Greendeck ML Assignment Submission Tarun Makkar.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "``PROBLEM STATEMENT:``\n",
    "========\n",
    "* Implementing ``HUSE`` (Hierarchical Universal Semantic Embeddings)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## ``Technology used:``\n",
    "* Pytorch"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "``PART 1:``\n",
    "===================================================\n",
    "Image Embeddings Input Model (10 points)\n",
    "---------------------------------------------\n",
    "\n",
    "You are free to use any pretrained image model for representing raw images into embedding space. Image embedding created from this part of the model are forwarded into the image tower for further processing. Here you will be tested on your skills to use transfer learning and image augmentation.\n",
    "\n",
    "Text  Embeddings Input Model (10 points)\n",
    "-------------------------------------------\n",
    "\n",
    "You are free to use any pretrained text model for representing text into embedding space. Text embedding created from this part of the model are forwarded into the text tower for further processing. Here you will be tested on your skills to use transfer learning and text preprocessing.\n",
    "\n",
    "\n",
    "``PART 2:``\n",
    "==============================================================\n",
    "Image Tower Model (10 points)\n",
    "-------------------------------------------\n",
    "It consists of sequence of dense and activation layers, for exact model description refer section 4 of the paper under the name Experiments . The final L2-normalized output is sent to a shared fully connected layer. You are required to build this part as it is, taking its input from image pretrained model of PART1.\n",
    "\n",
    "\n",
    "Text Tower Model (10 points)\n",
    "----------------------------------------\n",
    "It consists of sequence of dense and activation layers, for exact model description refer section 4 of the paper under the name Experiments . The final L2-normalized output is sent to a shared fully connected layer. You are required to build this part as it is, taking its input from text pretrained model of PART1.\n",
    "\n",
    "``Part 3``\n",
    "================\n",
    "Class Level Similarity (10 points)\n",
    "-------------------------------\n",
    "HUSE passes the embeddings from image tower and text tower through a shared fully connected layer and the model is trained using softmax cross entropy loss.\n",
    "\n",
    "Semantic Similarity (30 points)\n",
    "---------------------------\n",
    "\n",
    "You are required to create a semantic graph, which consists of Adjacency matrix which is nothing but matrix of cosine distance between the textl embedding representation of individual classes. Such that if you have 5 unique classes, your adjacency matrix will be a 5x5 matrix consisting of cosine distance of text embedding of one class name with all other classes names. This matrix is used in calculating Semantic Similarity loss.\n",
    "\n",
    "Cross Modal Gap (20 points)\n",
    "----------------------------------------\n",
    "\n",
    "You required to create a cross modal gap loss, which is the distance between image and text embeddings corresponding to the same product instance."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Importing ``Libraries``"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from __future__ import print_function, division\n",
    "\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.optim as optim\n",
    "from torch.optim import lr_scheduler\n",
    "import numpy as np\n",
    "import torchvision\n",
    "from torchvision import datasets, models, transforms\n",
    "import matplotlib.pyplot as plt\n",
    "import time\n",
    "import os\n",
    "import copy\n",
    "\n",
    "plt.ion()   # interactive mode\n",
    "\n",
    "import pandas as pd\n",
    "from skimage import io, transform\n",
    "from torch.utils.data import Dataset, DataLoader\n",
    "\n",
    "# For image embeddings\n",
    "from img2vec_pytorch import Img2Vec\n",
    "from PIL import Image\n",
    "\n",
    "import tqdm\n",
    "import re\n",
    "from scipy.spatial.distance import cosine\n",
    "from itertools import product\n",
    "\n",
    "import numpy as np\n",
    "import pandas as pd"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### ``Data Overview``\n",
    "* Creating a class ``ImageClassesDataset`` to load dataset\n",
    "\n",
    "### ``References:``\n",
    "* https://pytorch.org/tutorials/beginner/data_loading_tutorial.html"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Image name: 5da81f696504fb65d41c583d_0.jpg\n",
      "Description: Charlotte Tilbury  Hollywood Lips Matte Contour Liquid Lipstick – Show Girl  \n",
      "Classes: beauty<makeup<lipstick\n"
     ]
    }
   ],
   "source": [
    "# importing data\n",
    "training = pd.read_csv('training.csv')\n",
    "\n",
    "# Showing 65th row image_name , description, classes in data\n",
    "n = 65\n",
    "img_name = training.iloc[n, 0]\n",
    "name = training.iloc[n, 1]\n",
    "classes = training.iloc[n, 2]\n",
    "\n",
    "print('Image name: {}'.format(img_name))\n",
    "print('Description: {}'.format(name))\n",
    "print('Classes: {}'.format(classes))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "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"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Description: Charlotte Tilbury  Hollywood Lips Matte Contour Liquid Lipstick – Show Girl  \n",
      "Classes: beauty<makeup<lipstick\n"
     ]
    }
   ],
   "source": [
    "# Showing image of 65th data point\n",
    "\n",
    "def show_classes(image,name,classes):\n",
    "    \"\"\"Show image with landmarks\"\"\"\n",
    "    plt.imshow(image)\n",
    "    plt.show()\n",
    "    print('Description: {}'.format(name))\n",
    "    print('Classes: {}'.format(classes))\n",
    "\n",
    "plt.figure()\n",
    "\n",
    "show_classes(io.imread(os.path.join('images/', img_name)),name,\n",
    "               classes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ImageClassesDataset(Dataset):\n",
    "    \"\"\"Imageclasses dataset.\"\"\"\n",
    "\n",
    "    def __init__(self, csv_file, root_dir, transform=None):\n",
    "        \"\"\"\n",
    "        Args:\n",
    "            csv_file (string): Path to the csv file with annotations.\n",
    "            root_dir (string): Directory with all the images.\n",
    "            transform (callable, optional): Optional transform to be applied\n",
    "                on a sample.\n",
    "        \"\"\"\n",
    "        self.training = pd.read_csv(csv_file)\n",
    "        self.root_dir = root_dir\n",
    "        self.transform = transform\n",
    "\n",
    "    def __len__(self):\n",
    "        return len(self.training)\n",
    "\n",
    "    def __getitem__(self, idx):\n",
    "        if torch.is_tensor(idx):\n",
    "            idx = idx.tolist()\n",
    "\n",
    "        img_name = os.path.join(self.root_dir,\n",
    "                                self.training.iloc[idx, 0])\n",
    "        image = io.imread(img_name)\n",
    "        name = self.training.iloc[idx, 1]\n",
    "        classes = self.training.iloc[idx,2]\n",
    "        sample = {'image': image, 'name': name,'classes':classes}\n",
    "\n",
    "        if self.transform:\n",
    "            sample = self.transform(sample)\n",
    "\n",
    "        return sample"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "``Intializing`` Class and representing few data points"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 (400, 400, 3)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Description: Marc Jacobs Beauty  Eye-Conic Longwear Eyeshadow Palette - Scandalust 740  \n",
      "Classes: beauty<makeup<eyeshadow\n",
      "1 (400, 400, 3)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Description: Marc Jacobs Beauty  Eye-Conic Longwear Eyeshadow Palette - Provocouture 710  \n",
      "Classes: beauty<makeup<eyeshadow\n",
      "2 (400, 400, 3)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Description: Prada  Wool sweater   \n",
      "Classes: clothing<knitwear<fine knit\n",
      "3 (400, 400, 3)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Description: Prada  Printed silk-satin twill straight-leg pants  \n",
      "Classes: clothing<pants<straight leg\n"
     ]
    }
   ],
   "source": [
    "image_dataset = ImageClassesDataset(csv_file='training.csv',\n",
    "                                    root_dir='images')\n",
    "\n",
    "fig = plt.figure()\n",
    "\n",
    "for i in range(4):\n",
    "    sample = image_dataset[i]\n",
    "\n",
    "    print(i, sample['image'].shape)\n",
    "\n",
    "    ax = plt.subplot(1, 4, i + 1)\n",
    "    plt.tight_layout()\n",
    "    ax.set_title('Sample #{}'.format(i))\n",
    "    ax.axis('off')\n",
    "    show_classes(**sample)\n",
    "\n",
    "    if i == 3:\n",
    "        plt.show()\n",
    "        break"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## ``Part 1 (a)`` Data Augmentation:\n",
    "* ``Recaling`` of image\n",
    "* ``Random Crop`` of image\n",
    "* ``Normalization`` of image with ``mean : `` [0.485, 0.456, 0.406], ``std : `` [0.229, 0.224, 0.225]\n",
    "\n",
    "### `` References:``\n",
    "* https://pytorch.org/tutorials/beginner/data_loading_tutorial.html#transforms"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Rescale(object):\n",
    "    \"\"\"Rescale the image in a sample to a given size.\n",
    "\n",
    "    Args:\n",
    "        output_size (tuple or int): Desired output size. If tuple, output is\n",
    "            matched to output_size. If int, smaller of image edges is matched\n",
    "            to output_size keeping aspect ratio the same.\n",
    "    \"\"\"\n",
    "\n",
    "    def __init__(self, output_size):\n",
    "        assert isinstance(output_size, (int, tuple))\n",
    "        self.output_size = output_size\n",
    "\n",
    "    def __call__(self, sample):\n",
    "        image,name,classes = sample['image'],sample['name'],sample['classes']\n",
    "\n",
    "        h, w = image.shape[:2]\n",
    "        if isinstance(self.output_size, int):\n",
    "            if h > w:\n",
    "                new_h, new_w = self.output_size * h / w, self.output_size\n",
    "            else:\n",
    "                new_h, new_w = self.output_size, self.output_size * w / h\n",
    "        else:\n",
    "            new_h, new_w = self.output_size\n",
    "\n",
    "        new_h, new_w = int(new_h), int(new_w)\n",
    "\n",
    "        img = transform.resize(image, (new_h, new_w))\n",
    "\n",
    "        # h and w are swapped for landmarks because for images,\n",
    "        # x and y axes are axis 1 and 0 respectively\n",
    "        # landmarks = landmarks * [new_w / w, new_h / h]\n",
    "\n",
    "        return {'image': img,'name':name ,'classes': classes}\n",
    "\n",
    "\n",
    "class RandomCrop(object):\n",
    "    \"\"\"Crop randomly the image in a sample.\n",
    "\n",
    "    Args:\n",
    "        output_size (tuple or int): Desired output size. If int, square crop\n",
    "            is made.\n",
    "    \"\"\"\n",
    "\n",
    "    def __init__(self, output_size):\n",
    "        assert isinstance(output_size, (int, tuple))\n",
    "        if isinstance(output_size, int):\n",
    "            self.output_size = (output_size, output_size)\n",
    "        else:\n",
    "            assert len(output_size) == 2\n",
    "            self.output_size = output_size\n",
    "\n",
    "    def __call__(self, sample):\n",
    "        image,name,classes = sample['image'],sample['name'],sample['classes']\n",
    "\n",
    "        h, w = image.shape[:2]\n",
    "        new_h, new_w = self.output_size\n",
    "\n",
    "        top = np.random.randint(0, h - new_h)\n",
    "        left = np.random.randint(0, w - new_w)\n",
    "\n",
    "        image = image[top: top + new_h,\n",
    "                      left: left + new_w]\n",
    "\n",
    "        #landmarks = landmarks - [left, top]\n",
    "\n",
    "        return {'image': image,'name':name ,'classes': classes}\n",
    "\n",
    "\n",
    "class ToTensor(object):\n",
    "    \"\"\"Convert ndarrays in sample to Tensors.\"\"\"\n",
    "\n",
    "    def __call__(self, sample):\n",
    "        image,name,classes = sample['image'],sample['name'],sample['classes']\n",
    "\n",
    "        # swap color axis because\n",
    "        # numpy image: H x W x C\n",
    "        # torch image: C X H X W\n",
    "        image = image.transpose((2, 0, 1))\n",
    "        return {'image': torch.from_numpy(image),'name' :name ,'classes': classes}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Showing ``example`` on ``65th datapoint``"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "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"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Description: Charlotte Tilbury  Hollywood Lips Matte Contour Liquid Lipstick – Show Girl  \n",
      "Classes: beauty<makeup<lipstick\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Description: Charlotte Tilbury  Hollywood Lips Matte Contour Liquid Lipstick – Show Girl  \n",
      "Classes: beauty<makeup<lipstick\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Description: Charlotte Tilbury  Hollywood Lips Matte Contour Liquid Lipstick – Show Girl  \n",
      "Classes: beauty<makeup<lipstick\n"
     ]
    }
   ],
   "source": [
    "scale = Rescale(256)\n",
    "crop = RandomCrop(128)\n",
    "composed = transforms.Compose([Rescale(256),\n",
    "                               RandomCrop(224)])\n",
    "\n",
    "# Apply each of the above transforms on sample.\n",
    "fig = plt.figure()\n",
    "sample = image_dataset[65]\n",
    "for i, tsfrm in enumerate([scale, crop, composed]):\n",
    "    transformed_sample = tsfrm(sample)\n",
    "\n",
    "    ax = plt.subplot(1, 3, i + 1)\n",
    "    plt.tight_layout()\n",
    "    ax.set_title(type(tsfrm).__name__)\n",
    "    show_classes(**transformed_sample)\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 torch.Size([3, 224, 224]) beauty<makeup<eyeshadow\n",
      "1 torch.Size([3, 224, 224]) beauty<makeup<eyeshadow\n",
      "2 torch.Size([3, 224, 224]) clothing<knitwear<fine knit\n",
      "3 torch.Size([3, 224, 224]) clothing<pants<straight leg\n"
     ]
    }
   ],
   "source": [
    "transformed_dataset = ImageClassesDataset(csv_file='training.csv',\n",
    "                                           root_dir='images',\n",
    "                                           transform=transforms.Compose([\n",
    "                                               Rescale(256),\n",
    "                                               RandomCrop(224),\n",
    "                                               ToTensor()\n",
    "                                           ]))\n",
    "\n",
    "for i in range(len(transformed_dataset)):\n",
    "    sample = transformed_dataset[i]\n",
    "\n",
    "    print(i, sample['image'].size(), sample['classes'])\n",
    "\n",
    "    if i == 3:\n",
    "        break"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'image': tensor([[[0.9529, 0.9529, 0.9529,  ..., 0.9529, 0.9529, 0.9529],\n",
       "          [0.9529, 0.9529, 0.9529,  ..., 0.9529, 0.9529, 0.9529],\n",
       "          [0.9529, 0.9529, 0.9529,  ..., 0.9529, 0.9529, 0.9529],\n",
       "          ...,\n",
       "          [0.9529, 0.9529, 0.9529,  ..., 0.9529, 0.9529, 0.9529],\n",
       "          [0.9529, 0.9529, 0.9529,  ..., 0.9529, 0.9529, 0.9529],\n",
       "          [0.9529, 0.9529, 0.9529,  ..., 0.9529, 0.9529, 0.9529]],\n",
       " \n",
       "         [[0.9490, 0.9490, 0.9490,  ..., 0.9490, 0.9490, 0.9490],\n",
       "          [0.9490, 0.9490, 0.9490,  ..., 0.9490, 0.9490, 0.9490],\n",
       "          [0.9490, 0.9490, 0.9490,  ..., 0.9490, 0.9490, 0.9490],\n",
       "          ...,\n",
       "          [0.9490, 0.9490, 0.9490,  ..., 0.9490, 0.9490, 0.9490],\n",
       "          [0.9490, 0.9490, 0.9490,  ..., 0.9490, 0.9490, 0.9490],\n",
       "          [0.9490, 0.9490, 0.9490,  ..., 0.9490, 0.9490, 0.9490]],\n",
       " \n",
       "         [[0.9333, 0.9333, 0.9333,  ..., 0.9333, 0.9333, 0.9333],\n",
       "          [0.9333, 0.9333, 0.9333,  ..., 0.9333, 0.9333, 0.9333],\n",
       "          [0.9333, 0.9333, 0.9333,  ..., 0.9333, 0.9333, 0.9333],\n",
       "          ...,\n",
       "          [0.9333, 0.9333, 0.9333,  ..., 0.9333, 0.9333, 0.9333],\n",
       "          [0.9333, 0.9333, 0.9333,  ..., 0.9333, 0.9333, 0.9333],\n",
       "          [0.9333, 0.9333, 0.9333,  ..., 0.9333, 0.9333, 0.9333]]],\n",
       "        dtype=torch.float64),\n",
       " 'name': 'Charlotte Tilbury  Hollywood Lips Matte Contour Liquid Lipstick – Show Girl  ',\n",
       " 'classes': 'beauty<makeup<lipstick'}"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "transformed_dataset[65]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "``Part 1(a):`` Image Embeddings\n",
    "-------------------------------------------\n",
    "\n",
    "* Using ``resnet-18`` for image embeddings\n",
    "* Returns a vector of length 512\n",
    "\n",
    "### `` References :``\n",
    "* https://github.com/christiansafka/img2vec\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Showing a ``example``"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Tarun Makkar\\Anaconda3\\envs\\keras-gpu\\lib\\site-packages\\torchvision\\transforms\\transforms.py:219: UserWarning: The use of the transforms.Scale transform is deprecated, please use transforms.Resize instead.\n",
      "  warnings.warn(\"The use of the transforms.Scale transform is deprecated, \" +\n"
     ]
    }
   ],
   "source": [
    "# Initialize Img2Vec with GPU\n",
    "img2vec = Img2Vec(cuda=True)\n",
    "\n",
    "# Read in an image\n",
    "img = Image.open('images/5da816f86504fb65cea6fed6_0.jpg')\n",
    "# Get a vector from img2vec, returned as a torch FloatTensor\n",
    "vec = img2vec.get_vec(img, tensor=False)\n",
    "# Or submit a list\n",
    "# vectors = img2vec.get_vec(list_of_PIL_images)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(512,)"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "vec.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Below code take ``2 hrs`` , So running it once and saving it to ``numpy file``\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "image_embeddings = []\n",
    "for image in tqdm.tqdm(os.listdir('images')):\n",
    "    img = Image.open('images'+'/'+image)\n",
    "    vec = img2vec.get_vec(img, tensor=False)\n",
    "    image_embeddings.append(vec)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "np.save('image_embeddings.npy', image_embeddings, allow_pickle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "image_embeddings = np.load('image_embeddings.npy', allow_pickle=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "``Part 1(b):`` Text Preprocessing\n",
    "---------------------------------------------\n",
    "\n",
    "* Preprocessing ``name`` feature\n",
    "* Preprocessing includes: \n",
    "            * Removing spacial character,\n",
    "            * removing stopwords\n",
    "            * Coverting all characters into lowercase\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "training = pd.read_csv('training.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Charlotte Tilbury  Hollywood Lips Matte Contour Liquid Lipstick – Show Girl  \n"
     ]
    }
   ],
   "source": [
    "print(training['name'][65])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "sent = training['name'][65]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Charlotte Tilbury Hollywood Lips Matte Contour Liquid Lipstick Show Girl \n"
     ]
    }
   ],
   "source": [
    "#remove spacial character: https://stackoverflow.com/a/5843547/4084039\n",
    "sent = re.sub('[^A-Za-z0-9]+', ' ', sent)\n",
    "sent = re.sub(r'[0-9]+',' ',sent)\n",
    "print(sent)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "# https://gist.github.com/sebleier/554280\n",
    "# we are removing the words from the stop words list: 'no', 'nor', 'not'\n",
    "stopwords= ['i', 'me', 'my', 'myself', 'we', 'our', 'ours', 'ourselves', 'you', \"you're\", \"you've\",\\\n",
    "            \"you'll\", \"you'd\", 'your', 'yours', 'yourself', 'yourselves', 'he', 'him', 'his', 'himself', \\\n",
    "            'she', \"she's\", 'her', 'hers', 'herself', 'it', \"it's\", 'its', 'itself', 'they', 'them', 'their',\\\n",
    "            'theirs', 'themselves', 'what', 'which', 'who', 'whom', 'this', 'that', \"that'll\", 'these', 'those', \\\n",
    "            'am', 'is', 'are', 'was', 'were', 'be', 'been', 'being', 'have', 'has', 'had', 'having', 'do', 'does', \\\n",
    "            'did', 'doing', 'a', 'an', 'the', 'and', 'but', 'if', 'or', 'because', 'as', 'until', 'while', 'of', \\\n",
    "            'at', 'by', 'for', 'with', 'about', 'against', 'between', 'into', 'through', 'during', 'before', 'after',\\\n",
    "            'above', 'below', 'to', 'from', 'up', 'down', 'in', 'out', 'on', 'off', 'over', 'under', 'again', 'further',\\\n",
    "            'then', 'once', 'here', 'there', 'when', 'where', 'why', 'how', 'all', 'any', 'both', 'each', 'few', 'more',\\\n",
    "            'most', 'other', 'some', 'such', 'only', 'own', 'same', 'so', 'than', 'too', 'very', \\\n",
    "            's', 't', 'can', 'will', 'just', 'don', \"don't\", 'should', \"should've\", 'now', 'd', 'll', 'm', 'o', 're', \\\n",
    "            've', 'y', 'ain', 'aren', \"aren't\", 'couldn', \"couldn't\", 'didn', \"didn't\", 'doesn', \"doesn't\", 'hadn',\\\n",
    "            \"hadn't\", 'hasn', \"hasn't\", 'haven', \"haven't\", 'isn', \"isn't\", 'ma', 'mightn', \"mightn't\", 'mustn',\\\n",
    "            \"mustn't\", 'needn', \"needn't\", 'shan', \"shan't\", 'shouldn', \"shouldn't\", 'wasn', \"wasn't\", 'weren', \"weren't\", \\\n",
    "            'won', \"won't\", 'wouldn', \"wouldn't\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tqdm import tqdm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|█████████████████████████████████████████████████████████████████████████| 65714/65714 [00:02<00:00, 31923.37it/s]\n"
     ]
    }
   ],
   "source": [
    "\"\"\"Preprocessing Step\"\"\"\n",
    "\n",
    "preprocessed_names = []\n",
    "# tqdm is for printing the status bar\n",
    "for i in tqdm(range(len(training))):\n",
    "    sent = training['name'][i]\n",
    "    sent = re.sub('[^A-Za-z0-9]+', ' ', sent)\n",
    "    sent = re.sub(r'[0-9]+',' ',sent)\n",
    "    # https://gist.github.com/sebleier/554280\n",
    "    sent = ' '.join(e for e in sent.split() if e not in stopwords)\n",
    "    preprocessed_names.append(sent.lower().strip())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'charlotte tilbury hollywood lips matte contour liquid lipstick show girl'"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "preprocessed_names[65]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "``Part 1(b):`` Text Embeddings\n",
    "-------------------------------------------\n",
    "\n",
    "* Downloading ``glove.840B.300d.txt`` from https://nlp.stanford.edu/projects/glove/\n",
    "* Converting ``glove`` file to ``word2vec`` format\n",
    "* Loading word2vec model using ``gensim``\n",
    "* Using ``Stanford gloVe`` i.e a pretrained model\n",
    "* And encoding ``name``feature with ``TFIDF weighted-word2vec``\n",
    "* Saving embeddings to numpy file\n",
    "\n",
    "``TFIDF`` weighted-word2vec\n",
    "---\n",
    "\n",
    "For eg - \n",
    " \n",
    "Sentence = Computer is electronic device\n",
    "\n",
    "preprocessed_dev = Computer electronic device\n",
    "\n",
    "    w1 = Computer \n",
    "    w2 = electronin\n",
    "    w3 = device\n",
    "\n",
    "Result = \n",
    "    # tfidf(w1)*glove(w1) + tfidf(w2)*glove(w2) + tfidf(w3)*glove(w3) / (tfidf(w1)+tfidf(w2)+ tfdif(w3))\n",
    "\n",
    "\n",
    "### `` References:``\n",
    "* https://machinelearningmastery.com/develop-word-embeddings-python-gensim/\n",
    "* https://github.com/makkarss929/KNN-on-Donor-Choose-Dataset/blob/master/KNN_SUBMIT.ipynb"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Saving ``glove file`` in ``word2vec`` format"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from gensim.scripts.glove2word2vec import glove2word2vec\n",
    "glove_input_file = 'glove.840B.300d.txt'\n",
    "word2vec_output_file = 'glove.840B.300d.txt.word2vec'\n",
    "glove2word2vec(glove_input_file, word2vec_output_file)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Loading ``glove`` model using ``gensim``"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "from gensim.models import KeyedVectors\n",
    "# load the Stanford GloVe model\n",
    "filename = 'glove.840B.300d.txt.word2vec'\n",
    "glove_w2v = KeyedVectors.load_word2vec_format(filename, binary=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### TFIDF Weighted ``Word2Vec``"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.feature_extraction.text import TfidfVectorizer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "tfidf_model = TfidfVectorizer()\n",
    "tfidf_model.fit(preprocessed_names)\n",
    "# we are converting a dictionary with word as a key, and the idf as a value\n",
    "dictionary = dict(zip(tfidf_model.get_feature_names(), list(tfidf_model.idf_)))\n",
    "tfidf_words = set(tfidf_model.get_feature_names())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "``Tfidf with glove`` take ``10 hrs`` to execute"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# Compute tfidf weighted word2vec for each point\n",
    "preprocessed_names_tfidf_wt_w2v_vectors = []\n",
    "for sentence in tqdm(preprocessed_names): # for each sentence\n",
    "    vector = np.zeros(300) # as word vectors are of zero length\n",
    "    tf_idf_weight =0; # num of words with a valid vector in the sentence\n",
    "    for word in sentence.split(): # for each word in a sentence\n",
    "        if (word in list(glove_w2v.wv.vocab)) and (word in tfidf_words):\n",
    "            vec = glove_w2v[word] # getting the vector for each word\n",
    "            # here we are multiplying idf value(dictionary[word]) and the tf value((sentence.count(word)/len(sentence.split())))\n",
    "            tf_idf = dictionary[word]*(sentence.count(word)/len(sentence.split())) # getting the tfidf value for each word\n",
    "            vector += (vec * tf_idf) # calculating tfidf weighted w2v\n",
    "            tf_idf_weight += tf_idf\n",
    "    if tf_idf_weight != 0:\n",
    "        vector /= tf_idf_weight\n",
    "    preprocessed_names_tfidf_wt_w2v_vectors.append(vector)\n",
    "\n",
    "print(len(preprocessed_names_tfidf_wt_w2v_vectors))\n",
    "print(len(preprocessed_names_tfidf_wt_w2v_vectors[0]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Saving ``Text embeddings`` to numpy file"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "np.save('text_embeddings.npy', preprocessed_names_tfidf_wt_w2v_vectors, allow_pickle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "text_embeddings = np.load('text_embeddings.npy', allow_pickle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "text_embeddings = torch.from_numpy(text_embeddings)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "image_embeddings = np.load('image_embeddings.npy', allow_pickle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "image_embeddings = torch.from_numpy(image_embeddings)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "``Part 3(b): `` Label Embeddings (Semantic Similarity)\n",
    "---------------------------\n",
    "* ``Label Embeddings`` used for ``Semantic Similarity``\n",
    "* ``One hot encoding`` labels for compution of ``class level similarity``\n",
    "* Using ``InferSent : Facebook's pretrained model`` for compution of ``semantic similarity``\n",
    "* Downloading ``InferSent1.pkl`` from https://dl.fbaipublicfiles.com/infersent/infersent1.pkl \n",
    "* ``InferSent1.pkl`` contains pretrained weights\n",
    "* Loading Model using pretrained weights,creating vocab and encoding\n",
    "* Returns ``4096 dim`` of each label\n",
    "\n",
    "\n",
    "### ``References :``\n",
    "* https://github.com/facebookresearch/InferSent"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "filename = 'training.csv'\n",
    "mapping_csv = pd.read_csv(filename)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create a ``set`` of ``labels`` and ``sorting`` it alphabetically"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create a set of labels\n",
    "labels = set()\n",
    "for i in range(len(mapping_csv)):\n",
    "\t# convert spaced separated tags into an array of tags\n",
    "\ttags = mapping_csv['classes'][i].split('<')\n",
    "\t# add tags to the set of known labels\n",
    "\tlabels.update(tags)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# convert set of labels to a list to list\n",
    "labels = list(labels)\n",
    "# order set alphabetically\n",
    "labels.sort()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "67"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "``Note :`` There are 67 unique labels"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### ``Labels mapping``\n",
    "* Each ``label`` is assigned a ``unique number``. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# dict that maps labels to integers, and the reverse\n",
    "labels_map = {labels[i]:i for i in range(len(labels))}\n",
    "inv_labels_map = {i:labels[i] for i in range(len(labels))}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Part 3(a): ``One hot encoding`` for class level similarity"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "def one_hot_encode(tags, mapping):\n",
    "\t# create empty vector\n",
    "\tencoding = np.zeros(len(mapping), dtype='uint8')\n",
    "\t# mark 1 for each tag in the vector\n",
    "\tfor tag in tags:\n",
    "\t\tencoding[mapping[tag]] = 1\n",
    "\treturn encoding"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Showing ``Example``"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "tags = mapping_csv['classes'][65].split('<')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,\n",
       "       0], dtype=uint8)"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "one_hot_encode(tags,labels_map)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "one_hot_labels = []\n",
    "for i in range(len(mapping_csv)):\n",
    "    tags = mapping_csv['classes'][i].split('<')\n",
    "    one_hot_labels.append(one_hot_encode(tags,labels_map))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "one_hot_labels = torch.from_numpy(np.array(one_hot_labels))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([65714, 67])"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "one_hot_labels.size()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "``Labels embeddings``\n",
    "----------------------\n",
    "* Constructing label embedding using ``InferSent : Facebook's Pretrained Model``\n",
    "* Returns embedding of 4096 dimension."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Loading ``InferSent``"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "import InferSent"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "from InferSent import models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[nltk_data] Downloading package punkt to C:\\Users\\Tarun\n",
      "[nltk_data]     Makkar\\AppData\\Roaming\\nltk_data...\n",
      "[nltk_data]   Package punkt is already up-to-date!\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import nltk\n",
    "nltk.download('punkt')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Loading ``InferSent1.pkl`` : pretrained weights "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<All keys matched successfully>"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Load model\n",
    "#from models import InferSent\n",
    "model_version = 1\n",
    "MODEL_PATH = \"encoder/infersent%s.pkl\" % model_version\n",
    "params_model = {'bsize': 64, 'word_emb_dim': 300, 'enc_lstm_dim': 2048,\n",
    "                'pool_type': 'max', 'dpout_model': 0.0, 'version': model_version}\n",
    "model = models.InferSent(params_model)\n",
    "model.load_state_dict(torch.load(MODEL_PATH))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Keep it on CPU or put it on GPU\n",
    "use_cuda = False\n",
    "model = model.cuda() if use_cuda else model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "``InferSent`` uses ``glove`` file"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "# If infersent1 -> use GloVe embeddings. If infersent2 -> use InferSent embeddings.\n",
    "W2V_PATH = 'glove.840B.300d.txt' if model_version == 1 else 'fastText/crawl-300d-2M.vec'\n",
    "model.set_w2v_path(W2V_PATH)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Updating ``Vocabulary`` using ``Labels``"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "sentences = labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 76(/76) words with w2v vectors\n",
      "Vocab size : 76\n"
     ]
    }
   ],
   "source": [
    "# Load embeddings of K most frequent words\n",
    "model.build_vocab(sentences,tokenize=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['accessories']"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "[inv_labels_map[0]]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Showing example of `` accessories``"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 0.11502168, -0.06020666, -0.0304451 , ..., -0.03926746,\n",
       "       -0.03814262, -0.0289226 ], dtype=float32)"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.encode([inv_labels_map[0]])[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### ``Part 3(b):``Adjacency Matrix (Semantic Similarity)\n",
    "* Constructing Adjacency matrix for ``semantic graph``\n",
    "* Taking each ``label Embeddings`` as node and ``edge weight`` is cosine distance between 2 nodes\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(67, 4096)"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "emb = model.encode(labels,tokenize = True)\n",
    "\n",
    "emb.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "A = np.zeros((len(labels),len(labels)))\n",
    "for i in range(A.shape[0]):\n",
    "    for j in range(A.shape[1]):\n",
    "        A[i][j] = cosine(emb[i],emb[j])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[0.         0.65449536 0.38661617 ... 0.41142404 0.46810478 0.63951567]\n",
      " [0.65449536 0.         0.61885324 ... 0.62294078 0.60166988 0.34177005]\n",
      " [0.38661617 0.61885324 0.         ... 0.10404348 0.32019562 0.62520388]\n",
      " ...\n",
      " [0.41142404 0.62294078 0.10404348 ... 0.         0.30119473 0.61426258]\n",
      " [0.46810478 0.60166988 0.32019562 ... 0.30119473 0.         0.59520289]\n",
      " [0.63951567 0.34177005 0.62520388 ... 0.61426258 0.59520289 0.        ]]\n"
     ]
    }
   ],
   "source": [
    "print(A)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "``Part 2 :`` Defining Model Architecture\n",
    "----------------\n",
    "* Implementing ``Image Tower`` and ``Text Tower``.\n",
    "\n",
    "* ``The image tower`` consists of ``5 hidden layers`` of ``512 hidden units`` each.\n",
    "\n",
    "* ``Text tower consists`` of ``2 hidden layers`` of ``512 hidden units`` each.  \n",
    "* A ``dropout`` of ``0.15`` is used between all hidden layers of ``both towers``. \n",
    "* ``Both Towers`` consists of ``RELU`` non-linearity .\n",
    "* ``5th Layer`` of both towers should have ``same dimension`` as it ``shared hidden layer.``\n",
    "* The network is trained using the ``RMSProp optimizer`` with a ``learning rate`` of ``1.6192e-05``and  ``momentum``  set  to  ``0.9``  with  random  batches  of  1024 for 250,000 steps.  "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "``Shared Hidden Layer``(Late Fusion)\n",
    "----------------------------------\n",
    "* Last layer of both Towers is ``512``units.\n",
    "* ``Concatenating`` Last hidden Layers from ``both towers``.\n",
    "* After ``Concatenation`` it becomes ``1024``.\n",
    "* Then adding ``Extra Layer`` on ``concatenated Layer`` of ``512`` units."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch.nn.functional as F"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "class HUSE(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(HUSE, self).__init__()\n",
    "        \n",
    "        # Image Tower\n",
    "        self.image_tower = nn.Sequential(\n",
    "            # 1st Layer \n",
    "            nn.Linear(512,512),  # (N,512) -> (N,512)\n",
    "            nn.Dropout(0.15),\n",
    "            nn.ReLU(),\n",
    "            # 2nd Layer \n",
    "            nn.Linear(512,512),  # (N,512) -> (N,512)\n",
    "            nn.Dropout(0.15),\n",
    "            nn.ReLU(),\n",
    "            # 3rd Layer\n",
    "            nn.Linear(512,512),  # (N,512) -> (N,512)\n",
    "            nn.Dropout(0.15),\n",
    "            nn.ReLU(),\n",
    "            # 4th Layer\n",
    "            nn.Linear(512,512),  # (N,512) -> (N,512)\n",
    "            nn.Dropout(0.15),\n",
    "            nn.ReLU(),\n",
    "            # L2 Normalization \n",
    "            # nn.LayerNorm(512)\n",
    "        \n",
    "            )\n",
    "        \n",
    "        # Text Tower\n",
    "        self.text_tower = nn.Sequential(\n",
    "            # 1st Layer\n",
    "            nn.Linear(300,512),  # (N,300) -> (N,512)\n",
    "            nn.Dropout(0.15),\n",
    "            nn.ReLU(),\n",
    "            # L2 Normalization\n",
    "            # nn.LayerNorm(512)\n",
    "            )  \n",
    "        \n",
    "        self.shared_hidden_layer = nn.Sequential(\n",
    "            # Last Shared Layer\n",
    "            nn.Linear(1024,512),\n",
    "            nn.Dropout(0.15),\n",
    "            nn.ReLU(),\n",
    "            )\n",
    "        self.shared_output_layer = nn.Sequential(\n",
    "            nn.Linear(512,67),\n",
    "            nn.Dropout(0.15),\n",
    "            nn.ReLU(),\n",
    "            )\n",
    "        \n",
    "        self.softmax = nn.Softmax(dim=0)\n",
    "        \n",
    "    # Cross Modal Similarity\n",
    "    \n",
    "    \n",
    "        \n",
    "    # Cross entropy Similarity \n",
    "    def cross_entropy(self,predictions, targets):\n",
    "        # Converting into numpy\n",
    "        \n",
    "        predictions = predictions.detach().numpy()\n",
    "        targets = targets.detach().numpy()\n",
    "        \n",
    "        N = predictions.shape[0]\n",
    "        ce = -np.sum(targets * np.log(predictions)) / N\n",
    "        \n",
    "        # Converting back to tensor\n",
    "        ce = np.array(ce)\n",
    "        ce = torch.from_numpy(ce)\n",
    "        return ce\n",
    "    \n",
    "    # Semantic Similarity\n",
    "    def semantic_similarity(self,phi,A):\n",
    "        \n",
    "        # Converting to numpy array\n",
    "        phi = phi.detach().numpy()\n",
    "        \n",
    "        \n",
    "        D = np.zeros(A.shape)\n",
    "        sigma = np.zeros(A.shape)\n",
    "        threshold = 1\n",
    "\n",
    "        for m in tqdm(range(phi.shape[0])):\n",
    "            for i in range(phi.shape[1]):\n",
    "                for n in range(phi.shape[0]):\n",
    "                    for j in range(phi.shape[1]):\n",
    "                        D[i][j] = (cosine(phi[m][i],phi[n][j]))\n",
    "                        sigma[i][j] = 1.0 if (A[i][j]<threshold and D[i][j] < threshold) else 0.0\n",
    "\n",
    "        N = phi.shape[0]\n",
    "        result =  np.sum(sigma*((D - A)**2))/(N**2)\n",
    "        result = np.array(result)\n",
    "        \n",
    "        result = torch.from_numpy(result)\n",
    "        \n",
    "        return result\n",
    "        \n",
    "            \n",
    "    def forward(self,x1,x2):\n",
    "        #Calling Image Tower\n",
    "        x1 = self.image_tower(x1)\n",
    "        # L2 Normalization \n",
    "        x1 = F.normalize(x1)\n",
    "        \n",
    "        #Calling Text Tower\n",
    "        x2 = self.text_tower(x2)\n",
    "        # L2 normalization\n",
    "        x2 = F.normalize(x2)\n",
    "        \n",
    "        \n",
    "        # Late Fusion \n",
    "        x = torch.cat((x1,x2),dim = 1)\n",
    "        # Shared hidden layer\n",
    "        x = self.shared_hidden_layer(x)\n",
    "        # Unnormalized output layer\n",
    "        x = self.shared_output_layer(x)\n",
    "        \n",
    "        # Softmax on Unnormalized output layer\n",
    "        # x = F.softmax(x, dim=0)\n",
    "        x = self.softmax(x)\n",
    "        return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "net = HUSE()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "``Part 3(a):`` Class level similarity\n",
    "------\n",
    "* Loading ``BinaryCrossEntropyLoss`` from torch.nn\n",
    "* Using it for calculating ``loss`` between ``One hot labels`` and ``predicted labels.``\n",
    "\n",
    "\n",
    "``Part 3(c):`` Cross Modal gap\n",
    "-----\n",
    "* Using `` torch.nn.CosineSimilarity`` for calculating Cross Modal gap.\n",
    "* As we now,\n",
    "\n",
    "     ### cosine distance = (1 - cosine similarity) \n",
    "    \n",
    "    \n",
    " \n",
    "\n",
    "* Using ``above equation``, we calculate this loss"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Loading ``Loss functions`` and ``Optimizers``"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "loss_fn_1 = nn.BCELoss()\n",
    "optimizer = torch.optim.RMSprop(net.parameters(), lr=1.6192e-05, momentum =0.9)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Loading ``weights`` of loss function:\n",
    "* ``alpha`` for ``class level similarity``\n",
    "* ``beta`` for ``semantic similarity ``\n",
    "* ``gamma`` for ``cross modal gap``\n",
    "\n",
    "``Part 3:`` Total Loss Function\n",
    "----\n",
    "* Weighted Average of all 3 losses:\n",
    "         Total loss = alpha * class_level_similarity  + beta * semantic_similarity + gamma * cross_modal_gap"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "alpha = 10\n",
    "gamma = 1\n",
    "beta = 2"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### ``Note`` : Giving more ``weightage`` to ``class level similarity``"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Intializing few ``parameters``:\n",
    "* Number of epochs.\n",
    "* Batch_size"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### ``Training`` our Model: \n",
    "    # Number of epochs = 1000\n",
    "    # Batch_size = 128"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 199/1000,Total Loss : 1.92 ,Class Similarity Loss : 0.18, Cross Modal Loss: 0.17\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 399/1000,Total Loss : 1.85 ,Class Similarity Loss : 0.17, Cross Modal Loss: 0.17\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 599/1000,Total Loss : 1.82 ,Class Similarity Loss : 0.17, Cross Modal Loss: 0.16\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 799/1000,Total Loss : 1.79 ,Class Similarity Loss : 0.16, Cross Modal Loss: 0.16\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYIAAAEGCAYAAABo25JHAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy8QZhcZAAAgAElEQVR4nO3dd3zU9f3A8dc7l00GK4RtUARkyJClOAERt9XWUfeo1dr+cFTFWVvbqm2l1WpV1FarFOtAraAoKAqIjIDMhD0DAcJKAtnJ5/fH93uXG9+ERPLNHdz7+Xjkwd33+7m79x3Jvb+fLcYYlFJKRa+YcAeglFIqvDQRKKVUlNNEoJRSUU4TgVJKRTlNBEopFeViwx1AY7Vt29ZkZWWFOwyllDqqLF68eI8xJsPp3FGXCLKyssjOzg53GEopdVQRkS11ndOmIaWUinKaCJRSKsppIlBKqSiniUAppaKcJgKllIpymgiUUirKaSJQSqkoFzWJYM3OYp79Yg17D5aHOxSllIooUZMINhQc5O9frWfPwYpwh6KUUhElahJBnMd6q5XVNWGORCmlIksUJQIBoEITgVJKBYiaRBBv1wgqqjQRKKWUP9cSgYh0EZFZIpIrIqtEZJxDmXQR+UREltllbnYrnrhYbRpSSiknbq4+WgXcZ4xZIiKpwGIRmWGMyfErcxeQY4y5WEQygDUiMskY0+Q9uvHaR6CUUo5cqxEYY/KNMUvs28VALtApuBiQKiICpAD7sBJIk4vzNQ0ZN55eKaWOWs3SRyAiWcBAYEHQqReAk4AdwApgnDEm5JJdRG4XkWwRyS4oKPhBMcTHWp3FWiNQSqlAricCEUkBPgDuNsYUBZ0+D1gKdAQGAC+ISFrwcxhjJhpjBhtjBmdkOG6wc1hx2lmslFKOXE0EIhKHlQQmGWOmOBS5GZhiLOuBTUAvN2KJ185ipZRy5OaoIQFeB3KNMRPqKLYVGGWXzwR6AhvdiEcnlCmllDM3Rw2NAK4HVojIUvvYw0BXAGPMy8CTwBsisgIQ4EFjzB43gvE1DVVrZ7FSSvlzLREYY+ZifbnXV2YHMMatGPzp8FGllHIWNTOLfUtMaGexUkoFiJpEEOuJIUa0RqCUUsGiJhGA1U+gi84ppVSgqEoE8Z4YKnVmsVJKBYiqRBAXG6NNQ0opFSS6EoFHtLNYKaWCRFUiiNcagVJKhYiqRKCdxUopFSqqEkG8R2sESikVLKoSQZwnhkpdYkIppQJEVSKIj43RzmKllAoSVYkgziPaR6CUUkGiLBFoH4FSSgWLqkSgncVKKRUqqhJBnC4xoZRSIaIqEcTH6jwCpZQKFlWJIM6jo4aUUipYVCWC+FjRPgKllAoSVYlARw0ppVSoqEoE8TqzWCmlQkRVIojTmcVKKRUiuhKBvfqoMVorUEopr6hKBPEeAaCqRhOBUkp5RVUiiPNYb1c7jJVSqlZUJYL4WDsR6OxipZTyiapE4K0RlFdXhzkSpZSKHFGVCOJ9TUNaI1BKKa+oSgRxsVZncaUOIVVKKR/XEoGIdBGRWSKSKyKrRGRcHeXOFpGldplv3IoHIN7jAbSzWCml/MW6+NxVwH3GmCUikgosFpEZxpgcbwERaQn8AxhrjNkqIu1cjIc4e/iorkCqlFK1XKsRGGPyjTFL7NvFQC7QKajYT4EpxpitdrndbsUD1sxiQGcXK6WUn2bpIxCRLGAgsCDoVA+glYh8LSKLReSGOh5/u4hki0h2QUHBD45DO4uVUiqU64lARFKAD4C7jTFFQadjgVOAC4HzgMdEpEfwcxhjJhpjBhtjBmdkZPzgWHRCmVJKhXKzjwARicNKApOMMVMciuQBe4wxh4BDIjIb6A+sdSMe74Qy7SNQSqlabo4aEuB1INcYM6GOYh8DZ4hIrIgkA8Ow+hJc4W0aKq/UCWVKKeXlZo1gBHA9sEJEltrHHga6AhhjXjbG5IrIdGA5UAO8ZoxZ6VZAaUnW2y0qq3LrJZRS6qjjWiIwxswFpAHl/gz82a04/KUlxQFQVFrZHC+nlFJHhaiaWZwSH4uI1giUUspfVCWCmBghNSFWawRKKeUnqhIBWM1DRWWaCJRSyiv6EkFiHEWl2jSklFJe0ZcIkmK1RqCUUn6iLxEkxmkfgVJK+Ym+RJAUR7GOGlJKKZ/oSwRaI1BKqQDRlwiSYikur6K6RlcgVUopiMZEkGjNLi7UWoFSSgFRmAgy0xIB2FlYFuZIlFIqMkRdIujQ0koE+YWlYY5EKaUiQ9Qlgo7pSQDka41AKaWAKEwEGakJxMaI1giUUsoWdYnAEyNkpiWSf0BrBEopBVGYCAA6pCeyQ2sESikFRGki6NgyiR1aI1BKKSBKE0HnVknkF5bqpDKllCJqE0EyldWGnUVaK1BKqahMBF1aW0NI8/aVhDkSpZQKv6hMBJ1bJQOwbb92GCulVFQmgo4tExGBvP1aI1BKqahMBAmxHjJTE9m2T2sESikVlYkArH4CrREopVQ0J4JWyeRpH4FSSkVvIuiemcL2A6Vs2nMo3KEopVRYRW0iuKBvBwAWbNwb5kiUUiq8ojYRtE+39iXYe6gizJEopVR4uZYIRKSLiMwSkVwRWSUi4+opO0REqkXkx27FEywxzkNKQiwFxeXN9ZJKKRWR3KwRVAH3GWNOAoYDd4lI7+BCIuIBngE+dzEWRwfLq3hj3mZ26VITSqko5loiMMbkG2OW2LeLgVygk0PRXwEfALvdiqUuqQmxALyzcFtzv7RSSkWMZukjEJEsYCCwIOh4J+BHwMuHefztIpItItkFBQVNFtes+88mPjaGbTqfQCkVxVxPBCKSgnXFf7cxpijo9N+AB40x1fU9hzFmojFmsDFmcEZGRpPF1jYlgT4d03TbSqVUVIt188lFJA4rCUwyxkxxKDIYeEdEANoCF4hIlTHmIzfj8tchPZE56/ZwsLyKlARXPw6llIpIbo4aEuB1INcYM8GpjDGmmzEmyxiTBbwP/KI5kwDAZQM6UVxWxecrdzbnyyqlVMRws2loBHA9MFJElto/F4jIHSJyh4uv2yijT8qkVXIcCzbpxDKlVHRyrS3EGDMXkEaUv8mtWOoTEyP06ZjOh99v58nL+pIQ6wlHGEopFTZRO7PYX0JsDJXVhskLtoY7FKWUanaaCIC7RnYHYGeRzjJWSkUfTQTAoK6tiI0RXv5mA1XVNeEORymlmpUmAltVjQFg4eZ9YY5EKaWalyaCIHsP6mqkSqnooonAtvCRUQBsLNCNapRS0aVBiUBExolImlheF5ElIjLG7eCaU7vURHpmpvLW/M3aT6CUiioNrRHcYq8TNAbIAG4GnnYtqjC559wT2XOwggkz1oY7FKWUajYNTQTeiWEXAP8yxiyjEZPFjhbn9m5Pp5ZJzF7XdCucKqVUpGtoIlgsIl9gJYLPRSQVOObaTzwxwnl92rNh9yFq7FFESil1rGtoIrgVGA8MMcaUAHFYzUPHnIFdW1JaWc2M3F3hDkUppZpFQxPBqcAaY8wBEbkOeBQodC+s8LmgXwdSE2P5Zq02DymlokNDE8FLQImI9AceALYA/3YtqjDyxAj9O7fk2/V7qNTRQ0qpKNDQRFBljDHApcBzxpjngFT3wgqvK4d0YcveEr5dvyfcoSillOsamgiKReQhrP0FpomIB6uf4Jg0qlc7PDHCdxt1jwKl1LGvoYngKqAcaz7BTqAT8GfXogqzFgmxnN0jg3cWbqOorDLc4SillKsalAjsL/9JQLqIXASUGWOOyT4Cr1+O7E5haSV3TVpC3v6ScIejlFKuaegSE1cCC4GfAFcCC0Tkx24GFm79OqUDMGfdHsa9szTM0SillHsaulXlI1hzCHYDiEgGMBNrw/ljUqynNkeWVVaHMRKllHJXQ/sIYrxJwLa3EY89at1/Xk8AYkTIGj+NT5btCHNESinV9Br6ZT5dRD4XkZtE5CZgGvCpe2FFhrvO6c7YPu1Zsd2aO/evbzeFOSKllGp6DWoaMsbcLyJXACOwFpubaIz50NXIIkSX1km+2y0SGtqSppRSR48Gf7MZYz4APnAxlojUq32a73ZSnCeMkSillDvqTQQiUgw4LcMpgDHGpDmcO6YMzmrlu50Ur4lAKXXsqbePwBiTaoxJc/hJjYYkAHBcmxY8dXk/ABZt0o3tlVLHnmN+5E9TuGZoVwB2FJaxbldxmKNRSqmmpYmggXpkpgAwX9cfUkodY1xLBCLSRURmiUiuiKwSkXEOZa4VkeX2zzx7meuI9MbNQwF47ONVYY5EKaWalpvjIauA+4wxS+ytLReLyAxjTI5fmU3AWcaY/SJyPjARGOZiTD9Yu9QE3+1VOwopLK0kq00LOrZMqudRSikV+VxLBMaYfCDfvl0sIrlYq5bm+JWZ5/eQ+UBnt+I5UrGeGHp3SCMnv4gLn5/rO77pqQsQkTBGppRSR6ZZ+ghEJAsYCCyop9itwGd1PP52EckWkeyCgvBtIXnPuT1CjpVU6DpESqmjm+uJQERSsCai3W2MKaqjzDlYieBBp/PGmInGmMHGmMEZGRnuBXsYx7VJDjl2oFT3K1BKHd1cTQQiEoeVBCYZY6bUUeZk4DXgUmNMRA/J6ZGZyuJHR3PN0C6+Y4UlmgiUUkc3N0cNCfA6kGuMmVBHma7AFOB6Y8xat2JpSm1SEshq08J3/0BpRRijUUqpI+fmqKERWHscrxAR784uDwNdAYwxLwOPA22Af9gdrlXGmMEuxtQkbhqRRd7+Ut6av0VrBEqpo56bo4bmYq1JVF+Z24Db3IrBLQmxHu4b04O35m9h6z5rG8uD5VW8u2gbN52WRUyMjiJSSh09dGbxD9QyOZ52qQk89dlqyiqr+cO0XH43NYdLX/w23KEppVSjaCI4Ahf06wBAr8emM3nhVgBWbC/k3ext4QxLKaUaRRPBEXjikj4BM469Hnh/OXn7S8IQkVJKNZ4mgiN03fDjHI9XVjtt46CUUpFHE4FLSnXGsVLqKKGJ4Aj572Dm78mpOVRU1TRzNEop1XiaCI7QaSe0dTz+3ca9/HXmWm57M5tD5VXNHJVSSjWcJoImkBjn/DG+9PUGZubuYsqSvGaOSCmlGk4TQRM4r097AFY/OdbxfN7+0uYMRymlGsXNJSaixp9+fDL3n9eTxDiP4/lCXaFUKRXBtEbQBBJiPXRuZS1R/eV9Z9G9XUrA+X2HKtiy9xBV1dp5rJSKPJoImtgJGSnMvPesgGMrtxdy1p+/5q8zj4oFVpVSUUYTgUsuHdCRvp3SANhRWAbAi7M2UFKhI4iUUpFFE4FLnrt6IFN/dQY/P/P4gON/m7mO6hqddayUihzaWeyyzq2SAu5PnL2RVsnxrNxeyPAT2nB9HUtUKKVUc9FE4LKfDO7C3kMV/G3mOt+xZ6avBmDainxNBEqpsNOmIZclxnm4e3QPPht3RrhDUUopR5oImknH9KTDF1JKqTDQRNBMUhKdW+FqagyPfLiC1+duYldRGdmb9zVzZEqpaKd9BM3EEyP0yExh7a6DAcf3HCxn0gJrd7Mnp+YAsPDhUbRLS2z2GJVS0UlrBM3o87vPZPPTFwYc+8sXa0LKLdm6H4DdRWVMmLGWGofhpivyCtlVVOZOoEqpqKKJoBmJSMixd7Pz6JgeePVfYm9q8+AHy3n+y3V8v21/yOMufmEuI//ytStxKqWiizYNhUFCbAzlfpvW3D+2JzsLy33DSu99dxkVVTW+hFBexwY3h3QXNKVUE9AaQRhM+78zmHTbMJLiPKQnxTGmd3tuD5qBPH7KCnLyiwCoqjZ8+H0eL329gTU7iwPKzd+4l//YfQxKKfVDiDFH13IHgwcPNtnZ2eEOwxVb95Zw5p9nhRx/6dpB3Dlpie/+R3eN4LIXvw0oE9z3oJRS/kRksTFmsNM5rRFEkK5tkh2PF5UF7mewNqhWAHC0JXSlVOTQRHAUKCoNXLF0f0lFSJmDui+yUuoH0kRwFAgeNXTAYcezAyW6C5pS6odxLRGISBcRmSUiuSKySkTGOZQREXleRNaLyHIRGeRWPEezT1fsDLjv9KWviUAp9UO5WSOoAu4zxpwEDAfuEpHeQWXOB060f24HXnIxnqNCj8wUzuyRwcvXnVJnmQMOTUPB/QhKKdVQriUCY0y+MWaJfbsYyAU6BRW7FPi3scwHWopIB7diOhp8cc9Z/PuWoYzt2561vz+ff908JKSM09V/WWW179+s8dN467vNLkeqlDpWNEsfgYhkAQOBBUGnOgHb/O7nEZosEJHbRSRbRLILCgrcCjPixMfGcNaJGTx0fq+AfQu+27g3pOzzX67j1jcWkW9vi/ncl+t95zYWHAwpr5RSXq4nAhFJAT4A7jbGFAWfdnhIyDhIY8xEY8xgY8zgjIwMN8KMWDExws/POoHHL+7NR3eNqLPcsrxCvly9mwV2kqixh5NOX7mTkc9+wxerdtb5WKVUdHM1EYhIHFYSmGSMmeJQJA/o4ne/M7DDzZiOVnGeGAZ0aXnYcuOnrABqE4F3WetNew65F5xS6qjm5qghAV4Hco0xE+oo9j/gBnv00HCg0BiT71ZMx4JTj28TcH/8+b0cy1XbK5YWl1nzC1IT40LKFJdVMmnBFnYW6iqmSkUzN2sEI4DrgZEistT+uUBE7hCRO+wynwIbgfXAq8AvXIznmPDYRdbAq9SEWKbffQY3nFrbdzD5Z8N9t4vLqvjJy/P4b7bVBRMfG8OancX8d9FWKqutReye+F8Oj3y4kuFPfdmM70ApFWlcW33UGDMX5z4A/zIGuMutGI5F6cnWlX1SvIde7dOoqq5dmbRPp7SAsos2105Em5mzi1+/twyAXUXlXNy/I2t2BXfZKKWikS5DfZRpn5bI5QM7ceNpWQDEemordWkOzT9e0/06iyfMWMuEGWtDyqzfXUy7tESKy6ooKa/ixMzUpgtcKRWxNBEcZTwxwoSrBrjy3KMnzKZvpzRWbrdqCrqiqVLRQdcaOsaM7dP+Bz2uwt78xpsEvP793WYe+2jlkYallIpgmgiOMc9dM4C3bx3mu//kpX0a9Lh9h0KXrTDG8PjHq3hr/pYmi08pFXk0ERwDLh3QkQfHWsNIE2I9DO3W2nfu+lOzGvQcew+Vhxzbuq/Ed/vjpdsp1vWMlDomaSI4Bjx39UDuPPsE3/34WOf/1jNObBtyrFPLJAAufH5uyLnvtx7w3R73zlJtIlLqGKWdxVHgletPoWVSHKcc14qK6hou/8c8Vtu7nJ1xYlveWbTN8XF3/3dpwP3dxaG1BqXU0U8TwTHqmSv6+YZ/nufXgRzriaFdWqIvEXRvl9Lg50xNtH5dKqtrEAKHrjZERVUNsTFCTEy900uUUs1Mm4aOUVcN6cqgrq0cz/Xzm3jWNiUh5PyFJzuvBO5dpuLxj1fR5zefs/egVUPY79DRXFxWyeMfr2THgVLA6nju8ehnPPLRisa9EaWU6zQRRKG7R/fgmSv6sfjR0SEdwHeefQIvXDMQABH44M7TePfnpwIQGyNUVNUweeFWyqtq+GBJHlnjpzHwyRnMWr074Hmemb6af3+3hS/t44X29pqTFzo3QymlwkcTQRSK88Rw1ZCutElJoFOrpIBzd5x1AiLCksfOJfd3YznluFa+UUjvLNrGJS/Udir/8dPVvtvzNuwJeJ69B61aQqzdDJS336oZpCZYzUu7i8v4bsNeSiuqefqz1ZRUVDnGunpnEVnjp7F4y74jectKqXpoIohyI3tlMv3uM3z3U+wv6tYt4kmM84SU9/YtSFAz/yfL8pm6fAfTV+ZTWV3jWwa73N45zdtE1KpFPABXvzKfa16dz6QFW3j5mw28OnuTY3ze9ZLe0ZqEUq7RzmJFr/a1fQaeBnTktktNCBlBtLOojF/+53sAbj29G/Yq2DzxSQ4X9+/oWw47ziPU1Bg22vsjeCeyeZuOVm4vpEVCLN3atgCgrZ048nWpbKVcozUC1SDHZ7Tw3Z5xz1lcO6xrnWVn5OyipqZ2o7nL/vEte+yO5Q0Fhzj+4U995w6VWwnCuzT2RX+fyzl/+dp3vsI+vtehQ1op1TQ0EagGee6qgb7b6clx/O7Svnxcx9aZu4rKAnZE27avlLcXOC9Tsf2AdaX/1vwt3Pvu0pDzpRVW01JFVXWdsf3j6/VkjZ8WsCS3v7LK6jrPKaU0EagGSooP/FXxxAj9/bbOTPLrTyivqvE1/Xglxzm3Qnr3WAaYsmS773aZ3bdQav+7oeAQv35vmW9xPH9/m7kOgJz8IorLKvlq9S4AFm/Zx9JtB+j12HSufOW7kMcVlVWy/1BFQJKYs66AZdsOhJQ9nNveXMSPX5rX6McpFQk0ESgAPvzFafz39uF1nk+IDe049nfziKx6z5fVcUVfXO48WujBD5YDtYkA4P3FeXyztgCw5i6c9Nh05q2vHa10yQvf0u+JL7jljWwKSyq54qXvuOzFbwFY4rdcxqzVu1m27QAnP/EFA5+cwcMf1s5tuP71hVxqP2bHgVKe+jTXt+1nsHnr97BwkzWaaWbubrK37Kesspqs8dN4dfbGwPdZVunrB1Eq0mgiUAAM7NqKYUH7IftLiq8/EbRIqH/cgdOks/p8vHQHxhjKKgITSFllNQs37WPNrmJKK6v50+drHLfBK3JYIO/dRdtYtHkfN7+xyPdlD/Dh99sxxtDdr+/CGMP97y/jldkbWWrXEIwxAbWHn762IKSm4e3z+MOnuQHHT3lyJv1/+0W973ne+j288NU6x3N7Dpbz+6k5vr4U1TDGGBZs3IsxzslcWTQRqAZJchhK6q/FYRJFUZnzlT/AcW2SHY+/l53H81+tDzj2q8nfc+Ur33GgxEosu4ucRxOVVITWQB74YDk/eTm0iUhE2FFYRpXflf89/13qa4byfvk+/OEKuj/yWciXiv+kvAq/L+pXZ2/kD9NyWLbtQMDxuvz0tQX85YvAneMOlFSQt7+E30/N4bW5m5iRs4vb3szm+63763iW5rP/UAWFJZFdy/lo6XaumjifD7/ffvjCUUwTgWoQpzkFAL3aW+sZJds1gh+yjFBmaqLj8Y+W1v3H+197oby6EszBOpqcnMQIrM4P3JDno6U7iLEnS3j7K7yzovMLy/hk2Q5f2X5P1F7pl/oloD98msurczZxwz8X1vnaJRVVIXtBlPk1h138wlxOf2aWL0kt3XaAmbm7eGjKCu6atMS3IuzXa3bzxaqdVFbXcMdbi1nSiERRU2P4y+dryC8sbfBjAAY+OYP+v6u/lhNuW/da78l/8EJZZTVvzd8SMLIt2mkiUA1S1/yCj+4awbLHx9AqOT6gXEZq6BpGdclMd04E8zbsdTwOMGuN1VdQV9451IhEUFZZ45sJ7S/WYz27t23fuy7Tyu2F/Gry947PVdcGP173vruU3cW1tZhLXviWQU/OCCi/emcxt725iMLSSrbt867VZJ2baPc9dEhPZNqKfN6av4Xnv1zHTf9axO1vLWZ53gGmr9rZqCXDc/KLeGHWesZNrh21VVNjmDh7Q539GlOX73A8Hmm86yL69/O8OGs9j320kk+OgvfQXE1amghUg/16TA8+uPO0gGOJcR7Sk+MY2NUaQeS9iu7fuWXI4wGeurwfL/x0IBOu7M/g46xF8dqmxNf7uvef15MHxvZ0PFdcXkW5w0ii/SWN65OYuiI/5FiN/bQfLNnOC1+t882F+GR5aFmvpz5bHXLMv9YyZcl2fvTiPLLGT2P97mLW7z4IwBP/W+Ur8/upOczM3c1Hfs0ZO4OawGL8pnZPmFHbnJSTb838bp/mnFw3FhzkZ//OJmv8NM5/bg7GGN8s8W37S9hQYMUza81u/vjpap52eD+Ab/IgwPNfOvdrWO+9kvvfW0ZRWSVP/G8VvR+fXmdZr7nr9pA1fhrrdhXXW27E019x/nNz6i3jXem22u8L1ZusixyS3OSFW32jzvzNyNlF1vhp5O0vCTnnZN2u4kb1i81abdXo/P1nwVa6P/JZoy5qfiidWawa7JcjT6zzXNuUBB6/qDdDu7WmpKKa3h3TKCmvQkQY8oeZvnLXDK2diPafBVuB2vWH6jL8+Nacclxr/jR9zWFj7No6ma37Spi3vu7ahJPZ9mgkf96r4dlrCwLO+zcLBVu85fBNMtvt5Ta+86vxvDFvc8jr+vdZ+NciAN9ifiEx2wlQRCguq8QACzfu46yeGSzPO8AVL9X2keTmF5FfWOarbeQXljHq2W/Y/PSFvi9L/2Yqr+AO6xdnref/Rjn/brzx7WbeW5xHl9bJAe+xPt7Pd9Hm/czI3cXokzLpYS+p7m/7gVLfZxkcnzHWBk0eO8vV1BiKyypJivP4aq3Bo8GqawwPTbFGkG1++sKAcx/bzZTZm/fTuZVznxZgJ1bh3L/OpmN6IvMeGuU7t2anlfj9V/etqTF8kbOLO95eDMDDF/TihlOzSIzz+EazFRSXH3YwxpHSGoFqMrec3o2+ndIZ2q01KQmxtEtLDGgi+ihoAtrYvtY+Cd3tP/JBXVvStXXoH5l36Oqyx8cwpndmvTFM+YVVY/lvttWe7/98PTNTaZUc1+D3kxPUb9DU6hqWus6uJezz2z60uJ7Odn+H7D6Kr1bvot8TX3DyE19w27+zefD95b6F//zl7ChyrFF5h+0GjxZbvGU/izYHLgAY69BsmF9Yyqw1u32jrJze68dLt/PQlOUhk/28V+/lVdX8afoaxvx1tm9QwPyNe7nqle98NRew+mX871/4/Bx6PPoZgN+XvtWXM+6dpb7aVHVQSKt31v5/L9t2IGAgQlqS9XvjNBrNKze/iG4PfcrcddaQ5h1By6Kc97fZ3PWfJeTsKKLno5+Rt7+ESQu3+pIAWAs5/nVm4ICBUodk3NQ0ESjXvX7jYKb+6nQGdAlsLrr19G4se3wMne0VUGsMPH5R74Ay3dq28F0NpifHMeGqAQBcPaRLyF4K1wztSnpS7Rf9kKxWTPnFab5RSc9e2Z85D47kjZuHNPo9DOrq3NTl77w+tUnqntE9Dlv+iU9y6j3/4qwNvtsHGjg6p8RuRgj+3v1qzW7HyXjfrC2gPGiOx+erdvJ3e7TWtn0lAbWhK16ax09fXRBQ3qn/6KLn53Lzv0KAe5YAABNrSURBVBbhlOq87d7j3lnK5IXbeGjKCmbm1DbHeDtxf+v3+XxqN93d9+4yFmzax6hnv/Gde+CD5Yx69htfv8XaXbVJwZuAvI+ftiLflwiCP49se4HDdqkJXPritwz945f85uOVVNcY36ZMxWVV7C4qY+veEt7N3hawaq43AVz3euDnE+zRj1ZQXlXD1OX5bA6aeAmwpziwSclpBFxT06Yh5bpRJzlfxYsI6clxvj9MA4zuncnmpy8ka/w0AP5105CAPZhTEmL5/rFzSUuKY9Oe+ew5WM6ffnwyY/u2Jy0x8Gr/3Z+fioj4mlrapSWQkhDL2T3bBZT7dvxIRjz9lWOMZ/bIYPbaAk7u3JIlWw84LriXFOehtLKassoaYsT6Eu5QRwd4U7ukf0f+59dU9eZ3zkt5HCip5P73l4ccf2v+Ft8Cf14/f6v2CnXOuj3MWbeHc3tn8tK1gxyfu9Lv0npjwUGS4j2+taFy7T6LN7/b7CtTVWOI89Qmj/cW5/He4jxfc0yNQwep9yViPaFJx9uU9Mv/fE+XoGYbb23Hv4/F+/wHy2uT68dLt/ua9Vq3iPf9H7/53RaqjaGdPbLtQEkFQ//4pe9xOTuKeOKSPvZr1f+FHR8bQ0VVjW9y4/6SCsemt+ChxqWaCFQ0SLabH9qnhY40SkkM/RX1LmXdsaVVk+jdIS0gCcy890wKS63+CYBnrjiZ52auo00L55FMnVoG7snw02Fdff0Xw7q1ZvbaAhLjPHz30Ej2HargwufnBpQf0b0tXVsnc/OILEZP+Ibyqhr6dkrn12N6kLe/tM49oZtCpsNn1lhb9oZelQabkbOL4U996XiutLKaOesK+PtX630zrb1m5lpX+v41muV5Bxybo/o8Pp3SymqOzwjdPnXOugJm5Ow87NWx/xfr0m0HHMt7r+JfnLWBympD6xbxAZ3iwY95e/5W/m9kdwA2FgR+VgXF5ew7VEFuftFha22tkuPYVVR7EbHvYEVIzQ2gMuizaY6mIU0EKux6ZKYy4cr+jjWHlHo6yX53aR/O6dWOvp3SA453bxfYsXhen/YB+zaD1QE9f+M+Euzaxqxfn82Ls9bz/uI8kv3mTHibFjwx0CE9KWCpjeR4DyUV1cTHCo9fbDVpxXtiKK+qISnewy9Hnsis1bt/UCLw1iweHNuLjQUHeW9xHgD/uW0YN/xzIVU1hi6tk7ikfydenRO6l8P5fdvz2cqdIcedOHW4OtnjMMTW67GPVrJ5b8NG1Ph3WPvz9m94R1L5+3pNaGe+k11+tbXL/GaPB7xOee0X68SgpUDAeRa8d+RXblC/UaxHuOWNRSzddoBLB3SsM67tB0pDPj/v/2mw6at2stGvz6OuTZuakmt9BCLyTxHZLSKOA5pFJF1EPhGRZSKySkRudisWFfkuH9Q5oH3f26mbEFv3r2hqYhyX9K/7j68+/7xpCF/ccybfP34uYPVFeCfH1RjoZ1/RXz/8OMb0zuSWEd0Aq9nAO1S2jT3sVfyGcsbZ8Xqbs87p1c7Xd9A+LdH3GofjvVK86OQO/PbSPr7jp3Vv6+tMffyiPvTrnM6GP17A/ecFDq996bpTDrv+k9fMXOcRSAA9MkOvzr0m/6x2baqGJoEfwv/34nBenxP6xR5smsNQYX9O61+t2lEIwAGHIadr7WGu+QdCZ7n/fmoO5/zla0Y8/VWdgwOcfLKsNsbSimrmb9zLt+v31POII+NmZ/EbwNh6zt8F5Bhj+gNnA8+KSP0DylXU+Piu0/nHtYMCvmSbUnJ8LD0yU0mOr61x1I4wqeGTX53OL0eeSKsW8Uy8YTBt/Dqm/3XTEK4d1pXrhh0HBI7p946g8fgdO9meUzG6dzv6dAysvfiL94T+OWakJpAcH8ugri05u2cGUDu5zLushydGaOcwga8p1iX6x7Wn1Hlu+PGtj/j5u7ZOpk/HtHrLOCXPcUHDVZ+5oh8AW/aFJqTGJJK6eHfKC242+njpDt+xHQ4zs1+buylgVnND+Y8cKqmo5uqJ87n2tfo7oY+Ea4nAGDMbqG+jWQOkivWXnmKXdb8OpI4KXdskc0G/Docv2IS8X+JVh7lya5kczx9+1I/MtNoORK8x9tW//9pM3mapHpmpAWUBrhvelb9fY+314N9p+9zVAxh+fGvf0h4f3Hkab9w8NOCxyX7NZt69p3tmpvLlfWcBsL8J1gHq3i7FVwMCfCO8wKoJnWMnJwjta2mI124czH1j6h5h9exP+jOsW2DCGdClJfec24MnL+vrO3aC3a/g1E5f1+S6puY0PLcpNHTo8JEI5/DRF4CTgB3ACmCcMcbxEkZEbheRbBHJLihoWFuhUo11XBvri7h7u7qbQ/yddoK1Wqt/R/VvLu7D3AfPId1vvkL3diksemQ01w07LmTJhvvO7cn5fdtz7bCuvHx97dX3pQM68c7tp/ruO9WMsvwW6zvthLa8dsNg3rhliN+XYmhb99VDuvhuXz6oE8t+MybgeZz4L7/RM2hi1z9vGsILPx3I0KzWTLptGIlxjftKadMinrN7tOOpy/sF7J3tdcUpnWkbVNvxfhTX+e2S16lVUp3LoLRNjec3F/d2PBeJgmtAr/k1dzWmeakxwpkIzgOWAh2BAcALIuJYRzTGTDTGDDbGDM7IyHAqotQRO7NHBlN+cRo3nprVoPLt0hJ5745T+cOPaq9M4zwxjjNPM1ITiImRkDbmpHgPsZ4Y/vCjfiHDOA+nZXJgS+ro3pl0SK+9Kv/JKdaXvne+xZrfj+XnZ53gO5+eFEd6UhyfjjuD6XefwfePncuSx84NmA8BgRPGbjm9G5/88nTevnUYYCWoi07uyLt3nEpW2xasfvL8kDi9zTZOWiXHExMjXDO0a8De2b8e08P3Gq1bBL7PbnbC9k+OrZLjfWP9g7VpkcDNI7qx4OHaWb7eUUDNoU/HNCZc2T/k+I9P6exY3jsp0su/zyJ4GYqmEs5EcDMwxVjWA5uAXmGMRykGdW3lW5+mIYZktQ75Qq7PPaN7BHSA19cZXpepv6r9Iq7PZQM7sfnpC/ny3rN4745TSYj10LFlbTOJtyaTHB9Lr/ZptGoRT+sW8bxy/WDA6jAHePXGwTxywUmsfnIsI7q3pV/ndE4/sW2DYh19UiY92wde3yXFeZj6q9P521UDHD/rGLGWM/G+hjcRHJ/RgmuGduHhC08KeUxinMfXLNS5VRLfP3aub3KiN19kpiXSNiWBS/p3ZJzfhD//JR+smAPnmfw6qOlqQJeWvHzdIEb1CiznfW+f/PL0gGOv3TiYM04MvIB9745TQ/o5AC4f2Cmg3+q64YF7gzuNqGoK4Rw+uhUYBcwRkUygJ3D4Ln+ljmIXntyBC0/uwIKNe/lqze4f1BkePFz2cNKT4xiSZbWzJ8R6+MXZJ/CPrzfUu9jfwodH+eZwnJCR4mtuaowxvTN55MKTQtrtR9pDfp3ex7zxI0OSo3fIbov4WJ66/OSAc+3TEkMW5Hv5ulNo1SLemgT3zYaAYcmLHhnl+8xn338O//x2E49eeBLT7IUEHxjbkzNPzGDOuj0M6NKSod1ac2aPjIB9ItqlJjC2bwcGZ7Xm46U7eHJq7Qzo0spq+nVO581bhnLjPxcy+LhWvlrazHvPYvSEb3zPEbx+0Od3n0lPu1nomqFdmbxwKwmxHt9Q4nnjR9KyEUukNIZriUBEJmONBmorInnAb4A4AGPMy8CTwBsisgJrNeEHjTHujY9SKoIMO75NvTvCuenOs63moSv9+guCtWuCDtaJN1g1i6LSwoDjwU09/jo6dDj3ap9Kl9ZJPHxBaE3g87vP5JA9zj6rTTKb95b4RiF1aZ3MokdGB5T3T7xd2yT7ZgU/flFvSiur+cXZVpPRmt/XNnFVVddww6nHMaBLS+59dxnn2DWBtikJ3Hp6N56cmkPblHj2HKwIWX490+9z7N4uhZSEWA6WV5EQ6/FNpPTq6dc3cNoJbZi8cCuV1TVM/tlwpizZTof0RNdG0bmWCIwx1xzm/A5gjFuvr5RylpoYxwNjm68Vtm+nNO4Z3YPLB3Xi7QVbuOucxrXPt0iIZc4DIx3PpSfH+Trm37/zNAqKy3/Ql+Utp3er81ysJ4bfXWr1Aw3s2iqkc33OA+eQmhjLvA176WsPDx5xQhvuOOsEfnZG4PN6E0FMjNWc9c7tw7l64vyQ1zy/b3vGjTqRm0dk0TI53vWLBjna9vIcPHiwyc7ODncYSrli9toC0pPi6N/l8IvcRSrvOlHBSzkra/LZOwu38dhFJ/kS1p1vL2ZAl5YBHfluEJHFxpjBjuc0ESilmtKcdQXsO1TBpQM6hTsU5ae+RKBrDSmlmlTwCBkV+XQ/AqWUinKaCJRSKsppIlBKqSiniUAppaKcJgKllIpymgiUUirKaSJQSqkop4lAKaWi3FE3s1hECoAtP/DhbYFIXdguUmPTuBpH42ocjatxjiSu44wxjrP9jrpEcCREJLuuKdbhFqmxaVyNo3E1jsbVOG7FpU1DSikV5TQRKKVUlIu2RDAx3AHUI1Jj07gaR+NqHI2rcVyJK6r6CJRSSoWKthqBUkqpIJoIlFIqykVNIhCRsSKyRkTWi8j4Zn7tf4rIbhFZ6XestYjMEJF19r+t/M49ZMe5RkTOczGuLiIyS0RyRWSViIyLhNhEJFFEForIMjuu30ZCXH6v5RGR70VkaqTEJSKbRWSFiCwVkewIiquliLwvIqvt37NTwx2XiPS0PyfvT5GI3B3uuOzXucf+nV8pIpPtvwX34zLGHPM/gAfYABwPxAPLgN7N+PpnAoOAlX7H/gSMt2+PB56xb/e240sAutlxe1yKqwMwyL6dCqy1Xz+ssQECpNi344AFwPBwx+UX373Af4CpEfR/uRloG3QsEuJ6E7jNvh0PtIyEuPzi8wA7gePCHRfQCdgEJNn33wVuao64XPuAI+kHOBX43O/+Q8BDzRxDFoGJYA3Qwb7dAVjjFBvwOXBqM8X4MXBuJMUGJANLgGGREBfQGfgSGEltIoiEuDYTmgjCGheQZn+xSSTFFRTLGODbSIgLKxFsA1pjbSM81Y7P9biipWnI+wF75dnHwinTGJMPYP/bzj4ellhFJAsYiHX1HfbY7OaXpcBuYIYxJiLiAv4GPADU+B2LhLgM8IWILBaR2yMkruOBAuBfdlPaayLSIgLi8nc1MNm+Hda4jDHbgb8AW4F8oNAY80VzxBUtiUAcjkXquNlmj1VEUoAPgLuNMUX1FXU45kpsxphqY8wArCvwoSLSN9xxichFwG5jzOKGPsThmFv/lyOMMYOA84G7ROTMeso2V1yxWE2iLxljBgKHsJo2wh2X9WIi8cAlwHuHK+pwzI3fr1bApVjNPB2BFiJyXXPEFS2JIA/o4ne/M7AjTLF47RKRDgD2v7vt480aq4jEYSWBScaYKZEUG4Ax5gDwNTA2AuIaAVwiIpuBd4CRIvJ2BMSFMWaH/e9u4ENgaATElQfk2bU5gPexEkO44/I6H1hijNll3w93XKOBTcaYAmNMJTAFOK054oqWRLAIOFFEutlXAVcD/wtzTP8DbrRv34jVPu89frWIJIhIN+BEYKEbAYiIAK8DucaYCZESm4hkiEhL+3YS1h/I6nDHZYx5yBjT2RiThfU79JUx5rpwxyUiLUQk1Xsbq115ZbjjMsbsBLaJSE/70CggJ9xx+bmG2mYh7+uHM66twHARSbb/NkcBuc0Sl5sdMZH0A1yANSpmA/BIM7/2ZKw2v0qsLH4r0Aar03Gd/W9rv/KP2HGuAc53Ma7TsaqSy4Gl9s8F4Y4NOBn43o5rJfC4fTzsn5nf651NbWdxuD+v47FGjywDVnl/v8Mdl/06A4Bs+//yI6BVhMSVDOwF0v2ORUJcv8W66FkJvIU1Isj1uHSJCaWUinLR0jSklFKqDpoIlFIqymkiUEqpKKeJQCmlopwmAqWUinKaCJRymYicLfZKpUpFIk0ESikV5TQRKGUTkevE2gdhqYi8Yi98d1BEnhWRJSLypYhk2GUHiMh8EVkuIh9614gXke4iMlOsvRSWiMgJ9tOnSO26/JPsmaOIyNMikmM/z1/C9NZVlNNEoBQgIicBV2Et3jYAqAauBVpgrUczCPgG+I39kH8DDxpjTgZW+B2fBLxojOmPtU5Mvn18IHA31hryxwMjRKQ18COgj/08v3f3XSrlTBOBUpZRwCnAInv561FYX9g1wH/tMm8Dp4tIOtDSGPONffxN4Ex7vZ9OxpgPAYwxZcaYErvMQmNMnjGmBmspjyygCCgDXhORywFvWaWalSYCpSwCvGmMGWD/9DTGPOFQrr41WZyWBfYq97tdDcQaY6qwVgn9ALgMmN7ImJVqEpoIlLJ8CfxYRNqBb7/f47D+Rn5sl/kpMNcYUwjsF5Ez7OPXA98Yay+HPBG5zH6OBBFJrusF7X0g0o0xn2I1Gw1w440pdTix4Q5AqUhgjMkRkUexdvmKwVop9i6szVT6iMhioBCrHwGs5YBftr/oNwI328evB14Rkd/Zz/GTel42FfhYRBKxahP3NPHbUqpBdPVRpeohIgeNMSnhjkMpN2nTkFJKRTmtESilVJTTGoFSSkU5TQRKKRXlNBEopVSU00SglFJRThOBUkpFuf8HoKspfpxBQjIAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch: 999/1000,Total Loss : 1.78 ,Class Similarity Loss : 0.16, Cross Modal Loss: 0.16\n"
     ]
    },
    {
     "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": [
    "n_epochs = 1000\n",
    "batch_size = 128\n",
    "loss_arr = []\n",
    "loss_epoch_arr = []\n",
    "X = image_embeddings\n",
    "\n",
    "for epoch in range(n_epochs):\n",
    "\n",
    "    # X is a torch Variable\n",
    "    permutation = torch.randperm(X.size()[0])\n",
    "\n",
    "    for i in range(0,X.size()[0], batch_size):\n",
    "        \n",
    "\n",
    "        indices = permutation[i:i+batch_size]\n",
    "        \n",
    "        image_embeddings, text_embeddings,one_hot_labels = image_embeddings[indices],text_embeddings[indices],one_hot_labels[indices]\n",
    "        \n",
    "        #move inputs to gpu\n",
    "        \n",
    "        #inputs, labels = inputs.to(device), labels.to(device)\n",
    "\n",
    "        # zero the parameter gradients\n",
    "        optimizer.zero_grad()\n",
    "\n",
    "        phi_I = net.image_tower(image_embeddings)\n",
    "\n",
    "        phi_T = net.text_tower(text_embeddings.float())\n",
    "\n",
    "\n",
    "        phi = net.shared_hidden_layer(torch.cat((phi_I,phi_T),dim = 1))\n",
    "\n",
    "\n",
    "        z = net.shared_output_layer(phi)\n",
    "\n",
    "        outputs = net.softmax(z)\n",
    "\n",
    "        loss_class_similarity = loss_fn_1(outputs,one_hot_labels.float())\n",
    "\n",
    "        #loss_semantic_similarity = net.semantic_similarity(z,A)\n",
    "\n",
    "        loss_cross_modal_gap = torch.sum(1 - F.cosine_similarity(phi_I,phi_T))/phi_I.size(0)\n",
    "\n",
    "        loss = (alpha*loss_class_similarity + gamma*loss_cross_modal_gap )\n",
    "\n",
    "        loss.backward()\n",
    "\n",
    "        optimizer.step()\n",
    "        \n",
    "        loss_arr.append(loss.item())\n",
    "\n",
    "    loss_epoch_arr.append(loss.item())\n",
    "\n",
    "    \n",
    "    if (epoch+1)%200 == 0:\n",
    "        print('Epoch: %d/%d,Total Loss : %0.2f ,Class Similarity Loss : %0.2f, Cross Modal Loss: %0.2f' % (epoch,n_epochs ,loss,loss_class_similarity,loss_cross_modal_gap))\n",
    "\n",
    "        plt.plot(loss_epoch_arr)\n",
    "        plt.xlabel(\"epochs\")\n",
    "        plt.ylabel(\"loss\")\n",
    "        plt.title(\"\")\n",
    "        plt.show();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Only taking ``50 points`` to calculate ``semantic similarity``"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "``Part 3(b) :`` Semantic Similarity\n",
    "---\n",
    "* ``Showing example`` to compute ``semantic similarity`` on ``50 points``\n",
    "\n",
    "\n",
    "### ``Note :``\n",
    "\n",
    "We are not taking ``semantic loss`` in ``training`` because it takes ``lot of time`` to compute.For ``1 batch(128 points)`` it takes ``1 hr``."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {},
   "outputs": [],
   "source": [
    "out_I = net.image_tower(image_embeddings[:50,:])        \n",
    "out_T = net.text_tower(text_embeddings[:50,:].float())    \n",
    "out = net.shared_hidden_layer(torch.cat((out_I,out_T),dim = 1))\n",
    "z = net.shared_output_layer(out)\n",
    "outputs = net.softmax(z)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {},
   "outputs": [],
   "source": [
    "b = list(product(range(z.size(0)),range(z.size(1)),range(z.size(0)),range(z.size(1))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {},
   "outputs": [],
   "source": [
    "def semantic_similarity(z,A):\n",
    "    # This Code calculates \n",
    "    \n",
    "    phi = z.detach().numpy()\n",
    "    \n",
    "    D = np.zeros((50,67,50,67))\n",
    "    R = np.zeros((50,67,50,67))\n",
    "    sigma = np.zeros((50,67,50,67))\n",
    "    threshold = 2.0\n",
    "    \n",
    "    for m,i,n,j in b:\n",
    "        D[m][i][n][j] = cosine(phi[m][i],phi[n][j])\n",
    "        R[m][i][n][j] = np.power((D[m][i][n][j] - A[i][j]),2)\n",
    "        sigma[m][i][n][j] = 1.0 if ((A[i][j]< threshold) and (D[m][i][n][j] < threshold)) else 0.0\n",
    "    \n",
    "    return (np.sum(sigma*R))/(50**2)\n",
    "\n",
    "loss = semantic_similarity(z,A)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 97,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "4.34\n"
     ]
    }
   ],
   "source": [
    "print(loss)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Showing ``accuracy`` on 1 batch i.e 128 points"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "out_I = net.image_tower(image_embeddings)        \n",
    "out_T = net.text_tower(text_embeddings.float())    \n",
    "out = net.shared_hidden_layer(torch.cat((out_I,out_T),dim = 1))\n",
    "z = net.shared_output_layer(out)\n",
    "outputs = net.softmax(z)\n",
    "truth_labels = one_hot_labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "def accuracy(predicted_values,truth_values):\n",
    "    count = 0\n",
    "    for x,y in zip(truth_values,predicted_values):\n",
    "        x = torch.argsort(x)[-3:]\n",
    "        x = torch.sort(x).values\n",
    "        y = torch.argsort(y)[-3:]\n",
    "        y = torch.sort(y).values\n",
    "        \n",
    "        if torch.sum(x == y) == 3:\n",
    "            count +=3\n",
    "        elif torch.sum(x == y) == 2:\n",
    "            count +=2\n",
    "        elif torch.sum(x == y) == 1:\n",
    "            count +=1\n",
    "        \n",
    "    acc = (count)/(truth_values.size(0)*3)\n",
    "    return acc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.5442708333333334"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy(outputs,truth_labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "``Conclusion:``\n",
    "-----\n",
    "* Total loss is reduce from 3.5 to 1.69\n",
    "* Cross modal gap is 0.15.\n",
    "* Class level similarity loss is 0.16.\n",
    "* Accuracy is 62.5"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.7.3 64-bit ('keras-gpu': conda)",
   "language": "python",
   "name": "python37364bitkerasgpucondac15bbfaa9e7342f1a6aaed5bedd28ed1"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}