seq-tagging.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "c31bef24",
   "metadata": {},
   "outputs": [],
   "source": [
    "import localized_narratives\n",
    "import os\n",
    "import glob\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib.image as mpimg\n",
    "import numpy as np\n",
    "import pickle as pkl\n",
    "from scipy.spatial import ConvexHull\n",
    "from matplotlib.path import Path"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d81b912a",
   "metadata": {},
   "source": [
    "<h1>Step 0: Visualise Dataset</h1>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b41c2c15",
   "metadata": {},
   "source": [
    "<h3>I) Get Localized Narratives (ADE20k)</h3>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "1daa5ab1",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Already downloaded: ade20k_validation_localized_narratives.jsonl\n",
      "\n",
      "Localized Narrative sample:\n",
      "{\n",
      " dataset_id: ADE20k,\n",
      " image_id: ADE_val_00000175,\n",
      " annotator_id: 125,\n",
      " caption: In this image on the left side I can see a bed and a window....,\n",
      " timed_caption: [{'utterance': 'In', 'start_time': 0.0, 'end_time': 0.0}, ...],\n",
      " traces: [[{'x': 0.6408, 'y': 0.1371, 't': 0.013}, ...], ...],\n",
      " voice_recording: ade20k_validation/ade20k_validation_ADE_val_00000175_125.ogg\n",
      "}\n",
      "\n",
      "Voice recording URL:\n",
      " https://storage.googleapis.com/localized-narratives/voice-recordings/ade20k_validation/ade20k_validation_0000000000000175_125.ogg\n",
      "\n"
     ]
    }
   ],
   "source": [
    "\"\"\"https://github.com/google/localized-narratives/blob/master/demo.py\"\"\"\n",
    "\n",
    "# This folder is where you would like to download the annotation files to and\n",
    "# where to read them from.\n",
    "local_dir = os.getcwd()\n",
    "\n",
    "# The DataLoader class allows us to download the data and read it from file.\n",
    "data_loader = localized_narratives.DataLoader(local_dir)\n",
    "\n",
    "# Downloads the annotation files (it first checks if they are not downloaded).\n",
    "data_loader.download_annotations('ade20k_val')\n",
    "\n",
    "# Iterates through all or a limited number of (e.g. 1 in this case) annotations\n",
    "# for all files found in the local folder for a given dataset and split. E.g.\n",
    "# for `open_images_train` it will read only one shard if only one file was\n",
    "# downloaded manually.\n",
    "loc_narr = next(data_loader.load_annotations('ade20k_val', 1))\n",
    "\n",
    "print(f'\\nLocalized Narrative sample:\\n{loc_narr}')\n",
    "\n",
    "print(f'\\nVoice recording URL:\\n {loc_narr.voice_recording_url}\\n')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "60eea289",
   "metadata": {},
   "source": [
    "<b>Set paths</b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "3d36537c",
   "metadata": {},
   "outputs": [],
   "source": [
    "image_id = 'ADE_val_00000175'\n",
    "path_to_image = f'/Volumes/gordonssd/ADE20K_2021_17_01/images/ADE/validation/*/*/{image_id}.jpg'\n",
    "path_to_image_seg = f'/Volumes/gordonssd/ADE20K_2021_17_01/images/ADE/validation/*/*/{image_id}_seg.png'\n",
    "path_to_image_json = f'/Volumes/gordonssd/ADE20K_2021_17_01/images/ADE/validation/*/*/{image_id}.json'"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bf3ceb33",
   "metadata": {},
   "source": [
    "<h3>II) Show Image from ADE20k</h3>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "b34afdff",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "for filename in glob.glob(path_to_image):\n",
    "    path_to_image = filename\n",
    "\n",
    "image = mpimg.imread(path_to_image)\n",
    "plt.imshow(image)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "f10518a8",
   "metadata": {},
   "outputs": [],
   "source": [
    "image_height = image.shape[0] #1024\n",
    "image_width = image.shape[1] #768"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b69ecff9",
   "metadata": {},
   "source": [
    "<h3>III) Show Image Segmentation from ADE20k"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "b710cb10",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "for filename in glob.glob(path_to_image_seg):\n",
    "    path_to_image_seg = filename\n",
    "image_seg = mpimg.imread(path_to_image_seg)\n",
    "plt.imshow(image_seg)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "577b6496",
   "metadata": {},
   "source": [
    "<h3>IV) Convert to Object Labels</h3>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "35ef4ac3",
   "metadata": {},
   "source": [
    "From https://github.com/CSAILVision/ADE20K/blob/main/utils/utils_ade20k.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "e700c239",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[  0,   0, 447, ..., 447,   0,   0],\n",
       "       [  0,   0, 447, ..., 447,   0,   0],\n",
       "       [  0,   0, 447, ..., 447,   0,   0],\n",
       "       ...,\n",
       "       [  0,   0, 165, ..., 491, 491,   0],\n",
       "       [  0,   0, 165, ..., 491, 491,   0],\n",
       "       [  0,   0,   0, ...,   0,   0,   0]], dtype=int32)"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from PIL import Image\n",
    "with Image.open(path_to_image_seg) as io:\n",
    "        seg = np.array(io);\n",
    "R = seg[:,:,0];\n",
    "G = seg[:,:,1];\n",
    "B = seg[:,:,2];\n",
    "ObjectClassMasks = (R/10).astype(np.int32)*256+(G.astype(np.int32));\n",
    "ObjectClassMasks"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5c28b249",
   "metadata": {},
   "source": [
    "<h3>V) Object Name Lookup by Label</h3>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "01ac5d0c",
   "metadata": {},
   "outputs": [],
   "source": [
    "object_id = ObjectClassMasks[900][200] # 'bed' by visual inspection"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "bafb9685",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Get object name lookup table\n",
    "with open('/Volumes/gordonssd/ADE20K_2021_17_01/index_ade20k.pkl', 'rb') as f:\n",
    "    index_ade20k = pkl.load(f)\n",
    "# Replace \"-\" with \"other\"\n",
    "index_ade20k['objectnames'][0] = \"other\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "670945ed",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['other',\n",
       " 'aarm panel',\n",
       " 'abacus',\n",
       " 'accordion, piano accordion, squeeze box',\n",
       " 'acropolis']"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "index_ade20k['objectnames'][:5] # only showing first 5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "cd06b485",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'bed'"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# indexing is ahead by one\n",
    "index_ade20k['objectnames'][object_id - 1]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5f231145",
   "metadata": {},
   "source": [
    "<h1>Step 1: Word-trace alignments and segmentation mask</h1>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "7852f994",
   "metadata": {},
   "source": [
    "<h3>I) Get traces</h3>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "ac1dd51d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[{'x': 0.6408, 'y': 0.1371, 't': 0.013},\n",
       " {'x': 0.6408, 'y': 0.2455, 't': 0.029},\n",
       " {'x': 0.6354, 'y': 0.3292, 't': 0.045},\n",
       " {'x': 0.6152, 'y': 0.4101, 't': 0.062},\n",
       " {'x': 0.6098, 'y': 0.4897, 't': 0.079}]"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Concatenate all trace segments\n",
    "import itertools\n",
    "traces = list(itertools.chain.from_iterable(loc_narr.traces))\n",
    "traces[:5] # only showing first 5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "3fdab07e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "x = [trace['x'] for trace in traces]\n",
    "y = [trace['y'] for trace in traces]\n",
    "\n",
    "plt.scatter(x, y)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "729a5114",
   "metadata": {},
   "source": [
    "<h3>II) Align traces to word/phrase</h3>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "5565100e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'In this image on the left side I can see a bed and a window. On right side, I can see some object on the table and a door. At the top I can see the light.'"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "loc_narr.caption"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "ab372b24",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[{'utterance': 'In', 'start_time': 0.0, 'end_time': 0.0},\n",
       " {'utterance': 'this', 'start_time': 0.0, 'end_time': 0.8},\n",
       " {'utterance': 'image', 'start_time': 0.8, 'end_time': 1.0},\n",
       " {'utterance': 'on', 'start_time': 1.0, 'end_time': 1.4},\n",
       " {'utterance': 'the', 'start_time': 1.4, 'end_time': 1.6},\n",
       " {'utterance': 'left', 'start_time': 1.6, 'end_time': 2.1},\n",
       " {'utterance': 'side', 'start_time': 2.1, 'end_time': 2.3},\n",
       " {'utterance': 'I', 'start_time': 2.3, 'end_time': 2.6},\n",
       " {'utterance': 'can', 'start_time': 2.6, 'end_time': 3.0},\n",
       " {'utterance': 'see', 'start_time': 3.0, 'end_time': 3.3},\n",
       " {'utterance': 'a', 'start_time': 3.3, 'end_time': 4.4},\n",
       " {'utterance': 'bed', 'start_time': 4.4, 'end_time': 5.2},\n",
       " {'utterance': 'and', 'start_time': 5.2, 'end_time': 7.5},\n",
       " {'utterance': 'a', 'start_time': 7.5, 'end_time': 7.5},\n",
       " {'utterance': 'window.', 'start_time': 7.5, 'end_time': 8.1},\n",
       " {'utterance': 'On', 'start_time': 8.1, 'end_time': 11.5},\n",
       " {'utterance': 'right', 'start_time': 11.6, 'end_time': 12.3},\n",
       " {'utterance': 'side,', 'start_time': 12.3, 'end_time': 12.4},\n",
       " {'utterance': 'I', 'start_time': 12.4, 'end_time': 12.7},\n",
       " {'utterance': 'can', 'start_time': 12.7, 'end_time': 13.1},\n",
       " {'utterance': 'see', 'start_time': 13.1, 'end_time': 13.2},\n",
       " {'utterance': 'some', 'start_time': 13.2, 'end_time': 16.4},\n",
       " {'utterance': 'object', 'start_time': 16.4, 'end_time': 16.9},\n",
       " {'utterance': 'on', 'start_time': 16.9, 'end_time': 17.1},\n",
       " {'utterance': 'the', 'start_time': 17.1, 'end_time': 17.1},\n",
       " {'utterance': 'table', 'start_time': 17.1, 'end_time': 17.3},\n",
       " {'utterance': 'and', 'start_time': 17.3, 'end_time': 19.6},\n",
       " {'utterance': 'a', 'start_time': 19.6, 'end_time': 19.7},\n",
       " {'utterance': 'door.', 'start_time': 19.7, 'end_time': 19.9},\n",
       " {'utterance': 'At', 'start_time': 19.9, 'end_time': 20.6},\n",
       " {'utterance': 'the', 'start_time': 20.6, 'end_time': 20.8},\n",
       " {'utterance': 'top', 'start_time': 20.8, 'end_time': 20.9},\n",
       " {'utterance': 'I', 'start_time': 20.9, 'end_time': 21.2},\n",
       " {'utterance': 'can', 'start_time': 21.2, 'end_time': 21.5},\n",
       " {'utterance': 'see', 'start_time': 21.5, 'end_time': 21.7},\n",
       " {'utterance': 'the', 'start_time': 21.7, 'end_time': 21.9},\n",
       " {'utterance': 'light.', 'start_time': 21.9, 'end_time': 22.2}]"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "timed_caption = loc_narr.timed_caption\n",
    "timed_caption"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "97580006",
   "metadata": {},
   "source": [
    "<b>Get list of word_trace_align objects</b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "5292b3c6",
   "metadata": {},
   "outputs": [],
   "source": [
    "class word_trace_align:\n",
    "    word = ''\n",
    "    start_end = (0.0, 0.8)\n",
    "    word_traces = [{'x': 0.6408, 'y': 0.1371, 't': 0.013}]\n",
    "    hull = None\n",
    "    tag = 'dog'\n",
    "    \n",
    "    def __init__(self, w, start_end):\n",
    "        self.word=w\n",
    "        self.start_end = start_end"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "b0fa7987",
   "metadata": {},
   "outputs": [],
   "source": [
    "import string\n",
    "def clean_text(text):\n",
    "    return text.translate(str.maketrans('', '', string.punctuation)).lower()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "150c2c83",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'inthis'"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "clean_text('In;this?!')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "2ec04973",
   "metadata": {},
   "outputs": [],
   "source": [
    "narrative = []"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "31beb8e8",
   "metadata": {},
   "outputs": [],
   "source": [
    "save_word = ''\n",
    "for word in timed_caption:\n",
    "    w = save_word + word['utterance']\n",
    "    w = clean_text(w)\n",
    "    save_word = ''\n",
    "    # Check if start_time match end_time\n",
    "    if (word['start_time']==word['end_time']):\n",
    "        # Make phrase with next word\n",
    "        save_word = word['utterance'] + ' '\n",
    "        continue\n",
    "    narrative.append(word_trace_align(w, (word['start_time'], word['end_time'])))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "c8639443",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "in this (0.0, 0.8)\n",
      "image (0.8, 1.0)\n",
      "on (1.0, 1.4)\n",
      "the (1.4, 1.6)\n",
      "left (1.6, 2.1)\n",
      "side (2.1, 2.3)\n",
      "i (2.3, 2.6)\n",
      "can (2.6, 3.0)\n",
      "see (3.0, 3.3)\n",
      "a (3.3, 4.4)\n",
      "bed (4.4, 5.2)\n",
      "and (5.2, 7.5)\n",
      "a window (7.5, 8.1)\n",
      "on (8.1, 11.5)\n",
      "right (11.6, 12.3)\n",
      "side (12.3, 12.4)\n",
      "i (12.4, 12.7)\n",
      "can (12.7, 13.1)\n",
      "see (13.1, 13.2)\n",
      "some (13.2, 16.4)\n",
      "object (16.4, 16.9)\n",
      "on (16.9, 17.1)\n",
      "the table (17.1, 17.3)\n",
      "and (17.3, 19.6)\n",
      "a (19.6, 19.7)\n",
      "door (19.7, 19.9)\n",
      "at (19.9, 20.6)\n",
      "the (20.6, 20.8)\n",
      "top (20.8, 20.9)\n",
      "i (20.9, 21.2)\n",
      "can (21.2, 21.5)\n",
      "see (21.5, 21.7)\n",
      "the (21.7, 21.9)\n",
      "light (21.9, 22.2)\n"
     ]
    }
   ],
   "source": [
    "for word in narrative:\n",
    "    print(word.word, word.start_end) "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6bd7ce70",
   "metadata": {},
   "source": [
    "<b>Convert (start_time, end_time) to trace coordinates that fall within the time window</b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "dc3f6d8d",
   "metadata": {},
   "outputs": [],
   "source": [
    "for word in narrative:\n",
    "    start_end = word.start_end\n",
    "    start_time = start_end[0]\n",
    "    end_time = start_end[1]\n",
    "    # Filter trace_seg for items with t value within start_time:end_time\n",
    "    word.word_traces=list(filter(lambda coord: start_time<=coord['t']<= end_time, traces))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "a7950d17",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[{'x': 0.6408, 'y': 0.1371, 't': 0.013},\n",
       " {'x': 0.6408, 'y': 0.2455, 't': 0.029},\n",
       " {'x': 0.6354, 'y': 0.3292, 't': 0.045},\n",
       " {'x': 0.6152, 'y': 0.4101, 't': 0.062},\n",
       " {'x': 0.6098, 'y': 0.4897, 't': 0.079},\n",
       " {'x': 0.6098, 'y': 0.524, 't': 0.097},\n",
       " {'x': 0.6098, 'y': 0.5542, 't': 0.114},\n",
       " {'x': 0.6079, 'y': 0.5775, 't': 0.13},\n",
       " {'x': 0.5969, 'y': 0.5912, 't': 0.147},\n",
       " {'x': 0.5878, 'y': 0.6008, 't': 0.164},\n",
       " {'x': 0.586, 'y': 0.6022, 't': 0.18},\n",
       " {'x': 0.5841, 'y': 0.6036, 't': 0.214},\n",
       " {'x': 0.5805, 'y': 0.6049, 't': 0.23},\n",
       " {'x': 0.575, 'y': 0.6077, 't': 0.247},\n",
       " {'x': 0.5476, 'y': 0.6159, 't': 0.264},\n",
       " {'x': 0.4378, 'y': 0.6324, 't': 0.28},\n",
       " {'x': 0.339, 'y': 0.6434, 't': 0.297},\n",
       " {'x': 0.2183, 'y': 0.6461, 't': 0.314},\n",
       " {'x': 0.136, 'y': 0.6461, 't': 0.33},\n",
       " {'x': 0.0939, 'y': 0.6516, 't': 0.348},\n",
       " {'x': 0.0647, 'y': 0.6543, 't': 0.364},\n",
       " {'x': 0.0445, 'y': 0.6598, 't': 0.381},\n",
       " {'x': 0.0354, 'y': 0.6694, 't': 0.397},\n",
       " {'x': 0.0354, 'y': 0.6859, 't': 0.414},\n",
       " {'x': 0.0482, 'y': 0.7119, 't': 0.43},\n",
       " {'x': 0.0592, 'y': 0.7394, 't': 0.447},\n",
       " {'x': 0.0647, 'y': 0.7668, 't': 0.464},\n",
       " {'x': 0.072, 'y': 0.8066, 't': 0.481},\n",
       " {'x': 0.0903, 'y': 0.8244, 't': 0.498},\n",
       " {'x': 0.1378, 'y': 0.8327, 't': 0.514},\n",
       " {'x': 0.1982, 'y': 0.8327, 't': 0.53},\n",
       " {'x': 0.2768, 'y': 0.8286, 't': 0.546},\n",
       " {'x': 0.3427, 'y': 0.8162, 't': 0.563},\n",
       " {'x': 0.3866, 'y': 0.8025, 't': 0.581},\n",
       " {'x': 0.4085, 'y': 0.7915, 't': 0.597},\n",
       " {'x': 0.4177, 'y': 0.7778, 't': 0.614},\n",
       " {'x': 0.4195, 'y': 0.7641, 't': 0.631},\n",
       " {'x': 0.4195, 'y': 0.7517, 't': 0.648},\n",
       " {'x': 0.4049, 'y': 0.7435, 't': 0.664},\n",
       " {'x': 0.3482, 'y': 0.738, 't': 0.681},\n",
       " {'x': 0.2732, 'y': 0.738, 't': 0.697},\n",
       " {'x': 0.2019, 'y': 0.76, 't': 0.715},\n",
       " {'x': 0.1433, 'y': 0.7833, 't': 0.731},\n",
       " {'x': 0.1031, 'y': 0.8148, 't': 0.748},\n",
       " {'x': 0.0793, 'y': 0.8519, 't': 0.763},\n",
       " {'x': 0.072, 'y': 0.9067, 't': 0.78},\n",
       " {'x': 0.0866, 'y': 0.9534, 't': 0.797}]"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "narrative[0].word_traces"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "30f94f69",
   "metadata": {},
   "source": [
    "<h3>III) Get convex hull of traces</h3>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "d92ce8a4",
   "metadata": {},
   "outputs": [],
   "source": [
    "for word in narrative:\n",
    "    points = word.word_traces\n",
    "    points = np.array([[point['x'], point['y']] for point in points])\n",
    "    hull = ConvexHull(points)\n",
    "    word.hull = hull"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b3264d1f",
   "metadata": {},
   "source": [
    "<h3>IV) Get most frequent label within hull</h3>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "d9241dfe",
   "metadata": {},
   "outputs": [],
   "source": [
    "def hull_to_name(hull):    \n",
    "    # make a polygon from the hull verticies\n",
    "    points = hull.points\n",
    "    points = np.array([[point[0]*image_width, point[1]*image_height] for point in points])\n",
    "    tupVerts = [(points[vtx,0], points[vtx,1]) for vtx in hull.vertices]\n",
    "    p = Path(tupVerts)\n",
    "    # make a canvas of coordinates corresponding to the image\n",
    "    x, y = np.meshgrid(np.arange(image_width), np.arange(image_height)) \n",
    "    x, y = x.flatten(), y.flatten()\n",
    "    points = np.vstack((x,y)).T     \n",
    "    # get points in image within hull\n",
    "    grid = p.contains_points(points)\n",
    "    mask = grid.reshape(image_height,image_width) # now you have a mask with points inside a polygon\n",
    "    # get labels at those points\n",
    "    labels_within_hull = mask*ObjectClassMasks\n",
    "    labels = labels_within_hull[np.nonzero(labels_within_hull)]\n",
    "    # get count for unique labels\n",
    "    values, counts = np.unique(labels, return_counts=True)\n",
    "    # Edge case: some utterances the traces entirely out of the image \n",
    "    # if empty set label to 0\n",
    "    if len(counts)==0:\n",
    "        return index_ade20k['objectnames'][0] \n",
    "    # get most frequent label within hull\n",
    "    ind = np.argmax(counts)\n",
    "    label = values[ind]\n",
    "    # object name lookup with label\n",
    "    return index_ade20k['objectnames'][label - 1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "bf66a8ff",
   "metadata": {},
   "outputs": [],
   "source": [
    "for word in narrative:\n",
    "    hull = word.hull\n",
    "    word.tag = hull_to_name(hull)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cb37a223",
   "metadata": {},
   "source": [
    "<h3>V) Print word trace alignment</h3>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "1ecb0e32",
   "metadata": {},
   "outputs": [],
   "source": [
    "# for printing with colors\n",
    "class bcolors:\n",
    "    HEADER = '\\033[95m'\n",
    "    OKBLUE = '\\033[94m'\n",
    "    OKCYAN = '\\033[96m'\n",
    "    OKGREEN = '\\033[92m'\n",
    "    WARNING = '\\033[93m'\n",
    "    FAIL = '\\033[91m'\n",
    "    ENDC = '\\033[0m'\n",
    "    BOLD = '\\033[1m'\n",
    "    UNDERLINE = '\\033[4m'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "7ae26de8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "in this/\u001b[92mwall\u001b[0m image/\u001b[92mbed\u001b[0m on/\u001b[92mbed\u001b[0m the/\u001b[92mbed\u001b[0m left/\u001b[92mbed\u001b[0m side/\u001b[92mbed\u001b[0m i/\u001b[92mbed\u001b[0m can/\u001b[92mbed\u001b[0m see/\u001b[92mbed\u001b[0m a/\u001b[92mbed\u001b[0m bed/\u001b[92mbed\u001b[0m and/\u001b[92mwindowpane\u001b[0m a window/\u001b[92mwindowpane\u001b[0m on/\u001b[92mwindowpane\u001b[0m right/\u001b[92mdesk\u001b[0m side/\u001b[92mwall\u001b[0m i/\u001b[92mdesk\u001b[0m can/\u001b[92mdesk\u001b[0m see/\u001b[92mdesk\u001b[0m some/\u001b[92mdesk\u001b[0m object/\u001b[92mdesk\u001b[0m on/\u001b[92mdesk\u001b[0m the table/\u001b[92mdesk\u001b[0m and/\u001b[92mwall\u001b[0m a/\u001b[92mdoor\u001b[0m door/\u001b[92mwall\u001b[0m at/\u001b[92mcurtain\u001b[0m the/\u001b[92mceiling\u001b[0m top/\u001b[92mlamp\u001b[0m i/\u001b[92mlamp\u001b[0m can/\u001b[92mlamp\u001b[0m see/\u001b[92mlamp\u001b[0m the/\u001b[92mlamp\u001b[0m light/\u001b[92mlamp\u001b[0m "
     ]
    }
   ],
   "source": [
    "for word_trace_align in narrative:\n",
    "    word = word_trace_align.word\n",
    "    tag = word_trace_align.tag\n",
    "    # one tag may have multiple words eg. \tperson;individual;someone;somebody;mortal;soul\n",
    "    # taking only first\n",
    "    tag = tag.split()[0]\n",
    "    tag = clean_text(tag)\n",
    "    print(word+'/'+bcolors.OKGREEN + tag + bcolors.ENDC, end =\" \")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "e8c500b5",
   "metadata": {},
   "source": [
    "<h1>Step 2: Compute tf-idf to get $\\alpha_{w}$ and $\\alpha_{c}$</h1>\n",
    "<p>Note: $\\alpha_{w}$ is an average of $\\alpha_{w,d}$ across all documents</p>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5b89343c",
   "metadata": {},
   "source": [
    "Now that we've seen how the baseline tagging is done for one narrative, we can do the same for the others. In the seq-tagging-baseline notebook you may see the concise version of Step One. Previously for the sake of explanation, I have done Steps 2-4 in multiple iterations of a single narrative but in the other notebook those steps are done in a single iteration. Also in the notebook I have created lists for the words and labels in each narrative that will be used for calculating tf-idf. I put the code from that notebook into a python file seq_tagging_baseline in which I have exposed a function `get_documents(num_narratives)` that will allow me to get those lists so that I can calculate the tf-idf in this main notebook. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "b545b435",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Already downloaded: ade20k_validation_localized_narratives.jsonl\n"
     ]
    }
   ],
   "source": [
    "from seq_tagging_baseline import get_documents"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "f84ca28a",
   "metadata": {},
   "outputs": [],
   "source": [
    "raw_word_documents, raw_labels_documents = get_documents(10)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3597fc28",
   "metadata": {},
   "source": [
    "Calculating tf-idf in python\n",
    "https://towardsdatascience.com/natural-language-processing-feature-engineering-using-tf-idf-e8b9d00e7e76"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "26e03a10",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "fdd0527e",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "vectorizer = TfidfVectorizer()\n",
    "vectors = vectorizer.fit_transform(raw_word_documents)\n",
    "feature_names = vectorizer.get_feature_names()\n",
    "dense = vectors.todense()\n",
    "denselist = dense.tolist()\n",
    "df = pd.DataFrame(denselist, columns=feature_names)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "ed103d8d",
   "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>also</th>\n",
       "      <th>and</th>\n",
       "      <th>are</th>\n",
       "      <th>at</th>\n",
       "      <th>back</th>\n",
       "      <th>background</th>\n",
       "      <th>bag</th>\n",
       "      <th>bathtub</th>\n",
       "      <th>bed</th>\n",
       "      <th>behind</th>\n",
       "      <th>...</th>\n",
       "      <th>two</th>\n",
       "      <th>up</th>\n",
       "      <th>wall</th>\n",
       "      <th>water</th>\n",
       "      <th>we</th>\n",
       "      <th>which</th>\n",
       "      <th>window</th>\n",
       "      <th>windows</th>\n",
       "      <th>with</th>\n",
       "      <th>yellow</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.208255</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.139291</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.199404</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.199404</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.088109</td>\n",
       "      <td>0.235726</td>\n",
       "      <td>0.117863</td>\n",
       "      <td>0.198482</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>0.147617</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.117863</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.168728</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.056457</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.051039</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.095074</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>0.070710</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.112914</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.282838</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.080822</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.098194</td>\n",
       "      <td>0.131354</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.221202</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.376084</td>\n",
       "      <td>0.221202</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.131354</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.194833</td>\n",
       "      <td>0.305222</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.123036</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.194833</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.229191</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 119 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "       also       and       are        at      back  background       bag  \\\n",
       "0  0.000000  0.208255  0.000000  0.139291  0.000000    0.000000  0.000000   \n",
       "1  0.000000  0.088109  0.235726  0.117863  0.198482    0.000000  0.000000   \n",
       "2  0.000000  0.000000  0.000000  0.056457  0.000000    0.051039  0.000000   \n",
       "3  0.000000  0.098194  0.131354  0.000000  0.000000    0.000000  0.221202   \n",
       "4  0.194833  0.305222  0.000000  0.000000  0.000000    0.123036  0.000000   \n",
       "\n",
       "    bathtub       bed    behind  ...       two   up      wall     water  \\\n",
       "0  0.000000  0.199404  0.000000  ...  0.000000  0.0  0.000000  0.000000   \n",
       "1  0.000000  0.000000  0.000000  ...  0.147617  0.0  0.117863  0.000000   \n",
       "2  0.095074  0.000000  0.000000  ...  0.070710  0.0  0.112914  0.000000   \n",
       "3  0.000000  0.376084  0.221202  ...  0.000000  0.0  0.131354  0.000000   \n",
       "4  0.000000  0.000000  0.000000  ...  0.000000  0.0  0.000000  0.194833   \n",
       "\n",
       "         we  which    window  windows      with    yellow  \n",
       "0  0.000000    0.0  0.199404      0.0  0.000000  0.000000  \n",
       "1  0.000000    0.0  0.000000      0.0  0.168728  0.000000  \n",
       "2  0.282838    0.0  0.080822      0.0  0.000000  0.000000  \n",
       "3  0.000000    0.0  0.000000      0.0  0.000000  0.000000  \n",
       "4  0.000000    0.0  0.000000      0.0  0.000000  0.229191  \n",
       "\n",
       "[5 rows x 119 columns]"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a4b26e24",
   "metadata": {},
   "source": [
    "CountVectorizer in TfidfVectorizer ignores single characters by default  \n",
    "https://stackoverflow.com/questions/43601358/empty-vocabulary-for-single-letter-by-countvectorizer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "abe28c91",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.feature_extraction.text import TfidfTransformer\n",
    "from sklearn.feature_extraction.text import CountVectorizer\n",
    "from sklearn.pipeline import Pipeline\n",
    "vectors = Pipeline([('count', CountVectorizer(token_pattern = r\"(?u)\\b\\w+\\b\")), # keep single chars\n",
    "                  ('tfid', TfidfTransformer())]).fit_transform(raw_word_documents)\n",
    "feature_names = CountVectorizer(token_pattern = r\"(?u)\\b\\w+\\b\").fit(raw_word_documents).get_feature_names()\n",
    "dense = vectors.todense()\n",
    "denselist = dense.tolist()\n",
    "df = pd.DataFrame(denselist, columns=feature_names)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "da7f985f",
   "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>a</th>\n",
       "      <th>also</th>\n",
       "      <th>and</th>\n",
       "      <th>are</th>\n",
       "      <th>at</th>\n",
       "      <th>back</th>\n",
       "      <th>background</th>\n",
       "      <th>bag</th>\n",
       "      <th>bathtub</th>\n",
       "      <th>bed</th>\n",
       "      <th>...</th>\n",
       "      <th>two</th>\n",
       "      <th>up</th>\n",
       "      <th>wall</th>\n",
       "      <th>water</th>\n",
       "      <th>we</th>\n",
       "      <th>which</th>\n",
       "      <th>window</th>\n",
       "      <th>windows</th>\n",
       "      <th>with</th>\n",
       "      <th>yellow</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.276954</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.184636</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.123494</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.176789</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.176789</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.255549</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.085183</td>\n",
       "      <td>0.227899</td>\n",
       "      <td>0.113949</td>\n",
       "      <td>0.191892</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>0.142716</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.113949</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.163126</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.283327</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.054144</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.048947</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.091178</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>0.067812</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.108287</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.271248</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.077510</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.280360</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.093453</td>\n",
       "      <td>0.125013</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.210522</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.357926</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.125013</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.093777</td>\n",
       "      <td>0.179583</td>\n",
       "      <td>0.281331</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.113405</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.179583</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.211251</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 121 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "          a      also       and       are        at      back  background  \\\n",
       "0  0.276954  0.000000  0.184636  0.000000  0.123494  0.000000    0.000000   \n",
       "1  0.255549  0.000000  0.085183  0.227899  0.113949  0.191892    0.000000   \n",
       "2  0.283327  0.000000  0.000000  0.000000  0.054144  0.000000    0.048947   \n",
       "3  0.280360  0.000000  0.093453  0.125013  0.000000  0.000000    0.000000   \n",
       "4  0.093777  0.179583  0.281331  0.000000  0.000000  0.000000    0.113405   \n",
       "\n",
       "        bag   bathtub       bed  ...       two   up      wall     water  \\\n",
       "0  0.000000  0.000000  0.176789  ...  0.000000  0.0  0.000000  0.000000   \n",
       "1  0.000000  0.000000  0.000000  ...  0.142716  0.0  0.113949  0.000000   \n",
       "2  0.000000  0.091178  0.000000  ...  0.067812  0.0  0.108287  0.000000   \n",
       "3  0.210522  0.000000  0.357926  ...  0.000000  0.0  0.125013  0.000000   \n",
       "4  0.000000  0.000000  0.000000  ...  0.000000  0.0  0.000000  0.179583   \n",
       "\n",
       "         we  which    window  windows      with    yellow  \n",
       "0  0.000000    0.0  0.176789      0.0  0.000000  0.000000  \n",
       "1  0.000000    0.0  0.000000      0.0  0.163126  0.000000  \n",
       "2  0.271248    0.0  0.077510      0.0  0.000000  0.000000  \n",
       "3  0.000000    0.0  0.000000      0.0  0.000000  0.000000  \n",
       "4  0.000000    0.0  0.000000      0.0  0.000000  0.211251  \n",
       "\n",
       "[5 rows x 121 columns]"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "d01bfcfe",
   "metadata": {},
   "outputs": [],
   "source": [
    "alpha_w_scores = df.mean(0).to_dict()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "id": "113dbb17",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.26080934248688264"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "alpha_w_scores['the']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "id": "575711d4",
   "metadata": {},
   "outputs": [],
   "source": [
    "vectors = vectorizer.fit_transform(raw_labels_documents)\n",
    "feature_names = vectorizer.get_feature_names()\n",
    "dense = vectors.todense()\n",
    "denselist = dense.tolist()\n",
    "df = pd.DataFrame(denselist, columns=feature_names)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "8710bb69",
   "metadata": {},
   "outputs": [],
   "source": [
    "alpha_c_scores = df.mean(0).to_dict()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "0850f231",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.31790345902948713"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "alpha_c_scores['wall']"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6f367eaf",
   "metadata": {},
   "source": [
    "<h1>Step 3: Use IBM Model 1 to get translation probablities P(w|c) </h1>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "945999e8",
   "metadata": {},
   "source": [
    "Why use IBMModel1 when you already have the alignments betweens the words and labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "00ebe0ab",
   "metadata": {},
   "outputs": [],
   "source": [
    "from nltk.translate import AlignedSent\n",
    "from nltk.translate import IBMModel1"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3c744015",
   "metadata": {},
   "source": [
    "https://www.nltk.org/_modules/nltk/translate/ibm1.html"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "4139e183",
   "metadata": {},
   "outputs": [],
   "source": [
    "labels_documents = [document.split() for document in raw_labels_documents]\n",
    "words_documents = [document.split() for document in raw_word_documents]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "id": "0c4bfa6e",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "in this image on the left side i can see a bed and a window on right side i can see some object on the table and a door at the top i can see the light wall wall bed bed bed bed bed bed bed bed bed bed windowpane windowpane windowpane windowpane desk wall desk desk desk desk desk desk desk desk wall door wall curtain ceiling lamp lamp lamp lamp lamp lamp "
     ]
    }
   ],
   "source": [
    "for w in words_documents[0]:\n",
    "    print(w, end=\" \")\n",
    "for l in labels_documents[0]:\n",
    "    print(l, end=\" \")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "bcaf5975",
   "metadata": {},
   "outputs": [],
   "source": [
    "bitext = []\n",
    "for i in range(len(words_documents)):\n",
    "    bitext.append(AlignedSent(words_documents[i], labels_documents[i]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "id": "57059b5c",
   "metadata": {},
   "outputs": [],
   "source": [
    "ibm1 = IBMModel1(bitext, 10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "id": "151c68a9",
   "metadata": {},
   "outputs": [],
   "source": [
    "table = ibm1.translation_table"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "id": "10e47951",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1.2423940994124731e-11"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "table[\"up\"][None]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "id": "d124f535",
   "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>seems</th>\n",
       "      <th>see</th>\n",
       "      <th>building</th>\n",
       "      <th>trees</th>\n",
       "      <th>window</th>\n",
       "      <th>behind</th>\n",
       "      <th>a</th>\n",
       "      <th>to</th>\n",
       "      <th>text</th>\n",
       "      <th>food</th>\n",
       "      <th>...</th>\n",
       "      <th>handles</th>\n",
       "      <th>the</th>\n",
       "      <th>light</th>\n",
       "      <th>at</th>\n",
       "      <th>stand</th>\n",
       "      <th>metal</th>\n",
       "      <th>floor</th>\n",
       "      <th>could</th>\n",
       "      <th>fan</th>\n",
       "      <th>glasses</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>NaN</th>\n",
       "      <td>1.242394e-11</td>\n",
       "      <td>0.070144</td>\n",
       "      <td>4.487392e-08</td>\n",
       "      <td>2.000878e-08</td>\n",
       "      <td>1.221018e-07</td>\n",
       "      <td>2.220133e-08</td>\n",
       "      <td>0.041942</td>\n",
       "      <td>2.660010e-08</td>\n",
       "      <td>1.242394e-11</td>\n",
       "      <td>1.955278e-09</td>\n",
       "      <td>...</td>\n",
       "      <td>5.920915e-11</td>\n",
       "      <td>0.175424</td>\n",
       "      <td>4.325647e-10</td>\n",
       "      <td>0.001239</td>\n",
       "      <td>9.291743e-08</td>\n",
       "      <td>2.891390e-08</td>\n",
       "      <td>0.000003</td>\n",
       "      <td>1.242394e-11</td>\n",
       "      <td>9.291743e-08</td>\n",
       "      <td>9.291743e-08</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>building</th>\n",
       "      <td>2.099560e-02</td>\n",
       "      <td>0.033383</td>\n",
       "      <td>2.099560e-02</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.026540</td>\n",
       "      <td>1.289220e-08</td>\n",
       "      <td>2.099560e-02</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.028961</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.020973</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>2.099560e-02</td>\n",
       "      <td>0.020995</td>\n",
       "      <td>2.099560e-02</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>earth</th>\n",
       "      <td>3.508096e-02</td>\n",
       "      <td>0.001757</td>\n",
       "      <td>3.508096e-02</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.008949</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>3.508096e-02</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.003147</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.035044</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>3.508096e-02</td>\n",
       "      <td>0.035081</td>\n",
       "      <td>3.508096e-02</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>wall</th>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.063702</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>1.176603e-03</td>\n",
       "      <td>2.269599e-05</td>\n",
       "      <td>0.061638</td>\n",
       "      <td>1.000000e-12</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>1.366856e-07</td>\n",
       "      <td>...</td>\n",
       "      <td>5.128172e-04</td>\n",
       "      <td>0.090005</td>\n",
       "      <td>2.142062e-12</td>\n",
       "      <td>0.020100</td>\n",
       "      <td>5.180574e-05</td>\n",
       "      <td>5.128171e-04</td>\n",
       "      <td>0.003529</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>5.180574e-05</td>\n",
       "      <td>5.180574e-05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>bed</th>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.055439</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>9.271891e-03</td>\n",
       "      <td>6.286833e-05</td>\n",
       "      <td>0.068829</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.038593</td>\n",
       "      <td>1.168165e-02</td>\n",
       "      <td>0.009297</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 121 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                 seems       see      building         trees        window  \\\n",
       "NaN       1.242394e-11  0.070144  4.487392e-08  2.000878e-08  1.221018e-07   \n",
       "building  2.099560e-02  0.033383  2.099560e-02  0.000000e+00  0.000000e+00   \n",
       "earth     3.508096e-02  0.001757  3.508096e-02  0.000000e+00  0.000000e+00   \n",
       "wall      0.000000e+00  0.063702  0.000000e+00  0.000000e+00  1.176603e-03   \n",
       "bed       0.000000e+00  0.055439  0.000000e+00  0.000000e+00  9.271891e-03   \n",
       "\n",
       "                behind         a            to          text          food  \\\n",
       "NaN       2.220133e-08  0.041942  2.660010e-08  1.242394e-11  1.955278e-09   \n",
       "building  0.000000e+00  0.026540  1.289220e-08  2.099560e-02  0.000000e+00   \n",
       "earth     0.000000e+00  0.008949  0.000000e+00  3.508096e-02  0.000000e+00   \n",
       "wall      2.269599e-05  0.061638  1.000000e-12  0.000000e+00  1.366856e-07   \n",
       "bed       6.286833e-05  0.068829  0.000000e+00  0.000000e+00  0.000000e+00   \n",
       "\n",
       "          ...       handles       the         light        at         stand  \\\n",
       "NaN       ...  5.920915e-11  0.175424  4.325647e-10  0.001239  9.291743e-08   \n",
       "building  ...  0.000000e+00  0.028961  0.000000e+00  0.020973  0.000000e+00   \n",
       "earth     ...  0.000000e+00  0.003147  0.000000e+00  0.035044  0.000000e+00   \n",
       "wall      ...  5.128172e-04  0.090005  2.142062e-12  0.020100  5.180574e-05   \n",
       "bed       ...  0.000000e+00  0.038593  1.168165e-02  0.009297  0.000000e+00   \n",
       "\n",
       "                 metal     floor         could           fan       glasses  \n",
       "NaN       2.891390e-08  0.000003  1.242394e-11  9.291743e-08  9.291743e-08  \n",
       "building  2.099560e-02  0.020995  2.099560e-02  0.000000e+00  0.000000e+00  \n",
       "earth     3.508096e-02  0.035081  3.508096e-02  0.000000e+00  0.000000e+00  \n",
       "wall      5.128171e-04  0.003529  0.000000e+00  5.180574e-05  5.180574e-05  \n",
       "bed       0.000000e+00  0.000000  0.000000e+00  0.000000e+00  0.000000e+00  \n",
       "\n",
       "[5 rows x 121 columns]"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "table_as_df = pd.DataFrame.from_dict(table).fillna(0)\n",
    "table_as_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "aa5a7fa3",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "NaN           1.0\n",
       "building      1.0\n",
       "earth         1.0\n",
       "wall          1.0\n",
       "bed           1.0\n",
       "windowpane    1.0\n",
       "desk          1.0\n",
       "door          1.0\n",
       "curtain       1.0\n",
       "ceiling       1.0\n",
       "lamp          1.0\n",
       "bathtub       1.0\n",
       "rod           1.0\n",
       "shower        1.0\n",
       "bottle        1.0\n",
       "sink          1.0\n",
       "toilet        1.0\n",
       "floor         1.0\n",
       "wardrobe      1.0\n",
       "blanket       1.0\n",
       "cushion       1.0\n",
       "painting      1.0\n",
       "sky           1.0\n",
       "sea           1.0\n",
       "tree          1.0\n",
       "mountain      1.0\n",
       "sand          1.0\n",
       "person        1.0\n",
       "railing       1.0\n",
       "other         1.0\n",
       "path          1.0\n",
       "pane          1.0\n",
       "table         1.0\n",
       "chair         1.0\n",
       "embankment    1.0\n",
       "trunk         1.0\n",
       "house         1.0\n",
       "plant         1.0\n",
       "counter       1.0\n",
       "food          1.0\n",
       "price         1.0\n",
       "dtype: float64"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# verify that sum of P(w|c) for a given c is 1 for all w\n",
    "pd.DataFrame.from_dict(table).fillna(0).sum(axis=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4af1795d",
   "metadata": {},
   "source": [
    "<b>TODO: Use Add-1 smoothing</b>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "425bd2d5",
   "metadata": {},
   "source": [
    "<b>Get emission probabilities $\\alpha_w$ * P(w|c)</b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "7017ac26",
   "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>seems</th>\n",
       "      <th>see</th>\n",
       "      <th>building</th>\n",
       "      <th>trees</th>\n",
       "      <th>window</th>\n",
       "      <th>behind</th>\n",
       "      <th>a</th>\n",
       "      <th>to</th>\n",
       "      <th>text</th>\n",
       "      <th>food</th>\n",
       "      <th>...</th>\n",
       "      <th>handles</th>\n",
       "      <th>the</th>\n",
       "      <th>light</th>\n",
       "      <th>at</th>\n",
       "      <th>stand</th>\n",
       "      <th>metal</th>\n",
       "      <th>floor</th>\n",
       "      <th>could</th>\n",
       "      <th>fan</th>\n",
       "      <th>glasses</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>NaN</th>\n",
       "      <td>1.312900e-13</td>\n",
       "      <td>0.010925</td>\n",
       "      <td>1.393521e-09</td>\n",
       "      <td>4.226872e-10</td>\n",
       "      <td>3.105039e-09</td>\n",
       "      <td>4.673875e-10</td>\n",
       "      <td>0.006679</td>\n",
       "      <td>6.214512e-10</td>\n",
       "      <td>1.312900e-13</td>\n",
       "      <td>3.752018e-11</td>\n",
       "      <td>...</td>\n",
       "      <td>5.398598e-13</td>\n",
       "      <td>0.045752</td>\n",
       "      <td>8.995823e-12</td>\n",
       "      <td>0.000063</td>\n",
       "      <td>4.896176e-09</td>\n",
       "      <td>4.838551e-10</td>\n",
       "      <td>8.066589e-08</td>\n",
       "      <td>2.625800e-13</td>\n",
       "      <td>1.632059e-09</td>\n",
       "      <td>1.632059e-09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>building</th>\n",
       "      <td>2.218710e-04</td>\n",
       "      <td>0.005200</td>\n",
       "      <td>6.520000e-04</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.004227</td>\n",
       "      <td>3.011971e-10</td>\n",
       "      <td>2.218710e-04</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.007553</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.001067</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>3.513475e-04</td>\n",
       "      <td>5.816570e-04</td>\n",
       "      <td>4.437420e-04</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>earth</th>\n",
       "      <td>3.707181e-04</td>\n",
       "      <td>0.000274</td>\n",
       "      <td>1.089409e-03</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.001425</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>3.707181e-04</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000821</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.001782</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>5.870567e-04</td>\n",
       "      <td>9.718745e-04</td>\n",
       "      <td>7.414361e-04</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>wall</th>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.009922</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>2.992090e-05</td>\n",
       "      <td>4.778012e-07</td>\n",
       "      <td>0.009816</td>\n",
       "      <td>2.336274e-14</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>2.622885e-09</td>\n",
       "      <td>...</td>\n",
       "      <td>4.675787e-06</td>\n",
       "      <td>0.023474</td>\n",
       "      <td>4.454734e-14</td>\n",
       "      <td>0.001022</td>\n",
       "      <td>2.729843e-06</td>\n",
       "      <td>8.581657e-06</td>\n",
       "      <td>9.776704e-05</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>9.099477e-07</td>\n",
       "      <td>9.099477e-07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>bed</th>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.008635</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>2.357833e-04</td>\n",
       "      <td>1.323519e-06</td>\n",
       "      <td>0.010961</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.010065</td>\n",
       "      <td>2.429372e-04</td>\n",
       "      <td>0.000473</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 121 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                 seems       see      building         trees        window  \\\n",
       "NaN       1.312900e-13  0.010925  1.393521e-09  4.226872e-10  3.105039e-09   \n",
       "building  2.218710e-04  0.005200  6.520000e-04  0.000000e+00  0.000000e+00   \n",
       "earth     3.707181e-04  0.000274  1.089409e-03  0.000000e+00  0.000000e+00   \n",
       "wall      0.000000e+00  0.009922  0.000000e+00  0.000000e+00  2.992090e-05   \n",
       "bed       0.000000e+00  0.008635  0.000000e+00  0.000000e+00  2.357833e-04   \n",
       "\n",
       "                behind         a            to          text          food  \\\n",
       "NaN       4.673875e-10  0.006679  6.214512e-10  1.312900e-13  3.752018e-11   \n",
       "building  0.000000e+00  0.004227  3.011971e-10  2.218710e-04  0.000000e+00   \n",
       "earth     0.000000e+00  0.001425  0.000000e+00  3.707181e-04  0.000000e+00   \n",
       "wall      4.778012e-07  0.009816  2.336274e-14  0.000000e+00  2.622885e-09   \n",
       "bed       1.323519e-06  0.010961  0.000000e+00  0.000000e+00  0.000000e+00   \n",
       "\n",
       "          ...       handles       the         light        at         stand  \\\n",
       "NaN       ...  5.398598e-13  0.045752  8.995823e-12  0.000063  4.896176e-09   \n",
       "building  ...  0.000000e+00  0.007553  0.000000e+00  0.001067  0.000000e+00   \n",
       "earth     ...  0.000000e+00  0.000821  0.000000e+00  0.001782  0.000000e+00   \n",
       "wall      ...  4.675787e-06  0.023474  4.454734e-14  0.001022  2.729843e-06   \n",
       "bed       ...  0.000000e+00  0.010065  2.429372e-04  0.000473  0.000000e+00   \n",
       "\n",
       "                 metal         floor         could           fan       glasses  \n",
       "NaN       4.838551e-10  8.066589e-08  2.625800e-13  1.632059e-09  1.632059e-09  \n",
       "building  3.513475e-04  5.816570e-04  4.437420e-04  0.000000e+00  0.000000e+00  \n",
       "earth     5.870567e-04  9.718745e-04  7.414361e-04  0.000000e+00  0.000000e+00  \n",
       "wall      8.581657e-06  9.776704e-05  0.000000e+00  9.099477e-07  9.099477e-07  \n",
       "bed       0.000000e+00  0.000000e+00  0.000000e+00  0.000000e+00  0.000000e+00  \n",
       "\n",
       "[5 rows x 121 columns]"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "emission_df = table_as_df.copy()\n",
    "for column in table_as_df:\n",
    "    emission_df[column] = emission_df[column] * alpha_w_scores[column] \n",
    "emission_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "id": "c8f4ec72",
   "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>seems</th>\n",
       "      <th>see</th>\n",
       "      <th>building</th>\n",
       "      <th>trees</th>\n",
       "      <th>window</th>\n",
       "      <th>behind</th>\n",
       "      <th>a</th>\n",
       "      <th>to</th>\n",
       "      <th>text</th>\n",
       "      <th>food</th>\n",
       "      <th>...</th>\n",
       "      <th>handles</th>\n",
       "      <th>the</th>\n",
       "      <th>light</th>\n",
       "      <th>at</th>\n",
       "      <th>stand</th>\n",
       "      <th>metal</th>\n",
       "      <th>floor</th>\n",
       "      <th>could</th>\n",
       "      <th>fan</th>\n",
       "      <th>glasses</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>building</th>\n",
       "      <td>0.000222</td>\n",
       "      <td>0.005200</td>\n",
       "      <td>0.000652</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.004227</td>\n",
       "      <td>3.011971e-10</td>\n",
       "      <td>0.000222</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.007553</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.001067</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000351</td>\n",
       "      <td>0.000582</td>\n",
       "      <td>0.000444</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>earth</th>\n",
       "      <td>0.000371</td>\n",
       "      <td>0.000274</td>\n",
       "      <td>0.001089</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.001425</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000371</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000821</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.001782</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000587</td>\n",
       "      <td>0.000972</td>\n",
       "      <td>0.000741</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>wall</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.009922</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000030</td>\n",
       "      <td>4.778012e-07</td>\n",
       "      <td>0.009816</td>\n",
       "      <td>2.336274e-14</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>2.622885e-09</td>\n",
       "      <td>...</td>\n",
       "      <td>4.675787e-06</td>\n",
       "      <td>0.023474</td>\n",
       "      <td>4.454734e-14</td>\n",
       "      <td>0.001022</td>\n",
       "      <td>0.000003</td>\n",
       "      <td>0.000009</td>\n",
       "      <td>0.000098</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>9.099477e-07</td>\n",
       "      <td>9.099477e-07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>bed</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.008635</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000236</td>\n",
       "      <td>1.323519e-06</td>\n",
       "      <td>0.010961</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.010065</td>\n",
       "      <td>2.429372e-04</td>\n",
       "      <td>0.000473</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>windowpane</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.009915</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.001425</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000447</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>...</td>\n",
       "      <td>8.088907e-07</td>\n",
       "      <td>0.008036</td>\n",
       "      <td>1.447270e-06</td>\n",
       "      <td>0.001903</td>\n",
       "      <td>0.000044</td>\n",
       "      <td>0.000001</td>\n",
       "      <td>0.000164</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1.478622e-05</td>\n",
       "      <td>1.478622e-05</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 121 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "               seems       see  building  trees    window        behind  \\\n",
       "building    0.000222  0.005200  0.000652    0.0  0.000000  0.000000e+00   \n",
       "earth       0.000371  0.000274  0.001089    0.0  0.000000  0.000000e+00   \n",
       "wall        0.000000  0.009922  0.000000    0.0  0.000030  4.778012e-07   \n",
       "bed         0.000000  0.008635  0.000000    0.0  0.000236  1.323519e-06   \n",
       "windowpane  0.000000  0.009915  0.000000    0.0  0.001425  0.000000e+00   \n",
       "\n",
       "                   a            to      text          food  ...       handles  \\\n",
       "building    0.004227  3.011971e-10  0.000222  0.000000e+00  ...  0.000000e+00   \n",
       "earth       0.001425  0.000000e+00  0.000371  0.000000e+00  ...  0.000000e+00   \n",
       "wall        0.009816  2.336274e-14  0.000000  2.622885e-09  ...  4.675787e-06   \n",
       "bed         0.010961  0.000000e+00  0.000000  0.000000e+00  ...  0.000000e+00   \n",
       "windowpane  0.000447  0.000000e+00  0.000000  0.000000e+00  ...  8.088907e-07   \n",
       "\n",
       "                 the         light        at     stand     metal     floor  \\\n",
       "building    0.007553  0.000000e+00  0.001067  0.000000  0.000351  0.000582   \n",
       "earth       0.000821  0.000000e+00  0.001782  0.000000  0.000587  0.000972   \n",
       "wall        0.023474  4.454734e-14  0.001022  0.000003  0.000009  0.000098   \n",
       "bed         0.010065  2.429372e-04  0.000473  0.000000  0.000000  0.000000   \n",
       "windowpane  0.008036  1.447270e-06  0.001903  0.000044  0.000001  0.000164   \n",
       "\n",
       "               could           fan       glasses  \n",
       "building    0.000444  0.000000e+00  0.000000e+00  \n",
       "earth       0.000741  0.000000e+00  0.000000e+00  \n",
       "wall        0.000000  9.099477e-07  9.099477e-07  \n",
       "bed         0.000000  0.000000e+00  0.000000e+00  \n",
       "windowpane  0.000000  1.478622e-05  1.478622e-05  \n",
       "\n",
       "[5 rows x 121 columns]"
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# NaN shouldn't be applicable here so I shall remove it first.\n",
    "emission_df_nonan = emission_df[emission_df.index.notnull()]\n",
    "emission_df_nonan.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "id": "ab70268d",
   "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>a</th>\n",
       "      <th>also</th>\n",
       "      <th>and</th>\n",
       "      <th>are</th>\n",
       "      <th>at</th>\n",
       "      <th>back</th>\n",
       "      <th>background</th>\n",
       "      <th>bag</th>\n",
       "      <th>bathtub</th>\n",
       "      <th>bed</th>\n",
       "      <th>...</th>\n",
       "      <th>two</th>\n",
       "      <th>up</th>\n",
       "      <th>wall</th>\n",
       "      <th>water</th>\n",
       "      <th>we</th>\n",
       "      <th>which</th>\n",
       "      <th>window</th>\n",
       "      <th>windows</th>\n",
       "      <th>with</th>\n",
       "      <th>yellow</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>bathtub</th>\n",
       "      <td>0.000064</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000660</td>\n",
       "      <td>0.0</td>\n",
       "      <td>4.186138e-05</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000488</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>...</td>\n",
       "      <td>1.235988e-03</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>8.081592e-07</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.003096</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000746</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>bed</th>\n",
       "      <td>0.010961</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.010578</td>\n",
       "      <td>0.000007</td>\n",
       "      <td>0.000473</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000001</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.00698</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>3.158388e-08</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000236</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>blanket</th>\n",
       "      <td>0.000164</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000293</td>\n",
       "      <td>0.009839</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.001996</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.00142</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>4.763710e-05</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>bottle</th>\n",
       "      <td>0.000064</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000660</td>\n",
       "      <td>0.0</td>\n",
       "      <td>4.186138e-05</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000488</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>...</td>\n",
       "      <td>1.235988e-03</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>8.081592e-07</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.003096</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000746</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>building</th>\n",
       "      <td>0.004227</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.003560</td>\n",
       "      <td>0.002955</td>\n",
       "      <td>0.001067</td>\n",
       "      <td>0.0</td>\n",
       "      <td>5.018970e-09</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.00000</td>\n",
       "      <td>...</td>\n",
       "      <td>1.815327e-11</td>\n",
       "      <td>0.000222</td>\n",
       "      <td>1.079453e-03</td>\n",
       "      <td>3.343638e-14</td>\n",
       "      <td>0.001855</td>\n",
       "      <td>0.000222</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000222</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 121 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                 a  also       and       are        at  back    background  \\\n",
       "bathtub   0.000064   0.0  0.000000  0.000000  0.000660   0.0  4.186138e-05   \n",
       "bed       0.010961   0.0  0.010578  0.000007  0.000473   0.0  0.000000e+00   \n",
       "blanket   0.000164   0.0  0.000293  0.009839  0.000000   0.0  0.000000e+00   \n",
       "bottle    0.000064   0.0  0.000000  0.000000  0.000660   0.0  4.186138e-05   \n",
       "building  0.004227   0.0  0.003560  0.002955  0.001067   0.0  5.018970e-09   \n",
       "\n",
       "               bag   bathtub      bed  ...           two        up  \\\n",
       "bathtub   0.000000  0.000488  0.00000  ...  1.235988e-03  0.000000   \n",
       "bed       0.000001  0.000000  0.00698  ...  0.000000e+00  0.000000   \n",
       "blanket   0.001996  0.000000  0.00142  ...  0.000000e+00  0.000000   \n",
       "bottle    0.000000  0.000488  0.00000  ...  1.235988e-03  0.000000   \n",
       "building  0.000000  0.000000  0.00000  ...  1.815327e-11  0.000222   \n",
       "\n",
       "                  wall         water        we     which    window   windows  \\\n",
       "bathtub   8.081592e-07  0.000000e+00  0.003096  0.000000  0.000746  0.000000   \n",
       "bed       3.158388e-08  0.000000e+00  0.000000  0.000000  0.000236  0.000000   \n",
       "blanket   4.763710e-05  0.000000e+00  0.000000  0.000000  0.000000  0.000000   \n",
       "bottle    8.081592e-07  0.000000e+00  0.003096  0.000000  0.000746  0.000000   \n",
       "building  1.079453e-03  3.343638e-14  0.001855  0.000222  0.000000  0.000222   \n",
       "\n",
       "          with  yellow  \n",
       "bathtub    0.0     0.0  \n",
       "bed        0.0     0.0  \n",
       "blanket    0.0     0.0  \n",
       "bottle     0.0     0.0  \n",
       "building   0.0     0.0  \n",
       "\n",
       "[5 rows x 121 columns]"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# alphabetical order\n",
    "sorted_emission_df = emission_df_nonan.sort_index()\n",
    "sorted_emission_df = sorted_emission_df.sort_index(axis=1)\n",
    "sorted_emission_df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cac96c9a",
   "metadata": {},
   "source": [
    "<b>Lookups</b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "id": "6a71fcc6",
   "metadata": {},
   "outputs": [],
   "source": [
    "id2label = sorted_emission_df.index\n",
    "label2id = {k: v for v, k in enumerate(id2label)}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "id": "ded7b08f",
   "metadata": {},
   "outputs": [],
   "source": [
    "id2word=sorted_emission_df.columns\n",
    "word2id = {k: v for v, k in enumerate(id2word)}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "id": "fda2bfb5",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "bathtub       1.0\n",
       "bed           1.0\n",
       "blanket       1.0\n",
       "bottle        1.0\n",
       "building      1.0\n",
       "ceiling       1.0\n",
       "chair         1.0\n",
       "counter       1.0\n",
       "curtain       1.0\n",
       "cushion       1.0\n",
       "desk          1.0\n",
       "door          1.0\n",
       "earth         1.0\n",
       "embankment    1.0\n",
       "floor         1.0\n",
       "food          1.0\n",
       "house         1.0\n",
       "lamp          1.0\n",
       "mountain      1.0\n",
       "other         1.0\n",
       "painting      1.0\n",
       "pane          1.0\n",
       "path          1.0\n",
       "person        1.0\n",
       "plant         1.0\n",
       "price         1.0\n",
       "railing       1.0\n",
       "rod           1.0\n",
       "sand          1.0\n",
       "sea           1.0\n",
       "shower        1.0\n",
       "sink          1.0\n",
       "sky           1.0\n",
       "table         1.0\n",
       "toilet        1.0\n",
       "tree          1.0\n",
       "trunk         1.0\n",
       "wall          1.0\n",
       "wardrobe      1.0\n",
       "windowpane    1.0\n",
       "dtype: float64"
      ]
     },
     "execution_count": 54,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# turn back to probabilities: sum to 1\n",
    "norm_emission_df = sorted_emission_df.div(sorted_emission_df.sum(axis=1), axis=0)\n",
    "norm_emission_df.sum(axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "id": "c066a133",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[2.85699656e-03, 0.00000000e+00, 0.00000000e+00, ...,\n",
       "        0.00000000e+00, 0.00000000e+00, 0.00000000e+00],\n",
       "       [1.02703455e-01, 0.00000000e+00, 9.91093591e-02, ...,\n",
       "        0.00000000e+00, 0.00000000e+00, 0.00000000e+00],\n",
       "       [4.13538865e-03, 0.00000000e+00, 7.39775780e-03, ...,\n",
       "        0.00000000e+00, 0.00000000e+00, 0.00000000e+00],\n",
       "       ...,\n",
       "       [7.57622015e-02, 1.31493856e-05, 5.66383815e-02, ...,\n",
       "        0.00000000e+00, 1.84312750e-03, 0.00000000e+00],\n",
       "       [4.13538865e-03, 0.00000000e+00, 7.39775780e-03, ...,\n",
       "        0.00000000e+00, 0.00000000e+00, 0.00000000e+00],\n",
       "       [5.59323724e-03, 3.46105317e-04, 2.74579499e-02, ...,\n",
       "        0.00000000e+00, 5.76792269e-04, 0.00000000e+00]])"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "emission_matrix = norm_emission_df.to_numpy()\n",
    "emission_matrix"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6a21f656",
   "metadata": {},
   "source": [
    "<h1>Step 4: Find Count(c'->c)</h1>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "id": "38372502",
   "metadata": {},
   "outputs": [],
   "source": [
    "labels_docs = [[\"<start>\"] + document for document in labels_documents]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "id": "26dee755",
   "metadata": {},
   "outputs": [],
   "source": [
    "df = \"\"\n",
    "for i in range(len(labels_docs)):\n",
    "    doc = labels_docs[i]\n",
    "    if i==0: # for first iteration\n",
    "        df = pd.crosstab(pd.Series(doc[:-1], name='from'),\n",
    "            pd.Series(doc[1:], name='to')) \n",
    "    else:\n",
    "        df2 = pd.crosstab(pd.Series(doc[:-1], name='from'),\n",
    "                pd.Series(doc[1:], name='to'))\n",
    "        df = df.add(df2, fill_value=0).fillna(0)  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "id": "898d3935",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['bathtub', 'bed', 'blanket', 'bottle', 'building', 'ceiling', 'chair',\n",
       "       'counter', 'curtain', 'cushion', 'desk', 'door', 'earth', 'embankment',\n",
       "       'floor', 'food', 'house', 'lamp', 'mountain', 'other', 'painting',\n",
       "       'pane', 'path', 'person', 'plant', 'price', 'railing', 'rod', 'sand',\n",
       "       'sea', 'shower', 'sink', 'sky', 'table', 'toilet', 'tree', 'trunk',\n",
       "       'wall', 'wardrobe', 'windowpane'],\n",
       "      dtype='object', name='to')"
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.keys()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "85c25861",
   "metadata": {},
   "source": [
    "<b>Use Add-1 smoothing</b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "id": "a74a5e33",
   "metadata": {},
   "outputs": [],
   "source": [
    "smooth = df+1"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "187ff94b",
   "metadata": {},
   "source": [
    "<b>Get transition probabilities $\\alpha_{c'}$ * Count(c'->c)</b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "id": "f84a00ed",
   "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>to</th>\n",
       "      <th>bathtub</th>\n",
       "      <th>bed</th>\n",
       "      <th>blanket</th>\n",
       "      <th>bottle</th>\n",
       "      <th>building</th>\n",
       "      <th>ceiling</th>\n",
       "      <th>chair</th>\n",
       "      <th>counter</th>\n",
       "      <th>curtain</th>\n",
       "      <th>cushion</th>\n",
       "      <th>...</th>\n",
       "      <th>shower</th>\n",
       "      <th>sink</th>\n",
       "      <th>sky</th>\n",
       "      <th>table</th>\n",
       "      <th>toilet</th>\n",
       "      <th>tree</th>\n",
       "      <th>trunk</th>\n",
       "      <th>wall</th>\n",
       "      <th>wardrobe</th>\n",
       "      <th>windowpane</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>from</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>&lt;start&gt;</th>\n",
       "      <td>0.037052</td>\n",
       "      <td>0.090559</td>\n",
       "      <td>0.017696</td>\n",
       "      <td>0.003705</td>\n",
       "      <td>0.358684</td>\n",
       "      <td>0.036292</td>\n",
       "      <td>0.011636</td>\n",
       "      <td>0.080934</td>\n",
       "      <td>0.007114</td>\n",
       "      <td>0.026544</td>\n",
       "      <td>...</td>\n",
       "      <td>0.00741</td>\n",
       "      <td>0.018526</td>\n",
       "      <td>0.173710</td>\n",
       "      <td>0.104721</td>\n",
       "      <td>0.018526</td>\n",
       "      <td>0.091727</td>\n",
       "      <td>0.017697</td>\n",
       "      <td>1.589517</td>\n",
       "      <td>0.070783</td>\n",
       "      <td>0.054786</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>bathtub</th>\n",
       "      <td>0.055577</td>\n",
       "      <td>0.090559</td>\n",
       "      <td>0.017696</td>\n",
       "      <td>0.003705</td>\n",
       "      <td>0.179342</td>\n",
       "      <td>0.036292</td>\n",
       "      <td>0.011636</td>\n",
       "      <td>0.080934</td>\n",
       "      <td>0.007114</td>\n",
       "      <td>0.026544</td>\n",
       "      <td>...</td>\n",
       "      <td>0.00741</td>\n",
       "      <td>0.018526</td>\n",
       "      <td>0.086855</td>\n",
       "      <td>0.052361</td>\n",
       "      <td>0.018526</td>\n",
       "      <td>0.091727</td>\n",
       "      <td>0.017697</td>\n",
       "      <td>1.271614</td>\n",
       "      <td>0.035392</td>\n",
       "      <td>0.054786</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>bed</th>\n",
       "      <td>0.018526</td>\n",
       "      <td>1.086705</td>\n",
       "      <td>0.035392</td>\n",
       "      <td>0.003705</td>\n",
       "      <td>0.179342</td>\n",
       "      <td>0.036292</td>\n",
       "      <td>0.011636</td>\n",
       "      <td>0.080934</td>\n",
       "      <td>0.007114</td>\n",
       "      <td>0.053087</td>\n",
       "      <td>...</td>\n",
       "      <td>0.00741</td>\n",
       "      <td>0.018526</td>\n",
       "      <td>0.086855</td>\n",
       "      <td>0.052361</td>\n",
       "      <td>0.018526</td>\n",
       "      <td>0.091727</td>\n",
       "      <td>0.017697</td>\n",
       "      <td>0.317903</td>\n",
       "      <td>0.035392</td>\n",
       "      <td>0.109573</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>blanket</th>\n",
       "      <td>0.018526</td>\n",
       "      <td>0.181117</td>\n",
       "      <td>0.017696</td>\n",
       "      <td>0.003705</td>\n",
       "      <td>0.179342</td>\n",
       "      <td>0.036292</td>\n",
       "      <td>0.011636</td>\n",
       "      <td>0.080934</td>\n",
       "      <td>0.007114</td>\n",
       "      <td>0.026544</td>\n",
       "      <td>...</td>\n",
       "      <td>0.00741</td>\n",
       "      <td>0.018526</td>\n",
       "      <td>0.086855</td>\n",
       "      <td>0.052361</td>\n",
       "      <td>0.018526</td>\n",
       "      <td>0.091727</td>\n",
       "      <td>0.017697</td>\n",
       "      <td>0.635807</td>\n",
       "      <td>0.035392</td>\n",
       "      <td>0.054786</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>bottle</th>\n",
       "      <td>0.018526</td>\n",
       "      <td>0.090559</td>\n",
       "      <td>0.017696</td>\n",
       "      <td>0.003705</td>\n",
       "      <td>0.179342</td>\n",
       "      <td>0.036292</td>\n",
       "      <td>0.011636</td>\n",
       "      <td>0.080934</td>\n",
       "      <td>0.007114</td>\n",
       "      <td>0.026544</td>\n",
       "      <td>...</td>\n",
       "      <td>0.00741</td>\n",
       "      <td>0.018526</td>\n",
       "      <td>0.086855</td>\n",
       "      <td>0.052361</td>\n",
       "      <td>0.018526</td>\n",
       "      <td>0.091727</td>\n",
       "      <td>0.017697</td>\n",
       "      <td>0.635807</td>\n",
       "      <td>0.035392</td>\n",
       "      <td>0.054786</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 40 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "to        bathtub       bed   blanket    bottle  building   ceiling     chair  \\\n",
       "from                                                                            \n",
       "<start>  0.037052  0.090559  0.017696  0.003705  0.358684  0.036292  0.011636   \n",
       "bathtub  0.055577  0.090559  0.017696  0.003705  0.179342  0.036292  0.011636   \n",
       "bed      0.018526  1.086705  0.035392  0.003705  0.179342  0.036292  0.011636   \n",
       "blanket  0.018526  0.181117  0.017696  0.003705  0.179342  0.036292  0.011636   \n",
       "bottle   0.018526  0.090559  0.017696  0.003705  0.179342  0.036292  0.011636   \n",
       "\n",
       "to        counter   curtain   cushion  ...   shower      sink       sky  \\\n",
       "from                                   ...                                \n",
       "<start>  0.080934  0.007114  0.026544  ...  0.00741  0.018526  0.173710   \n",
       "bathtub  0.080934  0.007114  0.026544  ...  0.00741  0.018526  0.086855   \n",
       "bed      0.080934  0.007114  0.053087  ...  0.00741  0.018526  0.086855   \n",
       "blanket  0.080934  0.007114  0.026544  ...  0.00741  0.018526  0.086855   \n",
       "bottle   0.080934  0.007114  0.026544  ...  0.00741  0.018526  0.086855   \n",
       "\n",
       "to          table    toilet      tree     trunk      wall  wardrobe  \\\n",
       "from                                                                  \n",
       "<start>  0.104721  0.018526  0.091727  0.017697  1.589517  0.070783   \n",
       "bathtub  0.052361  0.018526  0.091727  0.017697  1.271614  0.035392   \n",
       "bed      0.052361  0.018526  0.091727  0.017697  0.317903  0.035392   \n",
       "blanket  0.052361  0.018526  0.091727  0.017697  0.635807  0.035392   \n",
       "bottle   0.052361  0.018526  0.091727  0.017697  0.635807  0.035392   \n",
       "\n",
       "to       windowpane  \n",
       "from                 \n",
       "<start>    0.054786  \n",
       "bathtub    0.054786  \n",
       "bed        0.109573  \n",
       "blanket    0.054786  \n",
       "bottle     0.054786  \n",
       "\n",
       "[5 rows x 40 columns]"
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "transition_df = smooth.copy()\n",
    "for column in transition_df:\n",
    "    transition_df[column] = transition_df[column] * alpha_c_scores[column] \n",
    "transition_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "id": "ffb8ee74",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "from\n",
       "<start>       1.0\n",
       "bathtub       1.0\n",
       "bed           1.0\n",
       "blanket       1.0\n",
       "bottle        1.0\n",
       "building      1.0\n",
       "ceiling       1.0\n",
       "chair         1.0\n",
       "counter       1.0\n",
       "curtain       1.0\n",
       "cushion       1.0\n",
       "desk          1.0\n",
       "door          1.0\n",
       "earth         1.0\n",
       "embankment    1.0\n",
       "floor         1.0\n",
       "food          1.0\n",
       "house         1.0\n",
       "lamp          1.0\n",
       "mountain      1.0\n",
       "other         1.0\n",
       "painting      1.0\n",
       "pane          1.0\n",
       "path          1.0\n",
       "person        1.0\n",
       "plant         1.0\n",
       "price         1.0\n",
       "railing       1.0\n",
       "rod           1.0\n",
       "sand          1.0\n",
       "sea           1.0\n",
       "shower        1.0\n",
       "sink          1.0\n",
       "sky           1.0\n",
       "table         1.0\n",
       "toilet        1.0\n",
       "tree          1.0\n",
       "trunk         1.0\n",
       "wall          1.0\n",
       "wardrobe      1.0\n",
       "windowpane    1.0\n",
       "dtype: float64"
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# normalize\n",
    "norm_transition_df = transition_df.div(transition_df.sum(axis=1), axis=0)\n",
    "norm_transition_df.sum(axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "id": "671d31ba",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[9.93091346e-03, 2.42724096e-02, 4.74298576e-03, ...,\n",
       "        4.26037559e-01, 1.89719431e-02, 1.46843910e-02],\n",
       "       [1.84115299e-02, 3.00000732e-02, 5.86220826e-03, ...,\n",
       "        4.21257161e-01, 1.17244165e-02, 1.81495291e-02],\n",
       "       [5.93199699e-03, 3.47965237e-01, 1.13324428e-02, ...,\n",
       "        1.01793396e-01, 1.13324428e-02, 3.50854987e-02],\n",
       "       ...,\n",
       "       [2.29635373e-03, 7.48343895e-03, 7.31156176e-04, ...,\n",
       "        8.66922252e-01, 1.46231235e-03, 1.35820561e-02],\n",
       "       [8.32773709e-03, 8.14161745e-02, 7.95462344e-03, ...,\n",
       "        1.42904429e-01, 6.36369875e-02, 2.46276937e-02],\n",
       "       [4.59295997e-03, 2.24515511e-02, 4.38717825e-03, ...,\n",
       "        4.72892681e-01, 8.77435649e-03, 8.14968171e-02]])"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "transition_matrix = norm_transition_df.to_numpy()\n",
    "transition_matrix"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a396e63a",
   "metadata": {},
   "source": [
    "<h1>Step 5: Construct Hidden Markov Model</h1>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "a823fa2a",
   "metadata": {},
   "source": [
    "https://costalab.org/wp-content/uploads/2017/01/hmm.txt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "id": "d8d2a8a7",
   "metadata": {},
   "outputs": [],
   "source": [
    "Pi = transition_matrix[0,:]# starting probabilities\n",
    "A = transition_matrix[1:,:]# transmission matrix\n",
    "B = emission_matrix # emission matrix    "
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0d2a0a04",
   "metadata": {},
   "source": [
    "https://stackoverflow.com/questions/9729968/python-implementation-of-viterbi-algorithm/9730083"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "id": "46798c0c",
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "\n",
    "def viterbi(y, A, B, Pi=None, constraint=np.ones(A.shape[0])):\n",
    "    \"\"\"\n",
    "    Parameters\n",
    "    ----------\n",
    "    y : array (T,)\n",
    "        Observation state sequence. int dtype.\n",
    "    A : array (K, K)\n",
    "        State transition matrix. See HiddenMarkovModel.state_transition  for\n",
    "        details.\n",
    "    B : array (K, M)\n",
    "        Emission matrix. See HiddenMarkovModel.emission for details.\n",
    "    Pi: optional, (K,)\n",
    "        Initial state probabilities: Pi[i] is the probability x[0] == i. If\n",
    "        None, uniform initial distribution is assumed (Pi[:] == 1/K).\n",
    "\n",
    "    Returns\n",
    "    -------\n",
    "    x : array (T,)\n",
    "        Maximum a posteriori probability estimate of hidden state trajectory,\n",
    "        conditioned on observation sequence y under the model parameters A, B,\n",
    "        Pi.\n",
    "    T1: array (K, T)\n",
    "        the probability of the most likely path so far\n",
    "    T2: array (K, T)\n",
    "        the x_j-1 of the most likely path so far\n",
    "    \"\"\"\n",
    "    # Cardinality of the state space\n",
    "    K = A.shape[0]\n",
    "    # Initialize the priors with default (uniform dist) if not given by caller\n",
    "    Pi = Pi if Pi is not None else np.full(K, 1 / K)\n",
    "    T = len(y)\n",
    "    T1 = np.empty((K, T), 'd')\n",
    "    T2 = np.empty((K, T), 'B')\n",
    "\n",
    "    # Initilaize the tracking tables from first observation\n",
    "    T1[:, 0] = Pi * B[:, y[0]]\n",
    "    T2[:, 0] = 0\n",
    "\n",
    "    # Iterate throught the observations updating the tracking tables\n",
    "    for i in range(1, T):\n",
    "        T1[:, i] = np.max(T1[:, i - 1] * constraint * A.T * B[np.newaxis, :, y[i]].T, 1)\n",
    "        T2[:, i] = np.argmax(T1[:, i - 1] * constraint * A.T, 1)\n",
    "\n",
    "    # Build the output, optimal model trajectory\n",
    "    x = np.empty(T, 'B')\n",
    "    x[-1] = np.argmax(T1[:, T - 1])\n",
    "    for i in reversed(range(1, T)):\n",
    "        x[i - 1] = T2[x[i], i]\n",
    "\n",
    "    return x, T1, T2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "id": "50c81d37",
   "metadata": {},
   "outputs": [],
   "source": [
    "test_sentence = \"The flower is on the table\"\n",
    "test_sentence = clean_text(test_sentence)\n",
    "test_sentence_arr = test_sentence.split()\n",
    "y = [word2id[word] for word in test_sentence_arr]\n",
    "x, _, _ = viterbi(y, A, B, Pi)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "id": "8d600557",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "the/\u001b[92mwall\u001b[0m flower/\u001b[92mtable\u001b[0m is/\u001b[92mwall\u001b[0m on/\u001b[92mwall\u001b[0m the/\u001b[92mwall\u001b[0m table/\u001b[92mwall\u001b[0m "
     ]
    }
   ],
   "source": [
    "for i in range(len(test_sentence_arr)):\n",
    "    word = test_sentence_arr[i]\n",
    "    label = id2label[x[i]]\n",
    "    print(word+'/'+bcolors.OKGREEN + label + bcolors.ENDC, end =\" \")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fe203afe",
   "metadata": {},
   "source": [
    "<h1>Step 6: Constrained Label Refinement</h1>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f7245229",
   "metadata": {},
   "source": [
    "Recall"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "id": "08916162",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(image_seg)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ce4e0f57",
   "metadata": {},
   "source": [
    "<b>Get all labels in image</b>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "id": "77e47009",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[  0,   0, 447, ..., 447,   0,   0],\n",
       "       [  0,   0, 447, ..., 447,   0,   0],\n",
       "       [  0,   0, 447, ..., 447,   0,   0],\n",
       "       ...,\n",
       "       [  0,   0, 165, ..., 491, 491,   0],\n",
       "       [  0,   0, 165, ..., 491, 491,   0],\n",
       "       [  0,   0,   0, ...,   0,   0,   0]], dtype=int32)"
      ]
     },
     "execution_count": 68,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ObjectClassMasks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "id": "be88ab38",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([   0,  165,  447,  491,  687,  724,  774,  976, 1186, 1223, 1395,\n",
       "       1407, 2679, 2978, 3055], dtype=int32)"
      ]
     },
     "execution_count": 69,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "unique_ids = np.unique(ObjectClassMasks)\n",
    "unique_ids"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "id": "02b3f168",
   "metadata": {},
   "outputs": [],
   "source": [
    "unique_labels = []\n",
    "for object_id in unique_ids:\n",
    "    if(object_id==0):\n",
    "        unique_labels.append(\"other\")\n",
    "        continue\n",
    "    label=index_ade20k['objectnames'][object_id-1]\n",
    "    unique_labels.append(label)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "id": "c0437e1d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['other',\n",
       " 'bed',\n",
       " 'ceiling',\n",
       " 'chest of drawers, chest, bureau, dresser',\n",
       " 'curtain, drape, drapery, mantle, pall',\n",
       " 'desk',\n",
       " 'door',\n",
       " 'floor, flooring',\n",
       " 'hanger',\n",
       " 'heater, warmer',\n",
       " 'lamp',\n",
       " 'laptop, laptop computer',\n",
       " 'swivel chair',\n",
       " 'wall',\n",
       " 'windowpane, window']"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "unique_labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "id": "5ce78f24",
   "metadata": {},
   "outputs": [
    {
     "ename": "KeyError",
     "evalue": "'chest'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyError\u001b[0m                                  Traceback (most recent call last)",
      "\u001b[0;32m/var/folders/z9/wjm53w_n3jl4wfzcyxrn54dm0000gn/T/ipykernel_37631/3213778828.py\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0munique_label_ids\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mlabel2id\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mclean_text\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlabel\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msplit\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mlabel\u001b[0m \u001b[0;32min\u001b[0m \u001b[0munique_labels\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/var/folders/z9/wjm53w_n3jl4wfzcyxrn54dm0000gn/T/ipykernel_37631/3213778828.py\u001b[0m in \u001b[0;36m<listcomp>\u001b[0;34m(.0)\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0munique_label_ids\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mlabel2id\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mclean_text\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlabel\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msplit\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mlabel\u001b[0m \u001b[0;32min\u001b[0m \u001b[0munique_labels\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;31mKeyError\u001b[0m: 'chest'"
     ]
    }
   ],
   "source": [
    "unique_label_ids = [label2id[clean_text(label).split()[0]] for label in unique_labels]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5bc67173",
   "metadata": {},
   "source": [
    "<b>Problem:</b> If a label was never used to tag the sentence in the beginning, it wouldn't be added to the dictionary of labels. I just realised that the \"bag of image labels\" in \"pairing each narrative and the corresponding bag of image labels\" may refer to all image labels annotated in the image."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 73,
   "id": "80f02c7a",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['hello;world', 'world']"
      ]
     },
     "execution_count": 73,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "\"hello;world world\".split()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1314280c",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}