtbW1wrLa2lpERETAxcVFZcExxp4ebZq1v2XLFgQGBqJHjx7ClbOGq22//fab6qJjjD012pSMvLy8kJeXh127diEjIwPAg5+qnj59OoyMjFQaIGPs6dCmZBQREQErKyu88sorcsu3b9+OkpISLFu2TCXBdaZNmzYB6LgSIoyxprXpnNFXX30FZ2fnRsvd3NywZcuWdgelDoWFhfDy8uIfcWRMTdqUjIqKitC9e/dGyy0sLFBYWNjuoNRBS0sLQ4cOVXcYjD212nSYZmdnh/j4+EaX9+Pj42FjY6OSwJ40YrEY48aNE54zxuS1KRmFhoZi0aJFqKurw5gxYwAAcXFxeOutt7BkyRKVBvikEIvF8PX1VXcYjGmsNiWjpUuX4tatW5g3b55wr5G+vj6WLVuG5cuXqzRAxtjToU3JSCQS4eOPP8a7776L9PR0GBgYoG/fvtDT01N1fE8MmUwmnE/r3r07Twdh7BHt+qkiY2NjPunbQvX19di6dSsAYMWKFdDV1VVzRIxpFv7zzBjTCJyMGGMagZMRY0wjcDJijGkETkaMMY3AyYgxphHadWmftZxYLMaoUaOE54wxeZyMOsnDyYgx1hgfpjHGNALvGXUSIkJJSQmAB6VWmvvtOcaeNmrfM9q0aRMcHBygr6+PYcOG4ezZs022j4yMRL9+/WBgYAA7OzssXrwY9+/f76Ro266urg6bN2/G5s2bUVdXp+5wGNM4ak1Ge/fuRVhYGMLDw3H+/HkMGDAAAQEBuHnzpsL2u3fvxttvv43w8HCkp6dj27Zt2Lt3L1asWNHJkTPGVE2tyWj9+vUIDQ1FSEgIXF1dsWXLFhgaGmL79u0K2yckJMDX1xfTp0+Hg4MDxo8fj2nTpjW7N8UY03xqS0a1tbVITk6Gv7///4LR0oK/vz9Onz6tcB0fHx8kJycLyScvLw+HDx+W+2XbR9XU1KC8vFzuwRjTPGo7gV1aWgqpVAorKyu55VZWVsLPHz1q+vTpKC0txYgRI0BEqK+vx2uvvdbkYVpERAQ++OADlcbOGFM9tZ/Abo3jx49jzZo12Lx5M86fP4/o6GgcOnQIq1atUrrO8uXLUVZWJjyuX7/eiREzxlpKbXtG3bp1g1gsRnFxsdzy4uJiWFtbK1zn3XffxcyZMzFnzhwAgLu7OyorK/HKK6/gnXfeUVg9UU9PjytQMvYYUNueka6uLgYPHoy4uDhhmUwmQ1xcHLy9vRWuU1VV1SjhNEytIKKOC1YFxGIxfHx84OPjw9NBGFNArTc9hoWFITg4GEOGDIGXlxciIyNRWVmJkJAQAMCsWbNga2uLiIgIAEBgYCDWr1+PgQMHYtiwYcjJycG7776LwMBAjf+Ci8VijB8/Xt1hMKax1JqMpk6dipKSErz33nsoKiqCp6cnYmJihJPaBQUFcntCK1euhEgkwsqVK3Hjxg1YWFggMDAQq1evVtcQGGMqIiJNP75RsfLycpiZmaGsrAympqbC8oEDB8LX1xdffPFFh2yXiFBWVgYAMDMz4+kg7LGl7DvUXo/V1bTHWV1dHSIjIxEZGcnTQRhTgJMRY0wjcDJijGkETkaMMY3AyYgxphE4GTHGNAInI8aYRuCys51ES0sLQ4cOFZ4zxuRxMuok2tramDRpkrrDYExj8Z9oxphG4D2jTkJEqKqqAgAYGhrydBDGHsF7Rp2krq4Oa9euxdq1a3k6CGMKcDJijGkETkaMMY3AyQgPSiKkpKSoOwzGnmqcjACh9O20adPUHAljTy9ORg9xcXFRdwiMPbU4GTHGNALfZ9RJtLS04OnpKTxnjMnjZNRJtLW1ERQUpO4wGNNY/CeaMaYReM+okxCRcOe1jo4OTwdh7BG8Z9RJ6urqsGbNGqxZs4angzCmACcjxphG4GTEGNMInIwYYxqBkxFjTCNwMmKMaQRORowxjcD3GXUSLS0tuLq6Cs8ZY/I4GXUSbW1tvPDCC+oOgzGNxX+iGWMagZMRY0wj8GFaJ6mtrcWaNWsAACtWrICurq6aI2JMs/CeEWNMI3AyYoxpBE5GjDGNwMmIMaYROBkxxjQCJyPGmEbgS/udREtLC3379hWeM8bkcTLqJNra2njppZfUHQZjGov/RDPGNAInI8aYRtCIZLRp0yY4ODhAX18fw4YNw9mzZ5W2HTVqFEQiUaPHpEmTOjHi1qutrcXq1auxevVq1NbWqjscxjSO2pPR3r17ERYWhvDwcJw/fx4DBgxAQEAAbt68qbB9dHQ0CgsLhcfly5chFosxZcqUTo689erq6vhnihhTQu3JaP369QgNDUVISAhcXV2xZcsWGBoaYvv27Qrbd+3aFdbW1sIjNjYWhoaGj0UyYowpp9arabW1tUhOTsby5cuFZVpaWvD398fp06db1Me2bdvw4osvwsjISOH7NTU1qKmpEV6XlZUBAMrLy4VlVVVVwjJt7Y75SGpra4U4ysvLedY+e2w1fHeISLUdkxrduHGDAFBCQoLc8qVLl5KXl1ez6585c4YA0JkzZ5S2CQ8PJwD84Ac/VPzIzc1tdw542GN9n9G2bdvg7u4OLy8vpW2WL1+OsLAw4bVMJsPt27dhbm4u/N59eXk57OzscP36dZiamnZ43B2Jx6KZnqSxlJWVoWfPnujatatK+1VrMurWrRvEYjGKi4vllhcXF8Pa2rrJdSsrK7Fnzx58+OGHTbbT09ODnp6e3DKJRKKwramp6WP/H6UBj0UzPUljUfVMArWewNbV1cXgwYMRFxcnLJPJZIiLi4O3t3eT6+7fvx81NTWYMWNGR4fJGOsEaj9MCwsLQ3BwMIYMGQIvLy9ERkaisrISISEhAIBZs2bB1tYWERERcutt27YNQUFBMDc3V0fYjDEVU3symjp1KkpKSvDee++hqKgInp6eiImJgZWVFQCgoKCg0e5gZmYmTp06hSNHjqgkBj09PYSHhzc6nHsc8Vg0E4+leSIiVV+fY4yx1lP7TY+MMQZwMmKMaQhORowxjcDJiDGmEZ6aZNSaMiXAg/uYnJ2doa+vD3d3dxw+fLiTIm1ea8Zy5coVPPfcc3BwcIBIJEJkZGTnBdoCrRnL1q1bMXLkSHTp0gVdunSBv79/s/+Onak1Y4mOjsaQIUMgkUhgZGQET09P7Ny5sxOjbVprvy8N9uzZA5FIhKCgoNZvVKWTSzTUnj17SFdXl7Zv305Xrlyh0NBQkkgkVFxcrLB9fHw8icVi+uSTTygtLY1WrlxJOjo6lJqa2smRN9basZw9e5befPNN+uGHH8ja2po+++yzzg24Ca0dy/Tp02nTpk104cIFSk9Pp9mzZ5OZmRn95z//6eTIG2vtWI4dO0bR0dGUlpZGOTk5FBkZSWKxmGJiYjo58sZaO5YG+fn5ZGtrSyNHjqTJkye3ertPRTLy8vKi+fPnC6+lUinZ2NhQRESEwvYvvPACTZo0SW7ZsGHD6NVXX+3QOFuitWN5mL29vUYlo/aMhYiovr6eTExM6Ntvv+2oEFusvWMhIho4cCCtXLmyI8JrlbaMpb6+nnx8fOibb76h4ODgNiWjJ/4wraFMib+/v7CsuTIlp0+flmsPAAEBAS0ua9JR2jIWTaWKsVRVVaGurk7lEzZbq71jISLExcUhMzMT//jHPzoy1Ga1dSwffvghLC0t8fLLL7d522q/A7ujlZaWQiqVCnd0N7CyskJGRobCdYqKihS2Lyoq6rA4W6ItY9FUqhjLsmXLYGNj0+gPR2dr61jKyspga2uLmpoaiMVibN68GePGjevocJvUlrGcOnUK27ZtQ0pKSru2/cQnI/Zk+uijj7Bnzx4cP34c+vr66g6nTUxMTJCSkoKKigrExcUhLCwMjo6OGDVqlLpDa7F79+5h5syZ2Lp1K7p169auvp74ZNSWMiXW1tZtKmvS0dpTckXTtGcs69atw0cffYSjR4/Cw8OjI8NskbaORUtLC3369AEAeHp6Ij09HREREWpNRq0dS25uLq5evYrAwEBhmUwmA/DgtwIzMzPRu3fvFm37iT9n1JYyJd7e3nLtASA2NrbZsiYdrT0lVzRNW8fyySefYNWqVYiJicGQIUM6I9RmqerfRSaTyZVIVofWjsXZ2RmpqalISUkRHs8++yxGjx6NlJQU2NnZtXzjbTjZ/tjZs2cP6enpUVRUFKWlpdErr7xCEomEioqKiIho5syZ9Pbbbwvt4+PjSVtbm9atW0fp6ekUHh6uUZf2WzOWmpoaunDhAl24cIG6d+9Ob775Jl24cIGys7PVNQRBa8fy0Ucfka6uLv34449UWFgoPO7du6euIQhaO5Y1a9bQkSNHKDc3l9LS0mjdunWkra1NW7duVdcQBK0dy6PaejXtqUhGREQbN26knj17kq6uLnl5eVFiYqLwnp+fHwUHB8u137dvHzk5OZGuri65ubnRoUOHOjli5Vozlvz8fIX1i/38/Do/cAVaMxZ7e3uFYwkPD+/8wBVozVjeeecd6tOnD+nr61OXLl3I29ub9uzZo4aoFWvt9+VhbU1GXEKEMaYRnvhzRoyxxwMnI8aYRuBkxBjTCJyMGGMagZMRY0wjcDJijGkETkaMMY3AyYgxphE4GT3Fjh8/DpFIhLt376o7FMTHx8Pd3R06OjpKS5ZWVVXhueeeg6mpaYvjVvcYR40ahUWLFqll24+bJ37WPntg1KhR8PT0lKuB7ePjg8LCQpiZmakvsP8KCwuDp6cnfv/9dxgbGyts8+233+LkyZNISEhAt27dNCJupjq8Z9QJamtrNXLburq6sLa2hkgk6sSIFMvNzcWYMWPQo0cPSCQSpW1cXFzQv39/jYmbqVCrZ7M95fz8/Gj+/Pk0f/58MjU1JXNzc1q5ciXJZDKhjb29PX344Yc0c+ZMMjExESYV/vjjj+Tq6kq6urpkb29P69atk+u7Yb0XX3yRDA0NycbGhr744gu5NteuXaNnn32WjIyMyMTEhKZMmSLMpiYiCg8PpwEDBtDWrVvJwcGBRCIRBQcHN5pcmp+fT8eOHSMAdOfOHWH9lsS4evVqCgkJIWNjY7Kzs6Ovvvqqyc/s/v37tHDhQrKwsCA9PT3y9fWls2fPEpHiibw7duxQ+LlDwUTf7777jgYPHkzGxsZkZWVF06ZNkysc/+gYr169Ss888wxJJBIyNDQkV1dXuUnQqampNGHCBDIyMiJLS0uaMWMGlZSUNDm+U6dOkZ+fHxkYGJBEIqHx48fT7du3hbjfeOMNoW1z8d6+fZumT59O3bp1I319ferTpw9t376diB5UYJg/fz5ZW1uTnp4e9ezZk9asWSOse+fOHXr55ZepW7duZGJiQqNHj6aUlBTh/ZSUFBo1ahQZGxuTiYkJDRo0iJKSkpocW2fiZNRKfn5+ZGxsTG+88QZlZGTQ999/T4aGhvT1118Lbezt7cnU1JTWrVtHOTk5lJOTQ+fOnSMtLS368MMPKTMzk3bs2EEGBgZyXzx7e3syMTGhiIgIyszMpA0bNpBYLKYjR44Q0YPC6J6enjRixAg6d+4cJSYm0uDBg+Vm4IeHh5ORkRFNmDCBzp8/TxcvXqS7d++St7c3hYaGCmU36uvrG31RWxpj165dadOmTZSdnU0RERGkpaVFGRkZSj+z119/nWxsbOjw4cN05coVCg4Opi5dutCtW7eovr6eCgsLydTUlCIjI6mwsJCqqqoa9XHr1i0KDQ0lb29vKiwspFu3bhER0bZt2+jw4cOUm5tLp0+fJm9vb/rnP/8prPfoGCdNmkTjxo2jS5cuUW5uLh08eJBOnDhBRA++zBYWFrR8+XJKT0+n8+fP07hx42j06NFKx3bhwgXS09OjuXPnUkpKCl2+fJk2btwoJLBHk1Fz8c6fP588PT0pKSmJ8vPzKTY2ln799VciIlq7di3Z2dnRX3/9RVevXqWTJ0/S7t27hXX9/f0pMDCQkpKSKCsri5YsWULm5ubCZ+Xm5kYzZsyg9PR0ysrKon379sklK3XjZNRKfn5+5OLiIrcntGzZMnJxcRFe29vbU1BQkNx606dPp3HjxsktW7p0Kbm6usqtN2HCBLk2U6dOFf6zHjlyhMRiMRUUFAjvX7lyhQAIexoNtZdu3rzZKO6HvxREjb+oLY1xxowZwmuZTEaWlpb05ZdfkiIVFRWko6NDu3btEpbV1taSjY0NffLJJ8IyMzMzhXtED3vjjTeaLX2SlJREAIQaR4+O0d3dnd5//32F665atYrGjx8vt+z69esEgDIzMxWuM23aNPL19VUaj6LPval4AwMDKSQkRGHbhQsX0pgxY+T+7zU4efIkmZqa0v379+WW9+7dW9hzNTExoaioKKWxqBufM2qD4cOHy52v8Pb2RnZ2NqRSqbDs0SqE6enp8PX1lVvm6+vbaL1Hq+l5e3sjPT1d6MPOzk6uep6rqyskEonQBgDs7e1hYWHR6nG1NMaHS72KRCJYW1vj5s2bCvvMzc1FXV2dXL86Ojrw8vKSi7mtkpOTERgYiJ49e8LExAR+fn4AgIKCAoXtX3/9dfzf//0ffH19ER4ejkuXLgnvXbx4EceOHYOxsbHwcHZ2FsahSEpKCsaOHauyeOfOnYs9e/bA09MTb731FhISEoR1Z8+ejZSUFPTr1w+vv/46jhw5Ihd7RUUFzM3N5eLPz88XYg8LC8OcOXPg7++Pjz76SOmY1IWTUQcxMjJ6Yreto6Mj91okEgl1jztTZWUlAgICYGpqil27diEpKQk//fQTAOUn7ufMmYO8vDzMnDkTqampGDJkCDZu3AgAqKioQGBgoFwJ1ZSUFGRnZyv9CSEDAwOVxvvPf/4T165dw+LFi/H3339j7NixePPNNwEAgwYNQn5+PlatWoXq6mq88MILeP7554XYu3fv3ij2zMxMLF26FADw/vvv48qVK5g0aRL+/PNPuLq6CtvXBJyM2uDMmTNyrxMTE9G3b1+IxWKl67i4uCA+Pl5uWXx8PJycnOTWS0xMbNS3i4uL0Mf169dx/fp14f20tDTcvXsXrq6uTcasq6srt3fTnhhbo3fv3tDV1ZXrt66uDklJSc3G3JyMjAzcunULH330EUaOHAlnZ2ele2gPs7Ozw2uvvYbo6GgsWbIEW7duBfDgy37lyhU4ODigT58+cg9lCd7Dw6NRvfT2xmthYYHg4GB8//33iIyMxNdffy28Z2pqiqlTp2Lr1q3Yu3cvDhw4gNu3b2PQoEEoKiqCtrZ2o9gf/tUOJycnLF68GEeOHMG///1v7Nixo0WxdwZORm1QUFCAsLAwZGZm4ocffsDGjRvxxhtvNLnOkiVLEBcXh1WrViErKwvffvstvvjiC+GvXoP4+Hh88sknyMrKwqZNm7B//36hb39/f7i7u+Oll17C+fPncfbsWcyaNQt+fn7NFqd3cHDAmTNncPXqVZSWlirck2lpjK1hZGSEuXPnYunSpYiJiUFaWhpCQ0NRVVXVrh/8A4CePXtCV1cXGzduRF5eHn799VesWrWqyXUWLVqEP/74A/n5+Th//jyOHTsmJPv58+fj9u3bmDZtGpKSkpCbm4s//vgDISEhShP58uXLkZSUhHnz5uHSpUvIyMjAl19+idLS0jbF+9577+GXX35BTk4Orly5gt9++02Ib/369fjhhx+QkZGBrKws7N+/H9bW1pBIJPD394e3tzeCgoJw5MgRXL16FQkJCXjnnXdw7tw5VFdXY8GCBTh+/DiuXbuG+Ph4JCUlCX1rBHWftHrc+Pn50bx58+i1114jU1NT6tKlC61YsaLRpX1FPyPdcNlcR0eHevbsSWvXrpV7397enj744AOaMmUKGRoakrW1NX3++edybVp6af9RmZmZNHz4cDIwMGjRpf2mYnx0bAMGDGiyDnV1dTUtXLiQunXr1ujSfoO2nsDevXs3OTg4kJ6eHnl7e9Ovv/5KAOjChQtE1PgE9oIFC6h3796kp6dHFhYWNHPmTCotLRX6y8rKon/9618kkUjIwMCAnJ2dadGiRQpPGjc4fvw4+fj4kJ6eHkkkEgoICBC29+gJ7ObiXbVqFbm4uJCBgQF17dqVJk+eTHl5eURE9PXXX5OnpycZGRmRqakpjR07ls6fPy/0XV5eTgsXLiQbGxvS0dEhOzs7eumll6igoIBqamroxRdfJDs7O9LV1SUbGxtasGABVVdXN/mZdyaugd1Kiu5kVhUHBwcsWrSIpw+wpxIfpjHGNAInI8aYRuDDNMaYRuA9I8aYRuBkxBRycHBo90n6999/H56ensLr2bNnK61VpMmuXr0KkUiElJQUdYfyRONkxBRKSkrCK6+80q4+3nzzzRbfENheIpEIP//8c7v76cyE+Wiy1vR+OxoXV3vMSaVSiEQiaGmp9u9KW+a2PaphfpQq1dXVNZqOwp4Q6r3N6cm1Y8cOMjMzo4MHD5KTkxMZGBjQc889R5WVlRQVFUX29vYkkUho4cKFVF9fL6x3+/ZtmjlzpnDT3YQJEygrK6tRv7/88gu5uLiQWCym/Px8un//Pi1ZsoRsbGzI0NCQvLy86NixY0rjk8lkFB4eLtwE1717d1q4cKHw/qM3NwKgLVu20KRJk4SbARMSEig7O5v8/PzI0NCQvL29KScnR1jn0Rswg4ODafLkycLr33//nXx9fcnMzIy6du1KkyZNklu/odbRnj176B//+Afp6ekpvDHS3t5ertaRvb298N7mzZvJ0dGRdHR0yMnJib777juln0l4eHij2krHjh0T4jhw4ACNGjWKDAwMyMPDgxISEuTWP3nyJI0YMYL09fWpR48etHDhQqqoqFC4rR07diit49RUXaKbN2+SlZUVrV69WugrPj6edHR06OjRo032q+k4GXWQHTt2kI6ODo0bN47Onz9PJ06cIHNzcxo/fjy98MILdOXKFTp48CDp6urSnj17hPWeffZZcnFxob/++otSUlIoICCA+vTpQ7W1tXL9+vj4UHx8PGVkZFBlZSXNmTOHfHx86K+//qKcnBxau3Yt6enpySWyh+3fv59MTU3p8OHDdO3aNTpz5kyjmkyPJiNbW1vau3cvZWZmUlBQEDk4ONCYMWMoJiaG0tLSaPjw4XIlUJpLRj/++CMdOHCAsrOz6cKFCxQYGEju7u4klUqJ6H/JyMHBgQ4cOEB5eXn0999/NxrLzZs3hS9dYWGhUD4lOjqadHR0aNOmTZSZmUmffvopicVi+vPPPxV+Jvfu3aMXXniBJkyYINR9qqmpEeJwdnam3377jTIzM+n5558ne3t7qqurIyKinJwcMjIyos8++4yysrIoPj6eBg4cSLNnz1a4raqqKlqyZAm5ubkJ22qo49RcXaJDhw6Rjo4OJSUlUXl5OTk6OtLixYub7VfTcTLqIA1/oR7+S//qq6+SoaGhULuGiCggIIBeffVVInowFQEAxcfHC++XlpaSgYEB7du3T67fh4tiXbt2jcRiMd24cUMuhrFjx9Ly5csVxvfpp5+Sk5OTkOQepSgZrVy5Unh9+vRpAkDbtm0Tlv3www+kr68vvG4uGT2qpKSEAFBqaioR/S8ZRUZGKl3n4fh++uknuWU+Pj4UGhoqt2zKlCk0ceJEpf0oirEhjm+++UZY1lBHKj09nYiIXn75ZXrllVfk1jt58iRpaWkpnXKhaOpOS+oSERHNmzePnJycaPr06eTu7i7XXtmUIE3HJ7A7kKGhIXr37i28trKygoODg9x5FCsrK2Hmdnp6OrS1tTFs2DDhfXNzc/Tr10+u9o+urq5cTaHU1FRIpVI4OTnJ1bI5ceKE0po1U6ZMQXV1NRwdHREaGoqffvoJ9fX1TY7n4W1aWVkBANzd3eWW3b9/H+Xl5U320yA7OxvTpk2Do6MjTE1N4eDgAKBxLaLmJgEro6w+U1vrKD08/u7duwOA8G938eJFREVFyX3+AQEBkMlkyM/Pb/E2WlKXCADWrVuH+vp67N+/H7t27YKenl6bxqRJ+AR2B1JU90cVtYAMDAzkirtVVFRALBYjOTm5UakPZSeQ7ezskJmZiaNHjyI2Nhbz5s3D2rVrceLECaUniB9e3rB9RctaOp7AwEDY29tj69atsLGxgUwmQ//+/RvVIlJnbaiHNTXWiooKvPrqq3j99dcbrdezZ88Wb6OhLtHx48cbvffwDxXk5ubi77//hkwmw9WrV+X+KDyuOBlpEBcXF9TX1+PMmTPw8fEBANy6dQuZmZlN1v4ZOHAgpFIpbt68iZEjR7Z4ewYGBggMDERgYCDmz58PZ2dnpKamYtCgQe0eS3MaxrV161Yh5lOnTrW5Px0dnUZlPhrqMwUHBwvL4uPjm/wsW1L3SZFBgwYhLS0Nffr0afE6irb1cF2ihj3FR9XW1mLGjBmYOnUq+vXrhzlz5iA1NRWWlpbtGoO68WGaBunbty8mT56M0NBQnDp1ChcvXsSMGTNga2uLyZMnK13PyckJL730EmbNmoXo6Gjk5+fj7NmziIiIwKFDhxSuExUVhW3btuHy5cvIy8vD999/DwMDA9jb23fU8OR06dIF5ubm+Prrr5GTk4M///wTYWFhbe7PwcEBcXFxKCoqwp07dwAAS5cuRVRUFL788ktkZ2dj/fr1iI6ObrI+k4ODAy5duoTMzEyUlpairq6uRdtftmwZEhISsGDBAqE65C+//IIFCxY0ua38/HykpKSgtLQUNTU1zdYlAoB33nkHZWVl2LBhA5YtWwYnJyf8v//3/5rs93HAyUjD7NixA4MHD8YzzzwDb29vEBEOHz7c7L01O3bswKxZs7BkyRL069cPQUFBSEpKUnqIIJFIsHXrVvj6+sLDwwNHjx7FwYMHYW5u3hHDakRLSwt79uxBcnIy+vfvj8WLF2Pt2rVt7u/TTz9FbGws7OzsMHDgQABAUFAQPv/8c6xbtw5ubm746quvsGPHDowaNUppP6GhoejXrx+GDBkCCwuLRpUvlfHw8MCJEyeQlZWFkSNHYuDAgXjvvfdgY2OjdJ3nnnsOEyZMwOjRo2FhYYEffvgBIpEIhw8fxj/+8Q+EhITAyckJL774Iq5duwYrKyscP34ckZGR2LlzJ0xNTaGlpYWdO3fi5MmT+PLLL5X2+zjgibKMMY3Ae0aMMY3AyYgxphE4GTHGNAInI8aYRuBkxBjTCJyMGGMagZMRY0wjcDJijGkETkaMMY3AyYgxphE4GTHGNML/BzxP+NHc8+m2AAAAAElFTkSuQmCC",
      "text/plain": [
       "<Figure size 310x260 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(3.1, 2.6))\n",
    "\n",
    "_ = plt.hist(p_vals, bins=1000, cumulative=1, density=True, histtype='step', color='k')\n",
    "\n",
    "plt.plot([0.05, 0.05], [0, 1], 'k--', alpha=0.5)\n",
    "plt.plot([0, 1], [0.957, 0.957], 'k--', alpha=0.5)\n",
    "\n",
    "plt.xlim([-1e-2, 0.4])\n",
    "plt.ylim([0.7, 1])\n",
    "\n",
    "plt.xlabel('proportion of false classes\\n more similar to the text')\n",
    "plt.ylabel('cdf')\n",
    "\n",
    "plt.tight_layout()\n",
    "\n",
    "# plt.savefig(os.path.join('figs', 'cdf_proportion_of_more_similar_false_classes.pdf'), bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Calc. proportion of more similar false classes per bin"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "image_names = [image_name for image_name in ic_imagename2wnid]\n",
    "\n",
    "tq_sims = ic_all_df.loc[image_names, text_query_sim_col].to_numpy()\n",
    "\n",
    "mean_bs = []\n",
    "mean_ps = []\n",
    "err_ps = []\n",
    "for i_b, b_0 in enumerate(ibins[1][:-1]):\n",
    "    b_1 = ibins[1][i_b + 1]\n",
    "    \n",
    "    mean_bs.append((b_0 + b_1)/2)\n",
    "    \n",
    "    p_vals_b = p_vals[np.logical_and(b_0 <= tq_sims,  tq_sims < b_1)]\n",
    "    mean_ps.append(np.mean(p_vals_b))\n",
    "    err_ps.append(1.96*np.sqrt(mean_ps[-1]*(1 - mean_ps[-1])/len(p_vals_b)))\n",
    "    \n",
    "mean_ps = np.array(mean_ps)\n",
    "err_ps = np.array(err_ps)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 330x250 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(3.3, 2.5))\n",
    "\n",
    "plt.plot(mean_bs, ibins[0]/250000, 'r', alpha=0.3)\n",
    "\n",
    "plt.plot(mean_bs, mean_ps, 'k')\n",
    "plt.fill_between(mean_bs, mean_ps + err_ps, mean_ps - err_ps, color='k', alpha=0.3)\n",
    "\n",
    "plt.xlabel('text-[name: def] similarity')\n",
    "plt.ylabel('proportion of false classes\\n more similar to the text')\n",
    "\n",
    "plt.xlim([0, 1])\n",
    "plt.ylim([0, 0.25])\n",
    "\n",
    "plt.tight_layout()\n",
    "\n",
    "plt.savefig(os.path.join('figs', 'text-query_sim_distribution_augmented.pdf'), bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Compare intra-class image similarities"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Settings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "sm_prefix = configs.LAIONConfig.SUBSET_SM_FILTERED_PREFIX\n",
    "ic_prefix = 'imagenet_captions_'\n",
    "\n",
    "cross_image_sim_prefix = 'img_img_sims'\n",
    "cross_image_sim_path = os.path.join('laion400m', 'processed', 'clip_image_similarities')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "def vec(x):\n",
    "    msk = np.tril(np.eye(len(x)) < 0.5)\n",
    "    return x[msk]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load similarities"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|███████████████████████████████████████████████████████████████████████████████████████████████████| 999/999 [00:04<00:00, 233.50it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "898\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "wnid2smsims = {}\n",
    "wnid2icsims = {}\n",
    "\n",
    "for wnid in tqdm(all_wnids):\n",
    "    sm_sim_path = os.path.join(cross_image_sim_path, sm_prefix + cross_image_sim_prefix + f'({wnid}).pkl')\n",
    "    ic_sim_path = os.path.join(cross_image_sim_path, ic_prefix + cross_image_sim_prefix + f'({wnid}).pkl')\n",
    "    \n",
    "    if (not os.path.exists(sm_sim_path)) or (not os.path.exists(ic_sim_path)):\n",
    "        continue\n",
    "\n",
    "    with open(sm_sim_path, 'rb') as f:\n",
    "        sm_sim_dict = pickle.load(f)\n",
    "    with open(ic_sim_path, 'rb') as f:\n",
    "        ic_sim_dict = pickle.load(f) \n",
    "        \n",
    "    if not isinstance(sm_sim_dict, dict):\n",
    "        continue\n",
    "        \n",
    "    sm_sim = sm_sim_dict['similarities']\n",
    "    ic_sim = ic_sim_dict['similarities']\n",
    "        \n",
    "    # Extract sims.\n",
    "    wnid2smsims[wnid] = vec(sm_sim)\n",
    "    wnid2icsims[wnid] = vec(ic_sim)\n",
    "    \n",
    "print(len(wnid2smsims))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Resample"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "rs_sm_sims = []\n",
    "rs_ic_sims = []\n",
    "\n",
    "for wnid, sm_sims in wnid2smsims.items():\n",
    "    ic_sims = wnid2icsims[wnid]\n",
    "    \n",
    "    n_s_i = np.min([len(ic_sims), 100])\n",
    "        \n",
    "    rs_sm_sims.extend(np.random.choice(sm_sims, size=(n_s_i,), replace=True))\n",
    "    rs_ic_sims.extend(np.random.choice(ic_sims, size=(n_s_i,), replace=True))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Plot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 340x250 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(3.4, 2.5))\n",
    "\n",
    "_ = plt.hist(rs_sm_sims, bins=70, color='b', alpha=0.3, density=True)\n",
    "_ = plt.hist(rs_ic_sims, bins=70, color='r', alpha=0.3, density=True)\n",
    "\n",
    "plt.legend(['LAIONet', 'ImageNet'])\n",
    "\n",
    "plt.xlabel('intra-class image-image CLIP similarity')\n",
    "plt.ylabel('\\npdf')\n",
    "\n",
    "plt.xlim([0, 1])\n",
    "\n",
    "plt.tight_layout()\n",
    "\n",
    "plt.savefig(os.path.join('figs', f'{sm_prefix}{ic_prefix}{cross_image_sim_prefix}_resampled.pdf'), bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Test per WNID"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████████████████████████████████████████████████████████████████████████████████████████████| 898/898 [00:00<00:00, 3774.48it/s]\n"
     ]
    }
   ],
   "source": [
    "sm_avg_intrasims = []\n",
    "sm_err_intrasims = []\n",
    "ic_avg_intrasims = []\n",
    "ic_err_intrasims = []\n",
    "\n",
    "for wnid, sm_intrasims in tqdm(wnid2smsims.items()):\n",
    "    ic_intrasims = wnid2icsims[wnid]\n",
    "    \n",
    "    if len(sm_intrasims) < 7 or len(ic_intrasims) < 7:\n",
    "        continue\n",
    "        \n",
    "    sm_avg_intrasims.append(np.mean(sm_intrasims))\n",
    "    sm_err_intrasims.append(1.96*np.std(sm_intrasims)/np.sqrt(len(sm_intrasims) - 1))\n",
    "    \n",
    "    ic_avg_intrasims.append(np.mean(ic_intrasims))\n",
    "    ic_err_intrasims.append(1.96*np.std(ic_intrasims)/np.sqrt(len(ic_intrasims) - 1))\n",
    "    \n",
    "sm_avg_intrasims = np.array(sm_avg_intrasims)\n",
    "sm_err_intrasims = np.array(sm_err_intrasims)\n",
    "ic_avg_intrasims = np.array(ic_avg_intrasims)\n",
    "ic_err_intrasims = np.array(ic_err_intrasims)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "avg < 0: 0.6887115165336374\n",
      "UCB < 0: 565 out of 877 (0.6442417331812998)\n",
      "LCB > 0: 226 out of 877 (0.2576966932725199)\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 340x250 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(3.4, 2.5))\n",
    "\n",
    "diff_avg_intrasims = sm_avg_intrasims - ic_avg_intrasims\n",
    "diff_err_intrasims = np.sqrt(sm_err_intrasims**2 + ic_err_intrasims**2)\n",
    "\n",
    "sorted_locs = np.argsort(diff_avg_intrasims)\n",
    "\n",
    "ucb_intrasims = np.sort(diff_avg_intrasims + diff_err_intrasims)\n",
    "lcb_intrasims = np.sort(diff_avg_intrasims - diff_err_intrasims)\n",
    "\n",
    "x = range(len(sorted_locs))\n",
    "\n",
    "plt.plot(x, diff_avg_intrasims[sorted_locs], color='k')\n",
    "plt.plot(x, ucb_intrasims, 'b')\n",
    "plt.plot(x, lcb_intrasims, 'r')\n",
    "\n",
    "plt.plot(x, [0]*len(x), 'k--', alpha=0.5)\n",
    "\n",
    "plt.legend(['LAIONet - ImageNet', 'UCB', 'LCB'])\n",
    "\n",
    "plt.xlabel(' ')\n",
    "plt.ylabel(r'$\\Delta$ intra-class image-image' + '\\n' + 'CLIP similarity')\n",
    "\n",
    "plt.tight_layout()\n",
    "\n",
    "plt.savefig(os.path.join('figs', 'intra-class_image-image_sim_diff.pdf'), bbox_inches='tight')\n",
    "\n",
    "print(f'avg < 0: {np.mean(diff_avg_intrasims < 0)}')\n",
    "print(f'UCB < 0: {np.sum(ucb_intrasims < 0)} out of {len(ucb_intrasims)} ({np.mean(ucb_intrasims < 0)})')\n",
    "print(f'LCB > 0: {np.sum(lcb_intrasims > 0)} out of {len(lcb_intrasims)} ({np.mean(lcb_intrasims > 0)})')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Compare intra-class cross image similarities for IC most similar texts"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Settings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "tt_sim_th = 0.7"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "ic_ms_prefix = configs.LAIONConfig.SUBSET_IC_MOST_SIMILAR_TXT_TXT_PREFIX\n",
    "\n",
    "ic_ms_imagename2laionindices_file_name = 'icimagename2laionindices.pkl'\n",
    "ic_ms_imagename2sims_file_name = 'icimagename2sims.pkl'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load and preprocess"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "with open(os.path.join(laion_labels_path, ic_ms_imagename2laionindices_file_name), 'rb') as f:\n",
    "    ic_ms_imagename2laionindices = pickle.load(f)\n",
    "    \n",
    "with open(os.path.join(laion_labels_path, ic_ms_imagename2sims_file_name), 'rb') as f:\n",
    "    ic_ms_imagename2ictextsims = pickle.load(f)\n",
    "    \n",
    "ic_ms_wnid2laionindices = {}\n",
    "ic_ms_wnid2ictextsims = {}\n",
    "\n",
    "for image_name, wnid in ic_imagename2wnid.items():\n",
    "    if wnid not in ic_ms_wnid2laionindices:\n",
    "        ic_ms_wnid2laionindices[wnid] = []\n",
    "        ic_ms_wnid2ictextsims[wnid] = []\n",
    "        \n",
    "    assert len(ic_ms_imagename2laionindices[image_name]) == 1\n",
    "    assert len(ic_ms_imagename2ictextsims[image_name]) == 1\n",
    "    \n",
    "    ic_ms_wnid2laionindices[wnid].extend(ic_ms_imagename2laionindices[image_name])\n",
    "    ic_ms_wnid2ictextsims[wnid].extend(ic_ms_imagename2ictextsims[image_name])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load and filter similarites"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|███████████████████████████████████████████████████████████████████████████████████████████████████| 999/999 [00:02<00:00, 362.97it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "962\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "wnid2icmssims = {}\n",
    "wnid2icsims = {}\n",
    "\n",
    "for wnid in tqdm(all_wnids):\n",
    "    ic_ms_sim_path = os.path.join(cross_image_sim_path, ic_ms_prefix + cross_image_sim_prefix + f'({wnid}).pkl')\n",
    "    ic_sim_path = os.path.join(cross_image_sim_path, ic_prefix + cross_image_sim_prefix + f'({wnid}).pkl')\n",
    "        \n",
    "    if (not os.path.exists(ic_ms_sim_path)) or (not os.path.exists(ic_sim_path)):\n",
    "        continue\n",
    "\n",
    "    with open(ic_ms_sim_path, 'rb') as f:\n",
    "        ic_ms_sim_dict = pickle.load(f)\n",
    "    with open(ic_sim_path, 'rb') as f:\n",
    "        ic_sim_dict = pickle.load(f)\n",
    "        \n",
    "    ic_ms_sim = ic_ms_sim_dict['similarities']\n",
    "    ic_sim = ic_sim_dict['similarities']\n",
    "    \n",
    "    # Choose tha samples\n",
    "    tt_sims = np.array(ic_ms_wnid2ictextsims[wnid])\n",
    "    laion_indices = np.array(ic_ms_wnid2laionindices[wnid])\n",
    "    laion_indices = laion_indices[tt_sims > tt_sim_th]\n",
    "    \n",
    "    # Filter sims\n",
    "    locs_filtered = [\n",
    "        ic_ms_sim_dict['index'].index(laion_index) for laion_index in laion_indices\n",
    "        if laion_index in ic_ms_sim_dict['index']\n",
    "    ]\n",
    "    \n",
    "    ic_ms_sim = ic_ms_sim[locs_filtered][:, locs_filtered]\n",
    "    \n",
    "    if len(ic_ms_sim) <= 1:\n",
    "        continue\n",
    "        \n",
    "    # Extract sims.\n",
    "    wnid2icmssims[wnid] = vec(ic_ms_sim)\n",
    "    wnid2icsims[wnid] = vec(ic_sim)\n",
    "    \n",
    "print(len(wnid2icmssims))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Resample"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "rs_ic_ms_sims = []\n",
    "rs_ic_sims = []\n",
    "\n",
    "for wnid, ic_ms_sims in wnid2icmssims.items():\n",
    "    if wnid not in wnid2icsims: \n",
    "        continue \n",
    "        \n",
    "    ic_sims = wnid2icsims[wnid]\n",
    "    \n",
    "    n_s_i = np.min([len(ic_sims), 100])\n",
    "        \n",
    "    rs_ic_ms_sims.extend(np.random.choice(ic_ms_sims, size=(n_s_i,), replace=True))\n",
    "    rs_ic_sims.extend(np.random.choice(ic_sims, size=(n_s_i,), replace=True))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Plot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 340x250 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(3.4, 2.5))\n",
    "\n",
    "_ = plt.hist(rs_ic_ms_sims, bins=70, color='b', alpha=0.3, density=True, range=(0, 0.99))\n",
    "_ = plt.hist(rs_ic_sims, bins=70, color='r', alpha=0.3, density=True, range=(0, 0.99))\n",
    "\n",
    "plt.legend(['LAION | ImageNet texts', 'ImageNet'])\n",
    "\n",
    "plt.xlabel('intra-class image-image CLIP similarity')\n",
    "plt.ylabel('\\npdf')\n",
    "\n",
    "plt.xlim([0, 1])\n",
    "\n",
    "plt.tight_layout()\n",
    "\n",
    "plt.savefig(os.path.join('figs', f'{ic_ms_prefix}{ic_prefix}{cross_image_sim_prefix}_resampled.pdf'), bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Test per WNID"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████████████████████████████████████████████████████████████████████████████████████████████| 962/962 [00:00<00:00, 3502.48it/s]\n"
     ]
    }
   ],
   "source": [
    "ic_ms_avg_intrasims = []\n",
    "ic_ms_err_intrasims = []\n",
    "ic_avg_intrasims = []\n",
    "ic_err_intrasims = []\n",
    "\n",
    "for wnid, ic_ms_intrasims in tqdm(wnid2icmssims.items()):\n",
    "    ic_intrasims = wnid2icsims[wnid]\n",
    "    \n",
    "    if len(ic_ms_intrasims) < 7 or len(ic_intrasims) < 7:\n",
    "        continue\n",
    "        \n",
    "    ic_ms_avg_intrasims.append(np.mean(ic_ms_intrasims))\n",
    "    ic_ms_err_intrasims.append(1.96*np.std(ic_ms_intrasims)/np.sqrt(len(ic_ms_intrasims) - 1))\n",
    "    \n",
    "    ic_avg_intrasims.append(np.mean(ic_intrasims))\n",
    "    ic_err_intrasims.append(1.96*np.std(ic_intrasims)/np.sqrt(len(ic_intrasims) - 1))\n",
    "    \n",
    "ic_ms_avg_intrasims = np.array(ic_ms_avg_intrasims)\n",
    "ic_ms_err_intrasims = np.array(ic_ms_err_intrasims)\n",
    "ic_avg_intrasims = np.array(ic_avg_intrasims)\n",
    "ic_err_intrasims = np.array(ic_err_intrasims)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "avg < 0: 0.8118918918918919\n",
      "UCB < 0: 715 out of 925 (0.772972972972973)\n",
      "LCB > 0: 141 out of 925 (0.15243243243243243)\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 340x250 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(3.4, 2.5))\n",
    "\n",
    "diff_avg_intrasims = ic_ms_avg_intrasims - ic_avg_intrasims\n",
    "diff_err_intrasims = np.sqrt(ic_ms_err_intrasims**2 + ic_err_intrasims**2)\n",
    "\n",
    "sorted_locs = np.argsort(diff_avg_intrasims)\n",
    "\n",
    "ucb_intrasims = np.sort(diff_avg_intrasims + diff_err_intrasims)\n",
    "lcb_intrasims = np.sort(diff_avg_intrasims - diff_err_intrasims)\n",
    "\n",
    "x = range(len(sorted_locs))\n",
    "\n",
    "plt.plot(x, diff_avg_intrasims[sorted_locs], color='k')\n",
    "plt.plot(x, ucb_intrasims, 'b')\n",
    "plt.plot(x, lcb_intrasims, 'r')\n",
    "\n",
    "plt.plot(x, [0]*len(x), 'k--', alpha=0.5)\n",
    "\n",
    "plt.legend(['LAIONet - ImageNet', 'UCB', 'LCB'])\n",
    "\n",
    "plt.xlabel(' ')\n",
    "plt.ylabel(r'$\\Delta$ intra-class image-image' + '\\n' + 'CLIP similarity')\n",
    "\n",
    "plt.tight_layout()\n",
    "\n",
    "plt.savefig(os.path.join('figs', 'intra-class_image-image_sim_diff_cond_text.pdf'), bbox_inches='tight')\n",
    "\n",
    "print(f'avg < 0: {np.mean(diff_avg_intrasims < 0)}')\n",
    "print(f'UCB < 0: {np.sum(ucb_intrasims < 0)} out of {len(ucb_intrasims)} ({np.mean(ucb_intrasims < 0)})')\n",
    "print(f'LCB > 0: {np.sum(lcb_intrasims > 0)} out of {len(lcb_intrasims)} ({np.mean(lcb_intrasims > 0)})')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Test per WNID per text-text th"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "tt_sim_ths = np.linspace(0.5, 0.75, 20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|                                                                                                               | 0/20 [00:00<?, ?it/s]/var/folders/0x/586kgp7s0gqbff4fcbg60twm0000gq/T/ipykernel_1438/2243046274.py:45: RuntimeWarning: invalid value encountered in scalar divide\n",
      "  ic_ms_err_intrasims.append(1.96*np.std(vec(ic_ms_sim))/np.sqrt(len(vec(ic_ms_sim)) - 1))\n",
      "100%|██████████████████████████████████████████████████████████████████████████████████████████████████████| 20/20 [01:18<00:00,  3.93s/it]\n"
     ]
    }
   ],
   "source": [
    "prop_ic_has_lower_avgs = np.zeros((len(tt_sim_ths),))\n",
    "prop_ic_ms_has_lower_avgs = np.zeros((len(tt_sim_ths),))\n",
    "n_wnid_included = np.zeros((len(tt_sim_ths),))\n",
    "\n",
    "for i_tt, tt_sim_th in enumerate(tqdm(tt_sim_ths)):\n",
    "    # Load and filter\n",
    "    ic_ms_avg_intrasims = []\n",
    "    ic_ms_err_intrasims = []\n",
    "    ic_avg_intrasims = []\n",
    "    ic_err_intrasims = []\n",
    "\n",
    "    for wnid in all_wnids:\n",
    "        ic_ms_sim_path = os.path.join(cross_image_sim_path, ic_ms_prefix + cross_image_sim_prefix + f'({wnid}).pkl')\n",
    "        ic_sim_path = os.path.join(cross_image_sim_path, ic_prefix + cross_image_sim_prefix + f'({wnid}).pkl')\n",
    "\n",
    "        if (not os.path.exists(ic_ms_sim_path)) or (not os.path.exists(ic_sim_path)):\n",
    "            continue\n",
    "\n",
    "        with open(ic_ms_sim_path, 'rb') as f:\n",
    "            ic_ms_sim_dict = pickle.load(f)\n",
    "        with open(ic_sim_path, 'rb') as f:\n",
    "            ic_sim_dict = pickle.load(f)\n",
    "\n",
    "        ic_ms_sim = ic_ms_sim_dict['similarities']\n",
    "        ic_sim = ic_sim_dict['similarities']\n",
    "\n",
    "        # Choose tha samples\n",
    "        tt_sims = np.array(ic_ms_wnid2ictextsims[wnid])\n",
    "        laion_indices = np.array(ic_ms_wnid2laionindices[wnid])\n",
    "        laion_indices = laion_indices[tt_sims > tt_sim_th]\n",
    "\n",
    "        # Filter sims\n",
    "        locs_filtered = [\n",
    "            ic_ms_sim_dict['index'].index(laion_index) for laion_index in laion_indices\n",
    "            if laion_index in ic_ms_sim_dict['index']\n",
    "        ]\n",
    "\n",
    "        ic_ms_sim = ic_ms_sim[locs_filtered][:, locs_filtered]\n",
    "\n",
    "        if len(ic_ms_sim) <= 1:\n",
    "            continue\n",
    "\n",
    "        # Extract and average sims.\n",
    "        ic_ms_avg_intrasims.append(np.mean(vec(ic_ms_sim)))\n",
    "        ic_ms_err_intrasims.append(1.96*np.std(vec(ic_ms_sim))/np.sqrt(len(vec(ic_ms_sim)) - 1))\n",
    "        \n",
    "        ic_avg_intrasims.append(np.mean(vec(ic_sim)))\n",
    "        ic_err_intrasims.append(1.96*np.std(vec(ic_sim))/np.sqrt(len(vec(ic_sim)) - 1))\n",
    "        \n",
    "    # Count\n",
    "    diff_avg_intrasims = np.array(ic_ms_avg_intrasims) - np.array(ic_avg_intrasims)\n",
    "    diff_err_intrasims = np.sqrt(np.array(ic_ms_err_intrasims)**2 + np.array(ic_err_intrasims)**2)\n",
    "\n",
    "    ucb_intrasims = diff_avg_intrasims + diff_err_intrasims\n",
    "    lcb_intrasims = diff_avg_intrasims - diff_err_intrasims\n",
    "    \n",
    "    prop_ic_has_lower_avgs[i_tt] = np.mean(lcb_intrasims > 0)\n",
    "    prop_ic_ms_has_lower_avgs[i_tt] = np.mean(ucb_intrasims < 0)\n",
    "    n_wnid_included[i_tt] = len(lcb_intrasims)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 300x250 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "chosen_sim_th = 0.7\n",
    "\n",
    "plt.figure(figsize=(3, 2.5))\n",
    "\n",
    "plt.plot(tt_sim_ths, n_wnid_included, 'b')\n",
    "plt.plot([chosen_sim_th]*2, [850, 1000], 'k--', alpha=0.5)\n",
    "plt.xlim([tt_sim_ths[0], tt_sim_ths[-1]])\n",
    "plt.xlabel('LAION text to ImageNet text\\nCLIP similarity')\n",
    "plt.ylabel('num. of classes included')\n",
    "\n",
    "plt.tight_layout()\n",
    "\n",
    "plt.savefig(os.path.join('figs', 'n-class_after_filtering_ic_most_similar.pdf'), bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAARsAAADwCAYAAAAw2cDUAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjcuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/bCgiHAAAACXBIWXMAAA9hAAAPYQGoP6dpAABJEElEQVR4nO2dd1gUV/v3v7vI7tJRpCpVxS4oEWtijETsNbHXqMRGbPGx5Kdi75hHY3k0Cho1lljIqwkWlKhE0YgiNlREQaXYqErbvd8/Jjuy7AI7uPTzua5zzc6ZM/fcO+x8OXPKfURERGAwGIxSRlzeDjAYjOoBExsGg1EmMLFhMBhlAhMbBoNRJjCxYTAYZQITGwaDUSYwsWEwGGUCExsGg1Em1ChvByoiCoUCL168gImJCUQiUXm7w2BUWIgI6enpsLOzg1hcdN2FiY0GXrx4AXt7+/J2g8GoNMTHx6Nu3bpFlmFiowETExMA3A00NTUtZ28YjIpLWloa7O3t+WemKJjYaED56mRqasrEhsHQAm2aG8q1gfjChQvo3bs37OzsIBKJcPz48WLPCQ0NRatWrSCVSlG/fn0EBgaqldm8eTOcnJwgk8nQpk0bXL16VffOMxgMQZSr2GRmZsLNzQ2bN2/WqnxsbCx69uyJzp074+bNm5g+fTrGjx+PU6dO8WUOHjyImTNnYtGiRYiIiICbmxu8vb2RnJxcWl+DwWBoA1UQANCxY8eKLPOf//yHmjZtqpI3ePBg8vb25vc9PT1pypQp/L5cLic7OztauXKl1r6kpqYSAEpNTdX6HAajOiLkWalU42wuX74MLy8vlTxvb29cvnwZAJCTk4Pr16+rlBGLxfDy8uLLaCI7OxtpaWkqicFg6JZKJTaJiYmwtrZWybO2tkZaWhrev3+PV69eQS6XayyTmJhYqN2VK1fCzMyMT6zbm1ERyMnJgZ+fH/z8/JCTk1Pe7nw0lUpsSot58+YhNTWVT/Hx8eXtEoNR5fjorm+5XI6oqCg4OjqiZs2auvCpUGxsbJCUlKSSl5SUBFNTUxgYGEBPTw96enoay9jY2BRqVyqVQiqVlorPDEZJ0dPTw5dffsl/ruwIrtlMnz4dO3fuBMAJTadOndCqVSvY29sjNDRU1/6p0K5dO4SEhKjknTlzBu3atQMASCQSeHh4qJRRKBQICQnhyzAYlQU9PT106NABHTp0qBJiI7g3qk6dOnTt2jUiIjp27BjZ2dlRdHQ0/d///R+1b99ekK309HS6ceMG3bhxgwCQv78/3bhxg54+fUpERHPnzqWRI0fy5R8/fkyGhoY0e/ZsunfvHm3evJn09PQoODiYL3PgwAGSSqUUGBhId+/eJR8fHzI3N6fExESt/WK9UQyGdgh5VgSLjVQqpfj4eCIimjBhAk2bNo2IOCEwMTERZOv8+fMEQC2NHj2aiIhGjx5NnTp1UjvH3d2dJBIJubi4UEBAgJrdTZs2kYODA0kkEvL09KQrV64I8ouJDaMiIJfL6dmzZ/Ts2TOSy+Xl7Y5GhDwrIiJhS7k4Ojpix44d6NKlC5ydnbF161b07NkTd+7cQceOHfH27VvdVr3KgbS0NJiZmSE1NZVNV2CUGzk5OVixYgUAYP78+ZBIJOXskTpCnhXBDcRjx47FoEGDYGtrC5FIxI9pCQ8PR6NGjUrmMYPBqPIIFhs/Pz80a9YM8fHx+Prrr/leHD09PcydO1fnDjIYjKpBibq+v/rqKwBAVlYWnzd69GjdeMRgMKokgru+5XI5li5dijp16sDY2BiPHz8GACxYsIDvEmcwGIyCCBab5cuXIzAwEGvWrFFpsGrWrBl+/vlnnTrHYDCqDoLFZs+ePdi+fTuGDx+uMtDIzc0N9+/f16lzDAaj6iC4zeb58+eoX7++Wr5CoUBubq5OnGIwGFyny+eff85/ruwIFpsmTZrg4sWLcHR0VMn/7bff0LJlS505VhE4cwawtQXMzABTUy6ZmADFBJFnMHRCfrGpCggWm4ULF2L06NF4/vw5FAoFjh49iujoaOzZswcnTpwoDR/LjX873dQwMVEVIAMDQCotPhkYAObmQK1aQM2aqlsTE4CtGsOoyggeQQwAFy9exJIlSxAZGYmMjAy0atUKCxcuRNeuXUvDxzJHOSqyRYtUZGSYIi0NSE0FSvMtUU+PEx6l+JiZAUZGgKEht83/ueDW3BywtORSrVqcLUblh4jw8uVLAIClpWWFXMNMyAjiEolNVUfTDSQCsrPBC0/+7fv33LHi0vv3QEoK8OYN8PYtt33zhjumK8RiTnCU4lMwmZtztTFlzSz/tgKOhq/WVPvpCvHx8RCJRPyCVFevXsX+/fvRpEkT+Pj4lMzjSoBIBMhkXLKy0q3t9+8/iM/bt1xKSQHeveNSZuaHbf7Pyu2bN8DLl9x5CgXw6hWX7t0T5odU+kF4jIy0f62zsACaNlVN5uZC7wKjqiNYbIYNGwYfHx+MHDkSiYmJ8PLyQrNmzbBv3z4kJiZi4cKFpeFnlcbAgEt2dh9nJzcXeP2aE57CUsFaWWoqJ1gAV8NSlhNKgTBDsLMDmjVTFaAmTTgxY1RPBIvN7du34enpCQA4dOgQmjdvjrCwMJw+fRoTJ05kYlOO6OsDNjZcEoJczglPfhFSClBxEAEvXgB37nxIz55xeS9eAKdPq5a3t1cVH+VWiwUVGZUcwWKTm5vLT748e/Ys+vTpAwBo1KgREhISdOsdo0zI3zitC1JTgbt3OeG5ffuDCCUkAPHxXAoOVj3HweGD+CgFyMUFqF2b9dJVFQSLTdOmTbFt2zb07NkTZ86cwdKlSwEAL168gIWFhc4dZFQ+zMyAdu24lJ83bzgRUgrRnTvc54QEIC6OSwVFyNAQcHLikrPzh8/KZGHBxKiyIFhsVq9ejf79+2Pt2rUYPXo03NzcAAC///47/3rFYGiiVi2gY0cu5UeTCN2/z72GvXv34ZgmjI25WpG9PVC3Lpfyf65bl2snYoJU/pSo61sulyMtLU1lNYUnT57A0NAQVrruqikHWKS+ikF2NvfK9eQJEBvLbfN/1vat3dj4gwA1agR4eHCpUSOgxkevL1J6yOVyPnh/ly5dKuSUhVIdZ/P+/XsQEQwNDQEAT58+xbFjx9C4cWN4e3uX3OsKBBObykFWFvD0Kff69ezZhxQf/+FzUVFqDQwANzdOeFq14rZNmnAN7QztKFWx6dq1KwYMGICJEyciJSUFjRo1gr6+Pl69egV/f39MmjTpo5yvCDCxqTpkZgLPn3MCFBcH3LoFRERwKSNDvbxUCrRoAXzyCdC+PffK5+jIXsMKo1TFpnbt2vjrr7/QtGlT/Pzzz9i0aRNu3LiBI0eOYOHChbgndCRZBYSJTdVHoQAePuRE5/p1LkVEcN3+BalbF/j0Uy517Mj1lpXFZFwiQmpqKgDAzMys0k9XEPzG+u7dO5j8Oyji9OnTGDBgAMRiMdq2bYunT5+WzGMGo4wRi4GGDbk0dCiXp1AAjx9zonP1KnDpEidCz54Bv/7KJYAbItChwwfx8fDgakS6Jjc3Fz/++COAijtdQQiCxaZ+/fo4fvw4+vfvj1OnTmHGjBkAgOTkZFYLYFRqxGKgfn0uDRrE5WVmAuHhwMWLXLp8mWsHOnGCSwD3ilWrFjeNxdKS2ypTwX1HR66tqDpSohATw4YNw4wZM9ClSxd+WdvTp09XuXg2DIaREfDFF1wCuCkhN29+EJ9Ll7h5aK9fc6m4VgR9fcDTE/jsMy61b199pnCUqOs7MTERCQkJcHNzg/jfl9erV6/C1NS0SqwdxdpsGNpCBCQnf5hTlpz8Yb/g56QkbnR1fsRioGXLD+LTsSM3ahqoerO+BS+/Wxr89NNP5OjoSFKplDw9PSk8PLzQsp06ddK4ZG+PHj34MqNHj1Y77u3trbU/bPldRmmgUBA9ekS0axfRmDFELi5EnFyppqZNiSZNIrpyJZsWLVpEixYtouzs7PJ2XyNCnpUSDWn6559/cOjQIcTFxSEnJ0fl2NGjRwXZOnjwIGbOnIlt27ahTZs2+PHHH+Ht7Y3o6GiNAwSPHj2qcs3Xr1/Dzc0NX3/9tUq5bt26ISAggN+XlkYLHoMhAJEIqFePS2PHcnnPnnGvYxcucCn/KOqtW7mZ88pXuEqPUCX79ddfSV9fn3r16kUSiYR69epFrq6uZGZmRmPGjBGsjJ6enjRlyhR+Xy6Xk52dHa1cuVKr8zds2EAmJiaUkZHB540ePZr69u0r2BclrGbDKC+Sk4mOHiUaOpQIyCZgEYnFi8jXN5tevSpv79QR8qwIHi2wYsUKbNiwAf/v//0/SCQS/Pe//8X9+/cxaNAgODg4CLKVk5OD69ev8+uFA4BYLIaXlxcuX76slY2dO3diyJAhMDIyUskPDQ2FlZUVGjZsiEmTJuH169eCfGMwygNLS6B/f2D/fuDqVTHc3VtDoWiNTZvEcHEBVq3igq1VSoQqmaGhIcXGxhIRUa1atejWrVtERHT37l2ysbERZOv58+cEgP7++2+V/NmzZ5Onp2ex54eHhxMAtTaeX3/9lYKCgujWrVt07Ngxaty4MbVu3Zry8vI02snKyqLU1FQ+xcfHs5oNo8Jw6hSRm9uHNp06dbh2n0J+zmVKqdZsatasifT0dABAnTp1cPv2bQBASkoK3r17pzsV1IKdO3eiefPmarPNhwwZgj59+qB58+bo168fTpw4gWvXriE0NFSjnZUrV8LMzIxP9vb2ZeA9g6EdXbtyAw1/+YWb4f78OfDNN4C7O3DyJCdBlQHBYvPZZ5/hzJkzAICvv/4a06ZNw4QJEzB06FB06dJFkK3atWtDT08PSUlJKvlJSUmwKSbcXGZmJg4cOIBx48YVex0XFxfUrl0bjx490nh83rx5SE1N5VN8fLz2X4LBKCWICJmZmcjMzIRIRBgxAoiOBtat40Yx374N9OrFNSDfuFHe3mqB0GrT69ev6fnz50TENeauXLmSevfuTTNnzqQ3b94IroZ5enrS1KlT+X25XE516tQptoE4ICCApFIpvdKi1Sw+Pp5EIhEFBQVp5RNrIGZUBLKzC+/6fvOGaPZsIqmUe7USiYjGjydKSipbH4U8K+U+zubAgQMklUopMDCQ7t69Sz4+PmRubk6JiYlERDRy5EiaO3eu2nkdO3akwYMHq+Wnp6fT999/T5cvX6bY2Fg6e/YstWrViho0aEBZWVla+cTEhlERKEpslDx9quy54pKpKdHatURlNSxH5+Ns0jRNhS0EoSNuBw8ejJcvX2LhwoVITEyEu7s7goODYW1tDQCIi4vjRykriY6OxqVLl3C6YDRtcEuW3rp1C7t370ZKSgrs7OzQtWtXLF26lI21YVQ5HBy4nqspU4Bp07iJo7NnA9u3A+vXc69ZFWWyuFbTFcRicbHT24kIIpEIcrlcZ86VF2y6AqMiIHS6gkIB7N4NzJ8PJCZyeV27Ahs2cEHBSgOdh5g4f/68ThxjMBilh1jMjUz+6itgxQrA359bSqdFC2DyZMDPj5udXl5oJTadOnUqbT8YDIaOMDEBVq4Exo/nXqmOHQM2bQL27QMWLQImTCifMBeCu74DAgJw+PBhtfzDhw9j9+7dOnGKwWB8PPXqAUePcquVNm/OrWIxbRq3Hte6dZrDopYmgsVm5cqVqK2cA58PKysr/v2SwWB8PGKxGO7u7nB3d1frJBHCF19wgwL/9z+uQTkxkavxODoCS5dy68qXBYLj2chkMty/fx9OTk4q+U+ePEHjxo3xvtJO3PgAayBmVFVyc7nXqRUruBjMAPfaNXUqMGMGNzdLCEKeFcFyaWVlhVu3bqnlR0ZGshUxGYwKjr4+MGYMF1Hw11+516v0dK6Nx9ERmD6dC3tRGgiOZzN06FB89913MDExwWeffQYA+OuvvzBt2jQMGTJE5w5WZORyOXJzc8vbDUYVhYj435e+vr7OV1fo1w8YOFAff/6ph2XLgGvXgP/+F9iyhevVWrCAW1lCVwh+jcrJycHIkSNx+PBh1Ph3OUGFQoFRo0Zh27ZtFTJ0oVCKqxoSERITE5FSVi+7jGoJldFSLubm5rC2tsHZsyIsX84F8QIAOzuucbmoSL+lum6UkocPH+LmzZswMDBA8+bN4ejoWBIzFZLibmBCQgJSUlJgZWUFQ0PDCrmeD6Pyo1Ao8OrVKwDcpOWPaSTWBBHh3bt3SE5Ohrm5OWxtbQFwkQMnTuSiBlpbA2fPchEDNVGq60YpadCgARo0aFDS0ystcrmcFxrWRsUoTRQKBf/2IJPJdC42AGDw74Cb5ORkWFlZQU9PD59+CoSGAl9+CURGAp07c4Lj5vZx1yqDdf2qFsp3aOVa5wxGZUf5W87f/mhpCZw7xy3A9+oVJzj//PNx12FiU0LYqxOjqlDYb7lWLa5G07YttzBfly7AlSslvw4TGwaDUSjm5tz8qo4duXXQv/ySW5ivJGglNgMGDODDTOzZswfZ2dkluxqDwah0mJgAwcHcq1RGBuDtDZRkbrZWYnPixAlkZmYCAMaOHct3xzEqD2PGjEG/fv2KLffs2TNIJBI0K6T7QSQS4fjx4yp5J06cQKdOnWBiYgJDQ0O0bt0agYGBKmWePHkCkUgEKysrPoa1End3d/j5+Qn4NsIIDAyEubl5qdnXNU5OThCJRAgPD4dMJoNMJoNIJML06dPx+eefa21Hec9v3rz50T4ZGXFrm3ftCrx7B/TowdV4hKBVb1SjRo0wb948dO7cGUSEQ4cOFdrNNWrUKGEeMCoUgYGBGDRoEC5cuIDw8HC0adOmyPKbNm3C9OnTMWfOHGzduhUSiQRBQUGYOHEibt++jXXr1qmUT09Px7p167B48WKd+ZyTk4M3b94UG7e6MiGTyTB37lz89ddf5e0Kj6EhEBTEhbA4eRLo3RvYu1eAAW1C/4WFhVGbNm2odu3aJBaLyczMjMzNzdVSzZo1SxBYsOJRVKjD9+/f0927d+n9+/fl4FnJ0WbhPoVCQS4uLhQcHExz5syhCRMmqJUBQMeOHSMiori4ONLX16eZM2eqldu4cSMBoCtXrhARUWxsLAGg2bNnk7GxMSXlC5br5uZGixYtEvyd/vnnH5o6dSpZWFjQjz/+WGi5gIAAMjMz4/cXLVpEbm5utHPnTrK3tycjIyOaNGkS5eXl0erVq8na2posLS1p2bJlKnbWr19PzZo1I0NDQ6pbty5NmjSJ0tPTVcps376d6tatSwYGBtSvXz9av369yrWJiI4fP04tW7YkqVRKzs7O5OfnR7m5ufxxR0dH+u6770gikdDJkyf5/GnTplGnTp1UbO3YsYMaNWpEUqmUGjZsSJs3b+aPocAS1AXPVSL0N52dTdS/PxeGtEaNUoxBLBKJVH4oVRGhYqNQEGVklH1SKLT/TtqITUhICNnY2FBeXh5FRUWprTRKpCo2/v7+BIBevHihZis7O5uMjY1p2rRpRPRBbCIiIsjd3V1lFVQhYvPixQtas2YNNW3alCQSCfXv35+OHTtGOTk5hZ6jSWyMjY3pq6++ojt37tDvv/9OEomEvL29ydfXl+7fv0+7du1SEUsibvXVc+fOUWxsLIWEhFDDhg1p0qRJ/PFLly6RWCymtWvXUnR0NG3evJlq1aqlcu0LFy6QqakpBQYGUkxMDJ0+fZqcnJzIz8+PL+Po6EgbNmyg7777jlq0aEFyuZyI1MVm7969ZGtrS0eOHKHHjx/TkSNHqFatWhQYGEhERFevXiUAdPbsWUpISKDXr19rvD8l+Qeak0M0eDARUIpi8+TJE1II+ZVXQoSKTUaG5gXiSzsV0IEi0UZshg0bRtOnT+f33dzcKCAgQKVMfrGZOHGi2n/t/LRo0YK6d+9ORB/E5saNGxQcHEz6+vr06NEj/jpFiU12djYdOHCAunfvTjVq1KC2bdvSli1btF7NQ5PYGBoaUlpaGp/n7e1NTk5O/INNRNSwYcMiV/k4fPgwWVhY8PuDBw+mnj17qpQZPny4yrW7dOlCK1asUCnzyy+/kK2tLb+vFJvExEQyNjam//73vySXy9XEpl69erR//34VW0uXLqV27doRkeo9L4qS1tZzc4kGD9ZxwPP8ODo6IiUlBTt37sS9e/cAAE2aNMG4ceNgZmYm1ByjgpCSkoKjR4/iUr5+zREjRmDnzp0YM2aMTq/l7e2Njh07YsGCBdi/f3+x5f/++28MGTIE9vb2OHfuHD799NOP9sHJyQkmJib8vrW1NfT09FRG6VpbWyM5OZnfP3v2LFauXIn79+8jLS0NeXl5yMrKwrt372BoaIjo6Gj0799f5Tqenp44ceIEvx8ZGYmwsDAsX76cz5PL5Sp2lFhaWmLixIlYt24dfHx8VOxmZmYiJiYG48aNw4QJE/j8vLy8MnsOa9QAtm4FDh7UsrzQC/zzzz/w9vaGgYEBvxLlhg0bsGLFCpw+fRqtWrUSarLSY2hY9lHPlNfVFfv370dWVpZKgzARQaFQ4MGDB3B1dVU7x9XVFampqXjx4gXs7OxUjuXk5CAmJgadO3fWeL1Vq1ahXbt2mD17drG+eXp6YseOHdi9eze++OILeHl5YeTIkejXr1+JR3Lr6+ur7ItEIo15CoUCANez06tXL0yaNAnLly9HrVq1cOnSJYwbNw45OTla+5GRkYHFixdjwIABasdkMplano+PD3bv3o2tW7eq2QGAHTt2qDXi6+npaeWLLhByKcGD+mbMmIE+ffrgyZMnOHr0KI4ePYrY2Fj06tUL06dPF2quSiAScV2DZZ10OYh5586dmDVrFm7evMmnyMhIfPrpp9i1a5fGcwYOHAh9fX2sX79e7di2bduQmZmJoUOHajzX09MTAwYMwNy5c4v1zdDQEOPHj8fFixdx//59tG7dGj/88ANsbGwwduxYnDt3jheF0uL69etQKBRYv3492rZtC1dXV7x48UKlTMOGDXHt2jWVvIL7rVq1QnR0NOrXr6+WNM19MjIywvTp07FixQqVIQPW1taws7PD48eP1ew4OzsDAB+BoaKseFKims2OHTv4CWIAUKNGDfznP//BJ598olPnGLolNTVVbcyFhYUFXr9+jYiICOzbtw+NCsQTGDp0KJYsWYJly5ap/M0BwMHBAWvWrMGsWbMgk8kwcuRI6OvrIygoCPPnz8esWbOK7Dpfvnw5mjZtqma3KOrVq4clS5Zg8eLF+OuvvxAYGIi+fftixYoV8PX11dqOUOrXr4/c3Fxs2rQJvXv3RlhYGLZt26ZSxtfXF5999hn8/f3Ru3dvnDt3Dn/++afKdICFCxeiV69ecHBwwFdffQWxWIzIyEjcvn0by5Yt03jt4cOHY9euXdi/f7/K/Vy8eDG+++47mJmZoVu3bsjOzsY///yDt2/fYubMmbCysoKBgQGCg4NRt25dyGSy8m3qENQiRERWVlZ06tQptfzg4GCysrISaq5CUlW7vlGgKxQAjRs3jqZOnUpNmjTReF5CQgKJxWJ+6WLkayBWEhQURJ9++ikZGRmRTCYjDw8P2rVrl0qZwhorfXx8CECJur6VZGRkUFxcXKHHC+v6zo+mBvROnTrxvWlEXO+bra0tGRgYkLe3N+3Zs4cA0Nu3b/ky27dvpzp16vBd38uWLSMbGxsVu8HBwdS+fXsyMDAgU1NT8vT0pO3bt/PHlQ3Ecrmcnj9/Ts+fP6e9e/dq7L7et28fubu7k0QioZo1a9Jnn31GR48e5Y/v2LGD7O3tSSwW66zrOz+luvyur68v1a1blw4cOEBxcXEUFxdHv/76K9WtW1flD1OZqYpiwygfxo8fTx07dizRufnFJn8vma4pK7ER/Bq1bt06iEQijBo1Cnl5eQC4xrZJkyZh1apVOqhrMRiVl3Xr1uHLL7+EkZER/vzzT+zevRtbtmwpkS2RSMQvGV0logyURAmJiDIzM+nWrVt069YtyszMLKkZIiL66aefyNHRkaRSKXl6elJ4eHihZQMCAtReBaRSqUoZhUJBCxYsIBsbG5LJZNSlSxd68OCB1v6wmg2jpHz99ddkaWlJMpmMmjRpQlu3bi1vl4qlwtZslBgaGqJ58+YfLXYHDx7EzJkzsW3bNrRp0wY//vgjvL29ER0dDSsrK43nmJqaIjo6mt8vqPpr1qzBxo0bsXv3bjg7O2PBggXw9vbG3bt3NXYvMhi64tChQ+XtQsWlJEqoSzw9PVWGrsvlcrKzsyt05GbBxr6CKBQKsrGxobVr1/J5KSkpJJVK6ddff9XKJ1azYVQnyqpmU67Bs3JycnD9+nV4eXnxeWKxGF5eXrh8+XKh52VkZMDR0RH29vbo27cv7ty5wx+LjY1FYmKiik0zMzO0adOmUJvZ2dlIS0tTSQxGeaNQKJCQkICEhIRSH0dUFpSr2Lx69QpyuRzW1tYq+dbW1khMTNR4TsOGDbFr1y4EBQVh7969UCgUaN++PZ79u7KW8jwhNleuXAkzMzM+2dvbf+xXYzB0AnE9xuXthk6odGFB27Vrh1GjRsHd3R2dOnXC0aNHYWlpif/9738ltjlv3jykpqbyKT4+XoceMxgMoIRLuTx8+BDnz59HcnKyWvVu4cKFWtupXbs29PT0kJSUpJKflJSkdSAkfX19tGzZEo8ePQIA/rykpCR+HRzlvru7u0YbUqmU72JkMBilg2Cx2bFjByZNmoTatWvDxsZGpSdIJBIJEhuJRAIPDw+EhITwISsVCgVCQkIwdepUrWzI5XJERUWhR48eAABnZ2fY2NggJCSEF5e0tDSEh4dj0qRJWvvGYDB0jNDWZwcHB1q1apXgVuvCOHDgAEmlUgoMDKS7d++Sj48PmZubU2JiIhERjRw5kubOncuXX7x4MZ06dYpiYmLo+vXrNGTIEJLJZHTnzh2+zKpVq8jc3JyCgoLo1q1b1LdvX3J2dta6tZ31RjEqAlVtBLHgNpu3b9/i66+/1pnYDR48GOvWrcPChQvh7u6OmzdvIjg4mG/gjYuLQ0JCgsr1J0yYgMaNG6NHjx5IS0vD33//jSZNmvBl/vOf/8DX1xc+Pj5o3bo1MjIyEBwcXK3H2LCA5+alZl/XKAOeXymwSFN5BjzXBYLF5uuvv8ZpoWHVi2Hq1Kl4+vQpsrOz1YJsh4aGqvxwN2zYwJdNTEzEyZMn0bJlSxV7IpEIS5YsQWJiIrKysnD27FmN8VgY6igDnitfPYtj06ZN6Nu3Lzp06IDw8HDcunULQ4YMwcSJE/H999+rlVcGPBdCSkpKtRuOIJPJMG/ePEgkEj5URGVHsNjUr18fCxYswJgxY7B+/Xps3LhRJTEqL0SEgIAAjBw5EsOGDcPOnTuLLB8fH49Zs2bx8VaaNGmC+vXrY9asWVi7di3Wr1+vJli+vr7w9/dXiYBXHJGRkbCxscGIESNw5syZjx5z4ufnB3d3d+zatQsODg4wNjbG5MmTIZfLsWbNGtjY2MDKykolmh4A+Pv7o3nz5jAyMoK9vT0mT57MB7FSsmPHDtjb28PQ0BD9+/eHv7+/Wq0qKCgIrVq1gkwmg4uLCxYvXszPM1Ti4+ODK1eu4OrVq6hdu3ah63z//PPPaNy4MWQyGRo1aqQyD0sZ16Zly5YQiUSCakWlgtB3NCcnp0KTs7Oz4He+iojgNptKEPG8Mgc8VygUFBoaSt988w2ZmppS3bp1ad68eXT//n2tvjsLeK77gOdKSjXERHVAsNhUgojnlTngeX7evXtH+/fvp27dulGNGjWoTZs2tHXrVkpJSSn0nMoa8Dw5OZlMTExoz549RKQuNuUd8JyoDKcrUBUa3VjdUQY8HzFiBJ+nDHiua/IHPBeKgYEBhg4dij///BN37txBbm4uJk2ahICAAEF2NAU8b9KkSbEBz7t06YI6derAxMQEI0eOxOvXr/Hu3TsAQHR0NB+XW0nB/cjISCxZsgTGxsZ8mjBhAhISEng7SiwsLDBx4kT88MMPyMrKUjmWP+B5flvLli1DTEyMoHtRVpRoUN+ePXuwdu1aPHz4EAAX+Hr27NkYOXKkTp2rNFSBiOcVOeB5fvLy8nD69Gn88ssvCAoKgouLC9asWYPhw4cLslNZAp5PmDABgYGBFTbguRAEi42/vz8WLFiAqVOnokOHDgCAS5cuYeLEiXj16hVmzJihcycrPMqI55UYZcDzgsu2TJ48Gbt27dIYGG3gwIGYM2cO1q9frxb0XJcBzwEgIiICv/zyC3799Vfk5eVh6NChuHDhQpnFvc4f8FxZ+ykYTkJowHNtyB/wvE+fPnx+/oDnhQltpQ94vmnTJmzdulVlTe8+ffqgadOm8PPzq55iU0morAHPL168iC5duqB79+7YsmULevXqVebdwSzguQ4Q2iAklUrp4cOHavkPHjxQi5hXWamKI4grc8DzV69eUXJysuDvrIQFPK8YAc9FRMJaeJs1a4Zhw4Zh/vz5KvnLli3DwYMHERUV9dECWN6kpaXBzMwMqampMDU1VTmWlZWF2NhYODs7V+sRyQztmDBhAu7fv4+LFy8KPlehUPBhUWxsbAoda/OxfMxvuqhnpSCCX6MWL16MwYMH48KFC3ybTVhYGEJCQlhIREa1R5cBz6sagsVm4MCBCA8Px4YNG/g5Mo0bN8bVq1fVpg0wGNWNq1evYs2aNUhPT4eLiws2btyI8ePHl9hewR6yykyJur49PDywd+9eXfvCYFR6dFm7F4vFsLS01Jm98kYrsUlLS+Pfx4qbEFfcexuDwaieaCU2NWvWREJCAqysrGBubq5xwSwigkgkqjB9+gwGo2KhldicO3cOtWrVAgCcP3++VB2qLFSFaPeMio1CocDLly8BAJaWlqXWG1VWv2WtxKZTp078Z2dnZ9jb26vVboioWgQKl0gkEIvFePHiBSwtLSGRSKrG0qiMCodCoUB2djYArnta12JDRMjJycHLly8hFotLfaCk4AZiZ2dn/pUqP2/evIGzs3OVf40Si8X8PXjx4kV5u8OowhARUlNTAXBzoUrrn5qhoSEcHBxKreakRLDYKNtmCpKRkVFtBrlJJBI4ODggLy+vyosro/zIycnBH3/8AYALplUaNQ89PT3UqFGjTGrnWovNzJkzAXAzYRcsWKAyy1UulyM8PLzQpVKqIspZwlVpHASjYiEWi/nZ3TKZrNKHB9VabG7cuAGAq9lERUWpfHGJRAI3NzeNMWcZDAYDECA2yl6osWPHYuPGjSqBhxgMBqM4BLUI5ebm4pdffsHTp09Lyx8Gg/EvIpEIlpaWsLS0rBI9noIaiPX19eHg4MAaRRmMMkBfXx9Tpkwpbzd0huC+rh9++AHz58/HmzdvSsMfBoNRRRHc9f3TTz/h0aNHsLOzg6OjI4wKhMOMiIjQmXMMBqPqIFhstFnCVSibN2/G2rVrkZiYCDc3N2zatEktKr2SHTt2YM+ePbh9+zYAbgb6ihUrVMqPGTMGu3fvVjnP29sbwcHBOvedwSgtcnNzsX37dgDcOJvKPsxCsNgsWrRIpw4cPHgQM2fOxLZt29CmTRv8+OOP8Pb2RnR0tNooZYBbjnfo0KFo3749ZDIZVq9eja5du+LOnTuoU6cOX65bt24qy3tIpVKd+s1glDZExM+NEhhQs0JSong2ABdt/t69ewCApk2bljhwlr+/PyZMmICxY8cC4KLynzx5Ert27dIYeX/fvn0q+z///DOOHDmCkJAQlSDsUqkUNjY2JfKJwWDoHsFik5ycjCFDhiA0NJRfwzglJQWdO3fGgQMHBAX7ycnJwfXr1zFv3jw+TywWw8vLC5cvX9bKxrt375Cbm8vPSlcSGhoKKysr1KxZE1988QWWLVsGCwsLrX1jMBi6RXBvlK+vL9LT03Hnzh28efMGb968we3bt5GWlobvvvtOkK1Xr15BLpfD2tpaJd/a2poP9Fwcc+bMgZ2dHby8vPi8bt26Yc+ePQgJCcHq1avx119/oXv37oV22WdnZyMtLU0lMRgM3SK4ZhMcHIyzZ8+icePGfF6TJk2wefNmdO3aVafOFceqVatw4MABhIaGqkwCHTJkCP+5efPmaNGiBerVq4fQ0FB06dJFzc7KlSuxePHiMvGZwaiuCK7ZKBQKja3i+vr6goPw1K5dG3p6ekhKSlLJT0pKKra9Zd26dVi1ahVOnz6NFi1aFFnWxcUFtWvXxqNHjzQenzdvHlJTU/lUHeLyMBhljWCx+eKLLzBt2jSVWC7Pnz/HjBkzNNYaikIikcDDwwMhISF8nkKhQEhICNq1a1foeWvWrMHSpUsRHBys1fKrz549w+vXr2Fra6vxuFQqhampqUpiMMobkUgEc3PzQkPxVjqEroAXFxdH7u7upK+vTy4uLuTi4kL6+vrUsmVLio+PF2qODhw4QFKplAIDA+nu3bvk4+ND5ubmlJiYSEREI0eOpLlz5/LlV61aRRKJhH777TdKSEjgU3p6OhERpaen0/fff0+XL1+m2NhYOnv2LLVq1YoaNGhAWVlZWvkkZJU/BqM6I+RZEdxmY29vj4iICJw9exb3798HwK0blb+BVgiDBw/Gy5cvsXDhQiQmJsLd3R3BwcF8o3FcXJxKBLGtW7ciJycHX331lYqdRYsWwc/PD3p6erh16xZ2796NlJQU2NnZoWvXrli6dCkba8NglCOCl9+tDghZUpTBqM4IeVZKFHQ0JCQEvXr1Qr169VCvXj306tULZ8+eLZGzDAZDM8rpCtu3b0dubm55u6OOXA78O21IGwSLzZYtW9CtWzeYmJhg2rRpmDZtGkxNTdGjRw9s3rxZqDkGg1EIRIQXL17gxYsXFWO6wuvXwMmTwP/9H9ClC2BuDnTooPXpgttsVqxYgQ0bNmDq1Kl83nfffYcOHTpgxYoVVSr+BoNRbZHLgTt3gMuXP6QHD9TLGRkBmZlamRQsNikpKejWrZtafteuXTFnzhyh5hgMRnmQlgbExxed3r9XP8/VFWjX7kOytwcKTBUqDMFi06dPHxw7dgyzZ89WyQ8KCkKvXr2EmmMwGLqGCHj7FnjyBIiN/bCNjQXi4jgh+Xc9qiIxNgbatPkgLG3aAAXnFwqY2iNYbJo0aYLly5cjNDSUH3h35coVhIWFYdasWdi4cSNfVuhcKQaDIYC0NOD6dSAq6oOYKIVFGxEwN+dqJvmTg8OHz05OgJ6eztwV3PXt7OysnWGRCI8fPy6RU+UN6/pmVARycnKwYsUKAMD86dMhuXMH+Ocf4No1bhsdXbQBa2tOMJydP2wdHT+IiQ5WSBHyrAiu2cTGxpbYMQaj2kME5OUBOTlAbu6HlH8/J4drdI2IgOHJk8CzZ8CSJdy5BXF0BFq1AurVUxeVfAtJVgRKHDwL+BA9rErM22AwioOI6/6NjuZ6Zh484D4nJKiLhyYRyc3lhEZLJAD+kz/DxgZo3ZpLn3zCJQHxo8qbEonNnj17sHbtWjx8+BAA4OrqitmzZ2PkyJE6dY7BKHOIuMbTp0+Bhw8/CIty+/at7q+ppwfo6wMSCbfV1wekUqBRI05QlOKSL+xtZUSw2Pj7+2PBggWYOnUqOvw7oOfSpUuYOHEiXr16hRkzZujcSQZDZ7x7p7mbV9lLEx8P/Lu+dqE4OHBdwA0bclt7e04c8otF/qQpX5lXowYgLtFA/kpHiRqIFy9erBLvFwB2794NPz+/KtGmwxqIKxA5OVzPSmpq4dv0dE5EMjM/bPN/zp+nbVethQXQoAEnJvmFpX79MmsLyc3N5WNuDx8+vEKurlCqDcQJCQlo3769Wn779u2RkJAg1ByjqpOVBbx8CSQnc9uUFM2iUZiQZGXp3idjY9Uu3oKpbl1uZGw5Q0R48uQJ/7myI1hs6tevj0OHDmH+/Pkq+QcPHkSDBg105hijgkLE1SQSE7mG0cREICmJE5P8SSkwuornbGQEmJkBpqbqWxMT7rihIbfN/7lgnpUVdx7r1ChzBIvN4sWLMXjwYFy4cIFvswkLC0NISAgOHTqkcwcrNETcf96cHG4uiVzO9TYUts3LA7KztUvKngttbBobc70SmpKFBdcuUBhZWZprGG/eqApKQsKHz+/eCbtPNWpwD7mlJVCzZuGiUZSYFPUdGJUCwX/BgQMH4urVq/D398fx48cBcMGzrl69WuK1oyosfn6FP4zKzxVx6n9+RCLuAbe05OawZGWp+p+TUzK7xsaArS3XHWttzSUrK/VkacmNVGU1iWqPILHJzc3Ft99+iwULFmDv3r2l5VPFYcMG4eeIxdx/YT29D9v8n6VS7ZJEom5H01Ys5npPXr5UTcnJXO2EiNu+eVO03yYmH2oSpqacQCjFpODWxoYTGwZDAILERl9fH0eOHMGCBQtKy5+KxaRJ3H/m/A+hpiq+TPZBVCrSf/C8PG4QmlKA3r7l2i0Kfg9j42rT/cooPwS/RvXr1w/Hjx+vHuNpVq3iHsjKSo0aH15xGJWSitjdXVIEi02DBg2wZMkShIWFwcPDA0YFugjZTG8GQzdIJBL88MMP5e2GztDprO/KPNM7P2xQH4OhHWzWN4PBqHCwWd8MRgUlLy8PBw8eBMCtr1ajko81KlEXxM6dO9GsWTPIZDLIZDI0a9YMP//8s659YzCqNQqFAg8fPsTDhw+hUCjK252PRrBULly4EP7+/vD19eXDgl6+fBkzZsxAXFwclixZonMnGQxG5Uew2GzduhU7duzA0KFD+bw+ffqgRYsW8PX1ZWLDYDA0Ivg1Kjc3F5988olavoeHB/IERCHLz+bNm+Hk5ASZTIY2bdrg6tWrRZY/fPgwGjVqBJlMhubNm+OPP/5QOU5EWLhwIWxtbWFgYAAvLy8+0BeDwSgfBIvNyJEjsXXrVrX87du3Y/jw4YIdOHjwIGbOnIlFixYhIiICbm5u8Pb2RnJyssbyf//9N4YOHYpx48bhxo0b6NevH/r164fb+ZYBXbNmDTZu3Iht27YhPDwcRkZG8Pb2RlZphCtgMBhaIXicja+vL/bs2QN7e3u0bdsWABAeHo64uDiMGjVKZcSjv79/sfbatGmD1q1b46effgLANYrZ29vD19cXc+fOVSs/ePBgZGZm4sSJE3xe27Zt4e7ujm3btoGIYGdnh1mzZuH7778HAKSmpsLa2hqBgYEYMmRIsT6xcTaMioDK6grz50MikZSzR+qU6jib27dvo1WrVgCAmJgYAEDt2rVRu3ZtldqFNt3hOTk5uH79OubNm8fnicVieHl54fLlyxrPuXz5MmbOnKmS5+3tzc9Aj42NRWJiIry8vPjjZmZmaNOmDS5fvqxRbLKzs5Gdnc3vp+kqBguDweARLDbnz5/X2cVfvXoFuVwO6wJzd6ytrXH//n2N5yQmJmosn5iYyB9X5hVWpiArV67E4sWLS/QdGIzSQiKRwM/Pr7zd0Blsqi+AefPmITU1lU/x8fHl7RKDUeUoV7GpXbs29PT0kJSUpJKflJQEGxsbjefY2NgUWV65FWJTKpXC1NRUJTEYDN1SrmIjkUjg4eGBkJAQPk+hUCAkJIQfMFiQdu3aqZQHgDNnzvDlnZ2dYWNjo1ImLS0N4eHhhdpkMBhlAJUzBw4cIKlUSoGBgXT37l3y8fEhc3NzSkxMJCKikSNH0ty5c/nyYWFhVKNGDVq3bh3du3ePFi1aRPr6+hQVFcWXWbVqFZmbm1NQUBDdunWL+vbtS87OzvT+/XutfEpNTSUAlJqaqtsvy2BUMYQ8K+UuNkREmzZtIgcHB5JIJOTp6UlXrlzhj3Xq1IlGjx6tUv7QoUPk6upKEomEmjZtSidPnlQ5rlAoaMGCBWRtbU1SqZS6dOlC0dHRWvvDxIbB0A4hz4rgcTbVATbOhsHQjlIdZ1MdUOovG2/DYBSN8hnRps7CxEYDr1+/BgDY29uXsycMRuUgPT0dZmZmRZZhYqOBWrVqAQDi4uKKvYEMzaSlpcHe3h7x8fHsVfQjqOj3kYiQnp4OOzu7YssysdGA+N9lTczMzCrkH7gywcYt6YaKfB+1/YfMRhAzGIwygYkNg8EoE5jYaEAqlWLRokWQSqXl7Uqlhd1D3VCV7iMbZ8NgMMoEVrNhMBhlAhMbBoNRJjCxYTAYZUK1FRshKzoEBgZCJBKpJJlMVobeVkyEroqRkpKCKVOmwNbWFlKpFK6urmorY1RHhNzHzz//XO23KBKJ0LNnzzL0uISU5ozQisqBAwdIIpHQrl276M6dOzRhwgQyNzenpKQkjeUDAgLI1NSUEhIS+KQMgVFdEXoPs7Oz6ZNPPqEePXrQpUuXKDY2lkJDQ+nmzZtl7HnFQuh9fP36tcrv8Pbt26Snp0cBAQFl63gJqJZi4+npSVOmTOH35XI52dnZ0cqVKzWWDwgIIDMzszLyrnIg9B5u3bqVXFxcKCcnp6xcrBQIvY8F2bBhA5mYmFBGRkZpuagzqt1rlHJFh/yrLxS3ogMAZGRkwNHREfb29ujbty/u3LlTFu5WSEpyD3///Xe0a9cOU6ZMgbW1NZo1a4YVK1ZALpeXldsVjpL+FvOzc+dODBkyBEZGRqXlps6odmJT1IoOha2+0LBhQ+zatQtBQUHYu3cvFAoF2rdvj2fPnpWFyxWOktzDx48f47fffoNcLscff/yBBQsWYP369Vi2bFlZuFwhKcl9zM/Vq1dx+/ZtjB8/vrRc1ClsIqYWtGvXTiV+cfv27dG4cWP873//w9KlS8vRs8qDQqGAlZUVtm/fDj09PXh4eOD58+dYu3YtFi1aVN7uVUp27tyJ5s2bw9PTs7xd0YpqV7MpyYoOBdHX10fLli3x6NGj0nCxwlOSe2hrawtXV1fo6enxeY0bN0ZiYiJycnJK1d+Kysf8FjMzM3HgwAGMGzeuNF3UKdVObEqyokNB5HI5oqKiYGtrW1puVmhKcg87dOiAR48eQaFQ8HkPHjyAra1thVxWtiz4mN/i4cOHkZ2djREjRpS2m7qjvFuoywOhKzosXryYTp06RTExMXT9+nUaMmQIyWQyunPnTnl9hXJH6D2Mi4sjExMTmjp1KkVHR9OJEyfIysqKli1bVl5foUIg9D4q6dixIw0ePLis3f0oqqXYEAlb0WH69Ol8WWtra+rRowdFRESUg9cVC6GrYvz999/Upk0bkkql5OLiQsuXL6e8vLwy9rriIfQ+3r9/nwDQ6dOny9jTj4PN+mYwGGVCtWuzYTAY5QMTGwaDUSYwsWEwGGUCExsGg1EmMLFhMBhlAhMbBoNRJjCxYTAYZQITGwaDUSYwsWEwGGUCExstGTNmDPr161dsuWfPnkEikaBZs2Yaj4tEIhw/flwl78SJE+jUqRNMTExgaGiI1q1bIzAwUKXMkydPIBKJYGVlhfT0dJVj7u7u8PPzK9QnPz8/uLu7F+u7ULS1q+29Kw7lPbh58+ZH2yoLxowZA5FIhFWrVqnkHz9+HCKRSJAtJycn/Pjjj8WW0/T70gW6sMvERscEBgZi0KBBSEtLQ3h4eLHlN23ahL59+6JDhw4IDw/HrVu3MGTIEEycOBHff/+9Wvn09HSsW7euNFxnlAIymQyrV6/G27dvy9uV8qe8J2dVFkaPHk19+/YtsoxCoSAXFxcKDg6mOXPm0IQJE9TKAKBjx44RETcTWl9fn2bOnKlWbuPGjQSAn5QXGxtLAGj27NlkbGysEhDbzc2NFi1apNGngIAAAqCSlMGx3759S+PGjaPatWuTiYkJde7cmQ9AnpycTNbW1rR8+XLeVlhYGOnr69PZs2eLtJufRYsWqZU7f/48ERHdunWLOnfuTDKZjGrVqkUTJkyg9PT0Qu+v8h7cuHGDiIjOnz9PACg4OJjc3d1JJpNR586dKSkpif744w9q1KgRmZiY0NChQykzM5O38+eff1KHDh3IzMyMatWqRT179qRHjx6pXCssLIzc3NxIKpWSh4cHHTt2TOXaRERRUVHUrVs3MjIyIisrKxoxYgS9fPmSPz569Gjq1asXNWrUiGbPns3nK23l5+LFi9SxY0eSyWRUt25d8vX15eMKd+rUSe0easLR0VGljKOjI3/s+PHj1LJlS5JKpeTs7Ex+fn6Um5tLRFxUA1tbW3r16hVfvkePHvT555+TXC4v0q4QmNhoiTZiExISQjY2NpSXl0dRUVEaA1HnFxt/f38CQC9evFCzlZ2dTcbGxjRt2jQi+vCgRUREkLu7u0qQ7KLE5t27dzRr1ixq2rQpH5H/3bt3RETk5eVFvXv3pmvXrtGDBw9o1qxZZGFhQa9fvyYiopMnT5K+vj5du3aN0tLSyMXFhWbMmFGs3fykp6fToEGDqFu3bny57OxsysjIIFtbWxowYABFRUVRSEgIOTs7q81wzk9hYtO2bVu6dOkSRUREUP369alTp07UtWtXioiIoAsXLpCFhQWtWrWKt/Pbb7/RkSNH6OHDh3Tjxg3q3bs3NW/enORyORERpaamUq1atWjEiBF0584d+uOPP8jV1VXl2m/fviVLS0uaN28e3bt3jyIiIujLL7+kzp0789dR/maOHj1KMpmM4uPjiUhdbB49ekRGRka0YcMGevDgAYWFhVHLli1pzJgxRMStqFC3bl1asmQJfw81kZyczIt+QkICJScnExHRhQsXyNTUlAIDAykmJoZOnz5NTk5O5OfnR0REeXl51K5dO+rXrx8REf30009kbm5OT58+LdKuUJjYaIk2YjNs2DCaPn06v+/m5qb23z6/2EycOLHIVRtatGhB3bt3JyLVBy04OJj09fX5/8ZFiQ0RV7twc3NTybt48SKZmppSVlaWSn69evXof//7H78/efJkcnV1pWHDhlHz5s1VymuyqwlN92779u1Us2ZNFTE+efIkicXiQpfJKUxszp49y5dZuXIlAaCYmBg+79tvvyVvb+9C/Xv58iUBoKioKCLiVoKwsLCg9+/f82V27Nihcu2lS5dS165dVezEx8cTAIqOjlb73m3btqVvvvmGiNTFZty4ceTj46Ni6+LFiyQWi3kfHB0dacOGDYV+ByX5f19KunTpQitWrFDJ++WXX8jW1pbfj4mJIRMTE5ozZw4ZGBjQvn37irUrFNZmoyNSUlJw9OhRlchpI0aMwM6dO3V+LW9vb3Ts2BELFiwosY3IyEhkZGTAwsICxsbGfIqNjUVMTAxfbt26dcjLy8Phw4exb98+SKVSXXwF3Lt3D25ubiqrAnTo0AEKhQLR0dGCbLVo0YL/bG1tDUNDQ7i4uKjkJScn8/sPHz7E0KFD4eLiAlNTUzg5OQEA4uLiAADR0dFo0aKFykKEBeP8RkZG4vz58yr3rlGjRgCgcv+UrF69Grt378a9e/fUjkVGRiIwMFDFlre3NxQKBWJjYwXdC01ERkZiyZIlKvYnTJiAhIQEvHv3DgDg4uKCdevWYfXq1ejTpw+GDRv20dctCAt4riP279+PrKwstGnThs8jIigUCjx48ACurq5q57i6uiI1NRUvXryAnZ2dyrGcnBzExMSgc+fOGq+3atUqtGvXDrNnzy6RvxkZGbC1tUVoaKjaMXNzc/5zTEwMXrx4AYVCgSdPnqB58+Ylul5poq+vz38WiUQq+8q8/OFIe/fuDUdHR+zYsQN2dnZQKBRo1qyZoFjIGRkZ6N27N1avXq12TFO42M8++wze3t6YN28exowZo2br22+/xXfffad2noODg9Y+FeXr4sWLMWDAALVj+QX1woUL0NPTw5MnT5CXl4caNXQrD0xsdMTOnTsxa9YstR/S5MmTsWvXLrXuTwAYOHAg5syZg/Xr12P9+vUqx7Zt24bMzEwMHTpU4/U8PT0xYMAAzJ07t1jfJBKJ2vpMrVq1QmJiImrUqMH/Zy9ITk4ORowYgcGDB6Nhw4YYP348oqKiYGVlVahdba/fuHFjBAYGIjMzk6/dhIWFQSwWo2HDhsXaLCmvX79GdHQ0duzYgU8//RQAcOnSJZUyDRs2xN69e5Gdnc3X5K5du6ZSplWrVjhy5AicnJy0fihXrVoFd3d3te/XqlUr3L17F/Xr1y/0XG3vtb6+vsa/dXR0dJH2Dx48iKNHjyI0NBSDBg3C0qVLsXjx4iLtCuajXsKqEaNHj6bPP/+cbty4oZLi4uLoxo0bBIDu3bundt6WLVvIxsaGb/lHgXffDRs2kFgspvnz59O9e/fo0aNHtH79epJKpTRr1iy+XMH2CiKi6OhoqlGjBslksiLbbPbt20dGRkZ048YNevnyJWVlZZFCoaCOHTuSm5sbnTp1imJjYyksLIzmz59P165dIyKi77//npycnCg1NZXkcjl17NiRevbsWaRdTSxfvpwcHBzo/v379PLlS8rJyaHMzEyytbWlgQMHUlRUFJ07d45cXFxK1ED89u1bvoym1Uvzty3J5XKysLCgESNG0MOHDykkJIRat26t8ndRNhCPGjWK7t69S8HBwdSoUSMCwPfWPX/+nCwtLemrr76iq1ev0qNHjyg4OJjGjBnDhzrV1FY1cuRIkslkKm02kZGRZGBgQFOmTKEbN27QgwcP6Pjx4yqdAF9++SX16dOHnj17ptLjVZAGDRrQpEmTKCEhgd68eUNERMHBwVSjRg3y8/Oj27dv0927d+nXX3+lH374gYi4tqaaNWvSxo0bVcpfvny5SLtCYWKjJaNHj1brfgRA48aNo6lTp1KTJk00npeQkEBisZiCgoKISHNDW1BQEH366adkZGREMpmMPDw8aNeuXSplNIkNEZGPjw8BKFJssrKyaODAgWRubq7SRZ2Wlka+vr5kZ2dH+vr6ZG9vT8OHD6e4uDg6f/481ahRgy5evKjig6mpKW3ZsqVIuwVJTk6mL7/8koyNjUul61uI2BARnTlzhho3bkxSqZRatGhBoaGhan+XsLAwatGiBUkkEvLw8KD9+/cTALp//z5f5sGDB9S/f38yNzcnAwMDatSoEU2fPp0UCgURaRab2NhYkkgkat3XV69e5e+RkZERtWjRQmXYweXLl6lFixYklUoL7fomIvr999+pfv36VKNGDZUu6uDgYGrfvj0ZGBiQqakpeXp60vbt20mhUFCXLl3I29ub95uIyNfXl+rVq8f/PQqzKwQWg5jB0IJ9+/Zh7NixSE1NhYGBQXm7UylhbTYMhgb27NkDFxcX1KlTB5GRkZgzZw4GDRrEhOYjYGLDYGggMTERCxcuRGJiImxtbfH1119j+fLl5e1WpYa9RjEYjDKBDepjMBhlAhMbRoUiMDBQZVBhSckfEkGXoSm0DfXAUIeJTTUjMTERvr6+cHFxgVQqhb29PXr37q2yuH1RD1TBB1e5r0wWFhbo2rUrbty4USL/Bg8ejAcPHpTo3PwkJCSge/fuH22nINeuXYOPjw+/X1rxY6oiTGyqEU+ePIGHhwfOnTuHtWvXIioqCsHBwejcuTOmTJnyUbbPnj2LhIQEnDp1ChkZGejevTtSUlIE2zEwMOBHKH8MNjY2OpvHBYCfymBpaQlDQ0Od2a1OMLGpRkyePBkikQhXr17FwIED4erqiqZNm2LmzJm4cuXKR9m2sLCAjY0NPvnkE6xbtw5JSUmFBg+LjIxE586dYWJiAlNTU3h4eOCff/4BoP4apYwGuGvXLjg4OMDY2BiTJ0+GXC7HmjVrYGNjAysrK7WeoqJqHHK5HOPGjYOzszMMDAzQsGFD/Pe//1Upo4wuuHz5ctjZ2fFTDPLX+pTTPPr37w+RSAQnJyc8efIEYrGY/z5KfvzxRzg6OqrM0apusK7vasKbN28QHByM5cuXq8y0VqKLdhIlyrEohU1sHD58OFq2bImtW7dCT08PN2/eVJs8mZ+YmBj8+eefCA4ORkxMDL766is8fvwYrq6u+Ouvv/D333/jm2++gZeXl8pE2MJQKBSoW7cuDh8+DAsLC/z999/w8fGBra0tBg0axJcLCQmBqakpzpw5o9HOtWvXYGVlhYCAAHTr1g16enqwtLSEl5cXAgIC8Mknn/BlAwICMGbMGIjF1ff/OxObasKjR49ARHwYhNIiJSUFS5cuhbGxsVpYBiVxcXGYPXs270uDBg2KtKlQKLBr1y6YmJigSZMm6Ny5M6Kjo/HHH3/wEzdXr16N8+fPayU2+vr6KpMMnZ2dcfnyZRw6dEhFbIyMjPDzzz9DIpFotGNpaQmAE2obGxs+f/z48Zg4cSL8/f0hlUoRERGBqKgoBAUFFetbVab6ymw1o7SHU7Vv3x7GxsaoWbMmIiMjcfDgQVhbW2ssO3PmTIwfPx5eXl5YtWqVxvgv+XFycoKJiQm/b21tjSZNmqjUEgrGrCmOzZs3w8PDA5aWljA2Nsb27dv5eDZKmjdvXqjQFEW/fv2gp6eHY8eOAeBeDTt37lzo7PrqAhObakKDBg0gEolw//79UrF/8OBBREZG4u3bt4iJiUGPHj0KLevn54c7d+6gZ8+eOHfuHJo0acI/mJrQFJ+muJg1RXHgwAF8//33GDduHE6fPo2bN29i7Nixaq99ml43tUEikWDUqFEICAhATk4O9u/fj2+++aZEtqoSTGyqCbVq1YK3tzc2b96MzMxMteMl6TnKj729PerVq6d124+rqytmzJiB06dPY8CAAQgICPio6wshLCwM7du3x+TJk9GyZUvUr1+/2NpVYRQW52X8+PE4e/YstmzZgry8PI2Bq6obTGyqEZs3b4ZcLoenpyeOHDmChw8f4t69e9i4cSPatWunUvb58+e4efOmStLFciTv37/H1KlTERoaiqdPnyIsLAzXrl1D48aNP9q2tjRo0AD//PMPTp06hQcPHmDBggVqwbG0xcnJCSEhIUhMTFS5P40bN0bbtm0xZ84cDB06lE3gBBObaoWLiwsiIiLQuXNnzJo1C82aNcOXX36JkJAQbN26VaXsunXr0LJlS5V08uTJj/ZBT08Pr1+/xqhRo+Dq6opBgwahe/fuKg22pc23336LAQMGYPDgwWjTpg1ev36NyZMnl8jW+vXrcebMGdjb26Nly5Yqx8aNG4ecnBz2CvUvbCImg1FKLF26FIcPH8atW7fK25UKAavZMBg6JiMjA7dv38ZPP/0EX1/f8nanwsDEhsHQMVOnToWHhwc+//xz9gqVD/YaxWAwygRWs2EwGGUCExsGg1EmMLFhMBhlAhMbBoNRJjCxYTAYZQITGwaDUSYwsWEwGGUCExsGg1EmMLFhMBhlwv8HrwJsHgRvvNoAAAAASUVORK5CYII=",
      "text/plain": [
       "<Figure size 300x250 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(3, 2.5))\n",
    "\n",
    "plt.plot(tt_sim_ths, prop_ic_ms_has_lower_avgs, 'b')\n",
    "plt.plot(tt_sim_ths, prop_ic_has_lower_avgs, 'r')\n",
    "plt.plot([chosen_sim_th]*2, [0, 1], 'k--', alpha=0.5)\n",
    "plt.xlim([tt_sim_ths[0], tt_sim_ths[-1]])\n",
    "plt.xlabel('LAION text to ImageNet text\\nCLIP similarity')\n",
    "plt.ylabel('proportion of classes')\n",
    "\n",
    "plt.legend(['LAION < ImageNet', 'LAION > ImageNet'])\n",
    "\n",
    "plt.tight_layout()\n",
    "\n",
    "plt.savefig(os.path.join('figs', 'prop_lower_sim_ic_most_similar.pdf'), bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Compare intra-class cross image similarities for ImageNetV2"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Settings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "sm_prefix = configs.LAIONConfig.SUBSET_SM_FILTERED_PREFIX\n",
    "iv2b_prefix = 'imagenetv2-matched-frequency_'\n",
    "iv2a_prefix = 'imagenetv2-threshold0.7_'\n",
    "iv2c_prefix = 'imagenetv2-top-images_'"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load similarities"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|███████████████████████████████████████████████████████████████████████████████████████████████████| 999/999 [00:02<00:00, 453.79it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "899\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "wnid2smsims = {}\n",
    "wnid2iv2asims = {}\n",
    "wnid2iv2bsims = {}\n",
    "wnid2iv2csims = {}\n",
    "\n",
    "for wnid in tqdm(all_wnids):\n",
    "    sm_sim_path = os.path.join(cross_image_sim_path, sm_prefix + cross_image_sim_prefix + f'({wnid}).pkl')\n",
    "    iv2a_sim_path = os.path.join(cross_image_sim_path, iv2a_prefix + cross_image_sim_prefix + f'({wnid}).pkl')\n",
    "    iv2b_sim_path = os.path.join(cross_image_sim_path, iv2b_prefix + cross_image_sim_prefix + f'({wnid}).pkl')\n",
    "    iv2c_sim_path = os.path.join(cross_image_sim_path, iv2c_prefix + cross_image_sim_prefix + f'({wnid}).pkl')\n",
    "    \n",
    "    if (not os.path.exists(sm_sim_path)) or (not os.path.exists(iv2a_sim_path)) or (not os.path.exists(iv2b_sim_path)) or (not os.path.exists(iv2c_sim_path)):\n",
    "        continue\n",
    "\n",
    "    with open(sm_sim_path, 'rb') as f:\n",
    "        sm_sim_dict = pickle.load(f)\n",
    "    with open(iv2a_sim_path, 'rb') as f:\n",
    "        iv2a_sim_dict = pickle.load(f) \n",
    "    with open(iv2b_sim_path, 'rb') as f:\n",
    "        iv2b_sim_dict = pickle.load(f) \n",
    "    with open(iv2c_sim_path, 'rb') as f:\n",
    "        iv2c_sim_dict = pickle.load(f)\n",
    "        \n",
    "    if not isinstance(sm_sim_dict, dict):\n",
    "        continue\n",
    "        \n",
    "    sm_sim = sm_sim_dict['similarities']\n",
    "    iv2a_sim = iv2a_sim_dict['similarities']\n",
    "    iv2b_sim = iv2b_sim_dict['similarities']\n",
    "    iv2c_sim = iv2c_sim_dict['similarities']\n",
    "        \n",
    "    # Extract sims.\n",
    "    wnid2smsims[wnid] = vec(sm_sim)\n",
    "    wnid2iv2asims[wnid] = vec(iv2a_sim)\n",
    "    wnid2iv2bsims[wnid] = vec(iv2b_sim)\n",
    "    wnid2iv2csims[wnid] = vec(iv2c_sim)\n",
    "    \n",
    "print(len(wnid2smsims))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Resample"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "rs_sm_sims = []\n",
    "rs_iv2a_sims = []\n",
    "rs_iv2b_sims = []\n",
    "rs_iv2c_sims = []\n",
    "\n",
    "np.random.seed(111)\n",
    "for wnid, sm_sims in wnid2smsims.items():\n",
    "    iv2a_sims = wnid2iv2asims[wnid]\n",
    "    iv2b_sims = wnid2iv2bsims[wnid]\n",
    "    iv2c_sims = wnid2iv2csims[wnid]\n",
    "    \n",
    "    n_s_i = np.min([len(iv2a_sims), len(iv2b_sims), len(iv2c_sims), 100])\n",
    "        \n",
    "    rs_sm_sims.extend(np.random.choice(sm_sims, size=(n_s_i,), replace=True))\n",
    "    rs_iv2a_sims.extend(np.random.choice(iv2a_sims, size=(n_s_i,), replace=True))\n",
    "    rs_iv2b_sims.extend(np.random.choice(iv2b_sims, size=(n_s_i,), replace=True))\n",
    "    rs_iv2c_sims.extend(np.random.choice(iv2c_sims, size=(n_s_i,), replace=True))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Plot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 340x250 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(3.4, 2.5))\n",
    "\n",
    "# _ = plt.hist(rs_sm_sims, bins=40, color='b', alpha=0.3, density=True)\n",
    "# _ = plt.hist(rs_iv2a_sims, bins=40, color='b', alpha=0.3, density=True)\n",
    "_ = plt.hist(rs_iv2b_sims, bins=40, color='g', alpha=0.3, density=True)\n",
    "_ = plt.hist(rs_iv2c_sims, bins=40, color='r', alpha=0.3, density=True)\n",
    "\n",
    "plt.legend(['ImageNetV2-b', 'ImageNetV2-c'])\n",
    "\n",
    "plt.xlabel('intra-class image-image CLIP similarity')\n",
    "plt.ylabel('\\npdf')\n",
    "\n",
    "plt.xlim([0, 1])\n",
    "\n",
    "plt.tight_layout()\n",
    "\n",
    "plt.savefig(os.path.join('figs', f'{sm_prefix}{iv2b_prefix}{iv2c_prefix}{cross_image_sim_prefix}_resampled.pdf'), bbox_inches='tight')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 158,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 340x250 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(3.4, 2.5))\n",
    "\n",
    "_ = plt.hist(rs_sm_sims, bins=40, color='b', alpha=0.3, density=True)\n",
    "_ = plt.hist(rs_iv2b_sims, bins=40, color='g', alpha=0.3, density=True)\n",
    "\n",
    "plt.legend(['LAIONet', 'ImageNetV2-c'])\n",
    "\n",
    "plt.xlabel('intra-class image-image CLIP similarity')\n",
    "plt.ylabel('\\npdf')\n",
    "\n",
    "plt.xlim([0, 1])\n",
    "\n",
    "plt.tight_layout()\n",
    "\n",
    "plt.savefig(os.path.join('figs', f'{sm_prefix}{iv2c_prefix}{cross_image_sim_prefix}_resampled.pdf'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Test per WNID"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████████████████████████████████████████████████████████████████████████████████████████████| 899/899 [00:00<00:00, 2396.52it/s]\n"
     ]
    }
   ],
   "source": [
    "sm_avg_intrasims = []\n",
    "sm_err_intrasims = []\n",
    "ic_avg_intrasims = []\n",
    "ic_err_intrasims = []\n",
    "iv2a_avg_intrasims = []\n",
    "iv2a_err_intrasims = []\n",
    "iv2b_avg_intrasims = []\n",
    "iv2b_err_intrasims = []\n",
    "iv2c_avg_intrasims = []\n",
    "iv2c_err_intrasims = []\n",
    "\n",
    "for wnid, sm_intrasims in tqdm(wnid2smsims.items()):\n",
    "    if wnid not in wnid2icsims:\n",
    "        continue\n",
    "        \n",
    "    ic_intrasims = wnid2icsims[wnid]\n",
    "    iv2a_intrasims = wnid2iv2asims[wnid]\n",
    "    iv2b_intrasims = wnid2iv2bsims[wnid]\n",
    "    iv2c_intrasims = wnid2iv2csims[wnid]\n",
    "    \n",
    "    if len(sm_intrasims) < 7 or len(iv2a_intrasims) < 7 or len(iv2b_intrasims) < 7 or len(iv2c_intrasims) < 7:\n",
    "        continue\n",
    "        \n",
    "    sm_avg_intrasims.append(np.mean(sm_intrasims))\n",
    "    sm_err_intrasims.append(1.96*np.std(sm_intrasims)/np.sqrt(len(sm_intrasims) - 1))\n",
    "    \n",
    "    ic_avg_intrasims.append(np.mean(ic_intrasims))\n",
    "    ic_err_intrasims.append(1.96*np.std(ic_intrasims)/np.sqrt(len(ic_intrasims) - 1))\n",
    "    \n",
    "    iv2a_avg_intrasims.append(np.mean(iv2a_intrasims))\n",
    "    iv2a_err_intrasims.append(1.96*np.std(iv2a_intrasims)/np.sqrt(len(iv2a_intrasims) - 1))\n",
    "    \n",
    "    iv2b_avg_intrasims.append(np.mean(iv2b_intrasims))\n",
    "    iv2b_err_intrasims.append(1.96*np.std(iv2b_intrasims)/np.sqrt(len(iv2b_intrasims) - 1))\n",
    "    \n",
    "    iv2c_avg_intrasims.append(np.mean(iv2c_intrasims))\n",
    "    iv2c_err_intrasims.append(1.96*np.std(iv2c_intrasims)/np.sqrt(len(iv2c_intrasims) - 1))\n",
    "    \n",
    "sm_avg_intrasims = np.array(sm_avg_intrasims)\n",
    "sm_err_intrasims = np.array(sm_err_intrasims)\n",
    "ic_avg_intrasims = np.array(ic_avg_intrasims)\n",
    "ic_err_intrasims = np.array(ic_err_intrasims)\n",
    "iv2a_avg_intrasims = np.array(iv2a_avg_intrasims)\n",
    "iv2a_err_intrasims = np.array(iv2a_err_intrasims)\n",
    "iv2b_avg_intrasims = np.array(iv2b_avg_intrasims)\n",
    "iv2b_err_intrasims = np.array(iv2b_err_intrasims)\n",
    "iv2c_avg_intrasims = np.array(iv2c_avg_intrasims)\n",
    "iv2c_err_intrasims = np.array(iv2c_err_intrasims)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 179,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "avg < 0: 0.4765258215962441\n",
      "UCB < 0: 273 out of 852 (0.3204225352112676)\n",
      "LCB > 0: 326 out of 852 (0.3826291079812207)\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 340x250 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(3.4, 2.5))\n",
    "\n",
    "# Change the next two lines + legened + file name for different ver of ImageNetV2\n",
    "diff_avg_intrasims = sm_avg_intrasims - iv2b_avg_intrasims\n",
    "diff_err_intrasims = np.sqrt(sm_err_intrasims**2 + iv2b_err_intrasims**2)\n",
    "\n",
    "sorted_locs = np.argsort(diff_avg_intrasims)\n",
    "\n",
    "ucb_intrasims = np.sort(diff_avg_intrasims + diff_err_intrasims)\n",
    "lcb_intrasims = np.sort(diff_avg_intrasims - diff_err_intrasims)\n",
    "\n",
    "x = range(len(sorted_locs))\n",
    "\n",
    "plt.plot(x, diff_avg_intrasims[sorted_locs], color='k')\n",
    "plt.plot(x, ucb_intrasims, 'b')\n",
    "plt.plot(x, lcb_intrasims, 'r')\n",
    "\n",
    "plt.plot(x, [0]*len(x), 'k--', alpha=0.5)\n",
    "\n",
    "# Change here\n",
    "plt.legend(['LAIONet - ImageNetV2-b', 'UCB', 'LCB'])\n",
    "\n",
    "plt.xlabel(' ')\n",
    "plt.ylabel(r'$\\Delta$ intra-class image-image' + '\\n' + 'CLIP similarity')\n",
    "\n",
    "plt.tight_layout()\n",
    "\n",
    "# Change here\n",
    "plt.savefig(os.path.join('figs', f'intra-class_image-image_sim_diff_{iv2b_prefix[:-1]}.pdf'), bbox_inches='tight')\n",
    "\n",
    "print(f'avg < 0: {np.mean(diff_avg_intrasims < 0)}')\n",
    "print(f'UCB < 0: {np.sum(ucb_intrasims < 0)} out of {len(ucb_intrasims)} ({np.mean(ucb_intrasims < 0)})')\n",
    "print(f'LCB > 0: {np.sum(lcb_intrasims > 0)} out of {len(lcb_intrasims)} ({np.mean(lcb_intrasims > 0)})')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Vs. MTurk selection frequency"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 340x250 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "diff_avg_intrasims = sm_avg_intrasims - iv2a_avg_intrasims\n",
    "diff_err_intrasims = np.sqrt(sm_err_intrasims**2 + iv2a_err_intrasims**2)\n",
    "ucb_a_intrasims = np.sort(diff_avg_intrasims + diff_err_intrasims)\n",
    "lcb_a_intrasims = np.sort(diff_avg_intrasims - diff_err_intrasims)\n",
    "\n",
    "diff_avg_intrasims = sm_avg_intrasims - iv2b_avg_intrasims\n",
    "diff_err_intrasims = np.sqrt(sm_err_intrasims**2 + iv2b_err_intrasims**2)\n",
    "ucb_b_intrasims = np.sort(diff_avg_intrasims + diff_err_intrasims)\n",
    "lcb_b_intrasims = np.sort(diff_avg_intrasims - diff_err_intrasims)\n",
    "\n",
    "diff_avg_intrasims = sm_avg_intrasims - iv2c_avg_intrasims\n",
    "diff_err_intrasims = np.sqrt(sm_err_intrasims**2 + iv2c_err_intrasims**2)\n",
    "ucb_c_intrasims = np.sort(diff_avg_intrasims + diff_err_intrasims)\n",
    "lcb_c_intrasims = np.sort(diff_avg_intrasims - diff_err_intrasims)\n",
    "\n",
    "mturk_selec_freq = {'b': 0.73, 'a': 0.85, 'c': 0.93}\n",
    "\n",
    "plt.figure(figsize=(3.4, 2.5))\n",
    "\n",
    "plt.plot(\n",
    "    [v for k, v in mturk_selec_freq.items()], \n",
    "    [\n",
    "        np.mean(ucb_b_intrasims < 0),\n",
    "        np.mean(ucb_a_intrasims < 0),\n",
    "        np.mean(ucb_c_intrasims < 0),\n",
    "    ],\n",
    "    'b',\n",
    "    marker='.'\n",
    ")\n",
    "\n",
    "plt.plot(\n",
    "    [v for k, v in mturk_selec_freq.items()], \n",
    "    [\n",
    "        np.mean(lcb_b_intrasims > 0),\n",
    "        np.mean(lcb_a_intrasims > 0),\n",
    "        np.mean(lcb_c_intrasims > 0),\n",
    "    ],\n",
    "    'g',\n",
    "    marker='.'\n",
    ")\n",
    "\n",
    "plt.xlabel('MTurk selection frequency')\n",
    "plt.ylabel('\\nproportion of classes')\n",
    "\n",
    "plt.legend(['LAIONet < ImageNetV2', 'LAIONet > ImageNetV2'])\n",
    "\n",
    "plt.ylim([0, 1])\n",
    "\n",
    "plt.tight_layout()\n",
    "\n",
    "plt.savefig(os.path.join('figs', f'prop_lower_sim_imagenetv2.pdf'), bbox_inches='tight')\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Vs. ImageNet"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 340x250 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "diff_avg_intrasims = ic_avg_intrasims - iv2a_avg_intrasims\n",
    "diff_err_intrasims = np.sqrt(ic_err_intrasims**2 + iv2a_err_intrasims**2)\n",
    "ucb_a_intrasims = np.sort(diff_avg_intrasims + diff_err_intrasims)\n",
    "lcb_a_intrasims = np.sort(diff_avg_intrasims - diff_err_intrasims)\n",
    "\n",
    "diff_avg_intrasims = ic_avg_intrasims - iv2b_avg_intrasims\n",
    "diff_err_intrasims = np.sqrt(ic_err_intrasims**2 + iv2b_err_intrasims**2)\n",
    "ucb_b_intrasims = np.sort(diff_avg_intrasims + diff_err_intrasims)\n",
    "lcb_b_intrasims = np.sort(diff_avg_intrasims - diff_err_intrasims)\n",
    "\n",
    "diff_avg_intrasims = ic_avg_intrasims - iv2c_avg_intrasims\n",
    "diff_err_intrasims = np.sqrt(ic_err_intrasims**2 + iv2c_err_intrasims**2)\n",
    "ucb_c_intrasims = np.sort(diff_avg_intrasims + diff_err_intrasims)\n",
    "lcb_c_intrasims = np.sort(diff_avg_intrasims - diff_err_intrasims)\n",
    "\n",
    "mturk_selec_freq = {'b': 0.73, 'a': 0.85, 'c': 0.93}\n",
    "\n",
    "plt.figure(figsize=(3.4, 2.5))\n",
    "\n",
    "plt.plot(\n",
    "    [v for k, v in mturk_selec_freq.items()], \n",
    "    [\n",
    "        np.mean(ucb_b_intrasims < 0),\n",
    "        np.mean(ucb_a_intrasims < 0),\n",
    "        np.mean(ucb_c_intrasims < 0),\n",
    "    ],\n",
    "    'r',\n",
    "    marker='.'\n",
    ")\n",
    "\n",
    "plt.plot(\n",
    "    [v for k, v in mturk_selec_freq.items()], \n",
    "    [\n",
    "        np.mean(lcb_b_intrasims > 0),\n",
    "        np.mean(lcb_a_intrasims > 0),\n",
    "        np.mean(lcb_c_intrasims > 0),\n",
    "    ],\n",
    "    'g',\n",
    "    marker='.'\n",
    ")\n",
    "\n",
    "plt.xlabel('MTurk selection frequency')\n",
    "plt.ylabel('\\nproportion of classes')\n",
    "\n",
    "plt.legend(['ImageNet < ImageNetV2', 'ImageNet > ImageNetV2'])\n",
    "\n",
    "plt.ylim([0, 1])\n",
    "\n",
    "plt.tight_layout()\n",
    "\n",
    "plt.savefig(os.path.join('figs', f'prop_lower_sim_imagenetv2_vs_imagenet.pdf'), bbox_inches='tight')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "laionvenv",
   "language": "python",
   "name": "laionvenv"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}