Omniglot.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "language_info": {
      "codemirror_mode": {
        "name": "ipython",
        "version": 3
      },
      "file_extension": ".py",
      "mimetype": "text/x-python",
      "name": "python",
      "nbconvert_exporter": "python",
      "pygments_lexer": "ipython3",
      "version": "3.6.9"
    },
    "colab": {
      "name": "Omniglot.ipynb",
      "provenance": [],
      "toc_visible": true,
      "include_colab_link": true
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/sidm-23/One-Shot-Learning-on-Omniglot-data/blob/master/Omniglot.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "-gcI6z3TXuE9",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Importing all the required  modules\n",
        "%tensorflow_version 1.x\n",
        "import cv2\n",
        "import numpy as np\n",
        "import os\n",
        "from matplotlib import pyplot as plt\n",
        "import pandas as pd\n",
        "from sklearn.utils import shuffle\n",
        "import tensorflow as tf\n",
        "from keras.models import Sequential\n",
        "from keras.optimizers import Adam\n",
        "from keras.layers import Conv2D, ZeroPadding2D, Activation, Input, concatenate\n",
        "from keras.models import Model\n",
        "from keras.layers.normalization import BatchNormalization\n",
        "from keras.layers.pooling import MaxPooling2D, AveragePooling2D\n",
        "from keras.layers.merge import Concatenate\n",
        "from keras.layers.core import Lambda, Flatten, Dense\n",
        "from keras.initializers import glorot_uniform\n",
        "from keras.regularizers import l2\n",
        "from keras import backend as K"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "t2_SnaQWWJsD",
        "colab_type": "text"
      },
      "source": [
        "The [Omniglot dataset](https://github.com/brendenlake/omniglot) is a collection of 1623 hand-drawn characters from 50 alphabets. For every character there are just 20 examples, each drawn by a different person at resolution 105x105. Each image is paired with stroke data, a sequence of [x,y,t] coordinates with time (t) in milliseconds. This data is split as 30 alphabets used for training and the other 20 used for validation of the model.\n"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "r6H66ZoAgxqY",
        "colab_type": "text"
      },
      "source": [
        "**Why use One-Shot Learning**     \n",
        "In conventional image processing, an image is put through a CNN(Convolution Neural Network) to extract features from which an object/edge is detected or classified, but this is very computationally intense and requires a very large training set to alleviate all bias. Another issue is that when a new set of training data is added the model has to be re-trained.\n",
        "So one-shot learning proposes to extract features from a small set of training data and make a faster prediction based on similarity.  \n",
        "*Ex* \n",
        "Face unlock in phones take only multiple images of and feature for your face and store it on device but when a new face is added it does not have re-train the entire model, it extracts the features from each image and compares for similarity thus reducing the time it takes to unlock while still being reliable."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "dTH6-YWP6zh9",
        "colab_type": "text"
      },
      "source": [
        "**Data Preprocessing**"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hFwMKcX77iB8",
        "colab_type": "text"
      },
      "source": [
        "**Raw data** \n",
        ">image_background  \n",
        ">>Gujarati\n",
        ">>>Character01\n",
        ">>>>20 images\n",
        "\n",
        "![Raw Dataset](https://raw.githubusercontent.com/brendenlake/omniglot/master/omniglot_grid.jpg)\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "v6YLDJJfXuFC",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def get_image_data(folder):\n",
        "\n",
        "    \"\"\"\n",
        "    Get Training and evaluation data for images\n",
        "    It returns the data in a DataFrame with multiple indexes\n",
        "    \"\"\"\n",
        "    \n",
        "    lang_list=[]#list of of languages\n",
        "    X={}\n",
        "    Y={}\n",
        "    for languages in os.listdir(folder):\n",
        "        if languages.startswith('.'):\n",
        "            continue#ignore all hidden files\n",
        "        else:\n",
        "            images={}\n",
        "            lang_list.append(languages)#for each language\n",
        "            chars=[]\n",
        "            for alphabets in os.listdir(folder+languages):\n",
        "                im=[]\n",
        "                chars.append(alphabets)#for each character\n",
        "                for character in os.listdir(folder+languages+\"/\"+alphabets):\n",
        "                    if(character.startswith('.')):\n",
        "                        continue\n",
        "                    else:\n",
        "                        im.append(cv2.imread(folder+languages+\"/\"+alphabets+\"/\"+character))#append image data\n",
        "                images[alphabets]=im#for each character\n",
        "                Y[languages]=chars#for each language type\n",
        "#             print(languages)\n",
        "            X[languages]=images\n",
        "            \n",
        "    #After getting all data return the data in Serires and a Dataframe to easily access the data\n",
        "    y=pd.Series(Y)\n",
        "    x=pd.DataFrame.from_dict({(i,j): X[i][j] \n",
        "                               for i in X.keys() \n",
        "                               for j in X[i].keys()},orient='index')\n",
        "    \n",
        "    return x,y,lang_list"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "fdTSXhajXuFF",
        "colab_type": "code",
        "outputId": "6722c7fe-341c-440b-e147-376082db8ec2",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        }
      },
      "source": [
        "#testing module\n",
        "print(\"Getting Trainind data\")\n",
        "x_train,y_train,languages=get_image_data('Fellowship/images_background/')\n",
        "print('Getting Evaluation data')\n",
        "x_val,y_val,languages_val=get_image_data('Fellowship/images_evaluation/')"
      ],
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Getting Trainind data\n",
            "Getting Evaluation data\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "V-TD9ZZr-_O-",
        "colab_type": "text"
      },
      "source": [
        "**After reading data**  \n",
        "The data is now stored a multi-indexed DataFrame, with each cell storing the image array to the corresponding language and character. This makes accessing the image data very easy and fast as pandas DataFrame structure is very well optimized."
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "w3JcAVoT-gpW",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "2978a546-22ce-463c-edb3-40c4d97fe281"
      },
      "source": [
        "x_val"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "execute_result",
          "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>0</th>\n",
              "      <th>1</th>\n",
              "      <th>2</th>\n",
              "      <th>3</th>\n",
              "      <th>4</th>\n",
              "      <th>5</th>\n",
              "      <th>6</th>\n",
              "      <th>7</th>\n",
              "      <th>8</th>\n",
              "      <th>9</th>\n",
              "      <th>10</th>\n",
              "      <th>11</th>\n",
              "      <th>12</th>\n",
              "      <th>13</th>\n",
              "      <th>14</th>\n",
              "      <th>15</th>\n",
              "      <th>16</th>\n",
              "      <th>17</th>\n",
              "      <th>18</th>\n",
              "      <th>19</th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th>(Kannada, character38)</th>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>(Kannada, character35)</th>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>(Kannada, character16)</th>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>(Kannada, character03)</th>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>(Kannada, character22)</th>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>...</th>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "      <td>...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>(Oriya, character20)</th>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>(Oriya, character06)</th>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>(Oriya, character17)</th>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>(Oriya, character02)</th>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>(Oriya, character13)</th>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "      <td>[[[255, 255, 255], [255, 255, 255], [255, 255,...</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "<p>659 rows × 20 columns</p>\n",
              "</div>"
            ],
            "text/plain": [
              "                                                                        0  ...                                                 19\n",
              "(Kannada, character38)  [[[255, 255, 255], [255, 255, 255], [255, 255,...  ...  [[[255, 255, 255], [255, 255, 255], [255, 255,...\n",
              "(Kannada, character35)  [[[255, 255, 255], [255, 255, 255], [255, 255,...  ...  [[[255, 255, 255], [255, 255, 255], [255, 255,...\n",
              "(Kannada, character16)  [[[255, 255, 255], [255, 255, 255], [255, 255,...  ...  [[[255, 255, 255], [255, 255, 255], [255, 255,...\n",
              "(Kannada, character03)  [[[255, 255, 255], [255, 255, 255], [255, 255,...  ...  [[[255, 255, 255], [255, 255, 255], [255, 255,...\n",
              "(Kannada, character22)  [[[255, 255, 255], [255, 255, 255], [255, 255,...  ...  [[[255, 255, 255], [255, 255, 255], [255, 255,...\n",
              "...                                                                   ...  ...                                                ...\n",
              "(Oriya, character20)    [[[255, 255, 255], [255, 255, 255], [255, 255,...  ...  [[[255, 255, 255], [255, 255, 255], [255, 255,...\n",
              "(Oriya, character06)    [[[255, 255, 255], [255, 255, 255], [255, 255,...  ...  [[[255, 255, 255], [255, 255, 255], [255, 255,...\n",
              "(Oriya, character17)    [[[255, 255, 255], [255, 255, 255], [255, 255,...  ...  [[[255, 255, 255], [255, 255, 255], [255, 255,...\n",
              "(Oriya, character02)    [[[255, 255, 255], [255, 255, 255], [255, 255,...  ...  [[[255, 255, 255], [255, 255, 255], [255, 255,...\n",
              "(Oriya, character13)    [[[255, 255, 255], [255, 255, 255], [255, 255,...  ...  [[[255, 255, 255], [255, 255, 255], [255, 255,...\n",
              "\n",
              "[659 rows x 20 columns]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 7
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "6r_S4cxTe4-x",
        "colab_type": "text"
      },
      "source": [
        "**Siamese Network**  \n",
        "This is the architecture we use for our one-shot classification. It takes an input of two sets of images and finds the similarity of each pair and finds the twins.\n",
        "This is done by passing it through 4 layer CNN which return 4096 features for each image, we then find the difference between features of each image and use activation function to signal the similarity if the images.\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jdt7CWTYXuFI",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def siamese_net():\n",
        "    \n",
        "    \"\"\"\n",
        "    Define the architecture of the model which is convolution neural network with 64 10x10 filters,\n",
        "    relu -> max pool -> convolution with 128 7x7 filters, relu -> max pool -> convolution with 128 4x4 filters,\n",
        "    relu -> max pool -> convolution with 256 4x4 filters.\n",
        "    \"\"\"\n",
        "    \n",
        "    def initialize_weights(shape, dtype=None ,name=None):\n",
        "        \"\"\"\n",
        "        The paper, http://www.cs.utoronto.ca/~gkoch/files/msc-thesis.pdf\n",
        "        suggests to initialize CNN layer weights with mean as 0.0 and standard deviation of 0.01\n",
        "        \"\"\"\n",
        "        return np.random.normal(loc = 0.0, scale = 1e-2, size = shape)\n",
        "\n",
        "    def initialize_bias(shape, dtype=None, name=None):\n",
        "        \"\"\"\n",
        "        The paper, http://www.cs.utoronto.ca/~gkoch/files/msc-thesis.pdf\n",
        "        suggests to initialize CNN layer bias with mean as 0.5 and standard deviation of 0.01\n",
        "        \"\"\"\n",
        "        return np.random.normal(loc = 0.5, scale = 1e-2, size = shape)\n",
        "\n",
        "    #Define Imput shape \n",
        "    input_shape=(105,105,3)\n",
        "    left_input=Input(input_shape)\n",
        "    right_input=Input(input_shape)\n",
        "\n",
        "    \"\"\"\n",
        "    Creating a 4 layer model \n",
        "    \"\"\"\n",
        "    model=Sequential()\n",
        "    model.add(Conv2D(filters=64,kernel_size=(10,10),activation='relu',\n",
        "                 input_shape=input_shape,kernel_initializer=initialize_weights,kernel_regularizer=l2(2e-4)))\n",
        "    model.add(MaxPooling2D())\n",
        "\n",
        "    model.add(Conv2D(128,(7,7),activation='relu',kernel_initializer=initialize_weights,\n",
        "                 bias_initializer=initialize_bias,kernel_regularizer=l2(2e-4)))\n",
        "    model.add(MaxPooling2D())\n",
        "\n",
        "    model.add(Conv2D(128,(4,4),activation='relu',kernel_initializer=initialize_weights,\n",
        "                 bias_initializer=initialize_bias,kernel_regularizer=l2(2e-4)))\n",
        "    model.add(MaxPooling2D())\n",
        "\n",
        "    model.add(Conv2D(256,(4,4),activation='relu',kernel_initializer=initialize_weights,\n",
        "                 bias_initializer=initialize_bias,kernel_regularizer=l2(2e-4)))\n",
        "    model.add(Flatten())\n",
        "\n",
        "    model.add(Dense(4096,activation='sigmoid',kernel_regularizer=l2(1e-3),kernel_initializer=initialize_weights,\n",
        "         bias_initializer=initialize_bias))\n",
        "\n",
        "    encode_l=model(left_input)\n",
        "    encode_r=model(right_input)\n",
        "\n",
        "    l1_layer=Lambda(lambda tensors:K.abs(tensors[0]-tensors[1]))\n",
        "    l1_distance=l1_layer([encode_l,encode_r])\n",
        "\n",
        "    # Add a dense layer with a sigmoid unit to generate the similarity score\n",
        "    prediction = Dense(1,activation='sigmoid',bias_initializer=initialize_bias)(l1_distance)\n",
        "    \n",
        "    # Connect the inputs with the outputs\n",
        "    siamese_net = Model(inputs=[left_input,right_input],outputs=prediction)\n",
        "\n",
        "    print(siamese_net.summary())\n",
        "    return siamese_net"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "S-RJVM_RXuFL",
        "colab_type": "code",
        "outputId": "48fc01a6-c1a9-4320-9a9a-f238520c2e33",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 374
        }
      },
      "source": [
        "#Create and complie model\n",
        "model=siamese_net()\n",
        "optimizer = Adam(lr = 0.00006)\n",
        "model.compile(loss=\"binary_crossentropy\",optimizer=optimizer)"
      ],
      "execution_count": 19,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"model_2\"\n",
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "input_3 (InputLayer)            (None, 105, 105, 3)  0                                            \n",
            "__________________________________________________________________________________________________\n",
            "input_4 (InputLayer)            (None, 105, 105, 3)  0                                            \n",
            "__________________________________________________________________________________________________\n",
            "sequential_2 (Sequential)       (None, 4096)         38960448    input_3[0][0]                    \n",
            "                                                                 input_4[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "lambda_2 (Lambda)               (None, 4096)         0           sequential_2[1][0]               \n",
            "                                                                 sequential_2[2][0]               \n",
            "__________________________________________________________________________________________________\n",
            "dense_4 (Dense)                 (None, 1)            4097        lambda_2[0][0]                   \n",
            "==================================================================================================\n",
            "Total params: 38,964,545\n",
            "Trainable params: 38,964,545\n",
            "Non-trainable params: 0\n",
            "__________________________________________________________________________________________________\n",
            "None\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "p0SUPKuAFZxZ",
        "colab_type": "text"
      },
      "source": [
        "**Training**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "dyTld0fZ8ugr",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Hyper parameters\n",
        "evaluate_every =200 # interval for evaluating on one-shot tasks\n",
        "batch_size = 32 #batch size to train data on\n",
        "n_iter = 2000 # No. of training iterations"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HWDGIV1kXuFX",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def get_batch(x):\n",
        "  \"\"\"\n",
        "  Create batch to train the model\n",
        "  We create 32 pair of images the first 16 are images with same class and for next\n",
        "  half we ensure that that the pairs are not from the same category.\n",
        "  \"\"\"\n",
        "  n_classes, n_examples= x.shape\n",
        "  w, h =105,105\n",
        "  categories=np.random.choice(n_classes,size=(batch_size,),replace=False)\n",
        "\n",
        "  # initialize 2 empty arrays for the input image batch\n",
        "  pairs=[np.zeros((batch_size, h, w,3)) for i in range(2)]\n",
        "    \n",
        "  # initialize vector for the targets\n",
        "  targets=np.zeros((batch_size,))\n",
        "\n",
        "  targets[batch_size//2:]=1\n",
        "  for i in range(batch_size):\n",
        "      category=categories[i]\n",
        "      idx_1=np.random.randint(0,n_examples)\n",
        "      pairs[0][i,:,:,:] = x.iat[category, idx_1]\n",
        "      idx_2 = np.random.randint(0, n_examples)\n",
        "      if i >= batch_size // 2:\n",
        "              category_2 = category  \n",
        "      else: \n",
        "          # add a random number to the category modulo n classes to ensure 2nd image has a different category\n",
        "          category_2 = (category + np.random.randint(1,n_classes)) % n_classes\n",
        "      pairs[1][i,:,:,:] = x.iat[category_2,idx_2]\n",
        "    \n",
        "  return pairs,targets\n",
        "    "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ErZrC8QfXuFZ",
        "colab_type": "code",
        "outputId": "89c8e850-1782-4924-d29d-47c89559b45c",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 714
        }
      },
      "source": [
        "print(\"Starting training process!\")\n",
        "l=[]\n",
        "for i in range(1, n_iter+1):\n",
        "    inputs,targets = get_batch(x_train)\n",
        "    loss = model.train_on_batch(inputs,targets)\n",
        "    l.append(loss)\n",
        "    if i % evaluate_every == 0:\n",
        "        print(\"\\n ------------- \\n\")\n",
        "        print(\"Train Loss: {0}\".format(loss))"
      ],
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Starting training process!\n",
            "\n",
            " ------------- \n",
            "\n",
            "Train Loss: 2.3495733737945557\n",
            "\n",
            " ------------- \n",
            "\n",
            "Train Loss: 1.7542649507522583\n",
            "\n",
            " ------------- \n",
            "\n",
            "Train Loss: 1.4156267642974854\n",
            "\n",
            " ------------- \n",
            "\n",
            "Train Loss: 1.1413341760635376\n",
            "\n",
            " ------------- \n",
            "\n",
            "Train Loss: 1.0137253999710083\n",
            "\n",
            " ------------- \n",
            "\n",
            "Train Loss: 1.0075879096984863\n",
            "\n",
            " ------------- \n",
            "\n",
            "Train Loss: 0.8550300002098083\n",
            "\n",
            " ------------- \n",
            "\n",
            "Train Loss: 0.908806324005127\n",
            "\n",
            " ------------- \n",
            "\n",
            "Train Loss: 0.6087885499000549\n",
            "\n",
            " ------------- \n",
            "\n",
            "Train Loss: 0.7647333145141602\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sfVLeTqAsuKL",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 312
        },
        "outputId": "f2a2329a-20cd-4249-b84b-ad9774f49ee7"
      },
      "source": [
        "plt.title('Training Loss')\n",
        "plt.xlabel('Epochs')\n",
        "plt.plot(l)"
      ],
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "[<matplotlib.lines.Line2D at 0x7f9a8086a588>]"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 23
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAEWCAYAAAB2X2wCAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nMVHCEroxxkQJS+jGGBMlLKEbY0yU+H9oNV/ktmkJkgAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "cfLBnMZ-gguG",
        "colab_type": "text"
      },
      "source": [
        "As seen loss has decreased for each epoch, now we can move on to test our model."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "e1UJRanNnUj2",
        "colab_type": "text"
      },
      "source": [
        "**Testing**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ou5_tIjMXuFc",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def make_test_batch(N):\n",
        "    \"\"\"\n",
        "    Create test batch\n",
        "    We create a pair of an image set first part containing the test_image(the image for which similarity is checked)\n",
        "    second part contains set of N images. We use shuffle so that the first image in support-set is not always from the same classs\n",
        "    \"\"\"\n",
        "    n_classes, n_examples= x_val.shape\n",
        "    w, h =105,105\n",
        "    categories=languages_val\n",
        "\n",
        "\n",
        "    indices=m=np.random.randint(0,n_examples,size=(N,))#choose random images to be selected\n",
        "    categories=np.random.randint(n_classes,size=(N,))\n",
        "    test_image=np.zeros((1,w,h,3))#placehoder for test image\n",
        "\n",
        "    targets=np.zeros((N,))\n",
        "    support_set=[]\n",
        "    targets[0] = 1\n",
        "    true_category = categories[0]\n",
        "    ex1, ex2 = np.random.choice(n_examples,replace=False,size=(2,))\n",
        "    test_image[0] = np.asarray(x_val.iat[true_category,ex1])\n",
        "    for c,i in zip(categories,indices):\n",
        "        support_set.append(x_val.iat[c,i])\n",
        "    support_set[0] = x_val.iat[true_category,ex2]\n",
        "    support_set=np.array(support_set)\n",
        "    targets,support_set = shuffle(targets,support_set)#shuffle the images \n",
        "    pairs = [test_image,support_set]\n",
        "    return pairs,targets"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JNL3axTTXuFe",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def testing(k,N):\n",
        "  \"\"\"\n",
        "  Used to test model for k times on N-batch size data\n",
        "  \"\"\"\n",
        "  n_correct=0\n",
        "  for i in range(k):\n",
        "      inputs_t, targets_t = make_test_batch(N)\n",
        "      probs = model.predict_on_batch(inputs_t)\n",
        "      if np.argmax(probs) == np.argmax(targets_t):\n",
        "          n_correct+=1\n",
        "  ImageLoader(inputs_t)#showcase last batch images for example\n",
        "  percent_correct = (100.0 * n_correct / k)\n",
        "  print(\"Got an average of \"+str(percent_correct)+\"% for \"+str(N)+\" way one-shot learning accuracy \\n\")\n",
        "  # return percent_correct\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eiPCNzDKkSSs",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def ImageLoader(imlist):\n",
        "  \"\"\"\n",
        "  The images are loaded and displayed showcasing the test-image \n",
        "  and the support-set of images to which it is compared to.\n",
        "  \"\"\"\n",
        "  plt.subplots()\n",
        "  plt.title(\"Test Image\")\n",
        "  plt.imshow(imlist[0][0])\n",
        "  plt.subplots()\n",
        "  img=concat_images(imlist[1])\n",
        "  plt.title(\"Support-set image\")\n",
        "  plt.imshow(img)\n",
        "  plt.show()"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "G9ddtS4aFuYN",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def concat_images(X):\n",
        "    \"\"\"\n",
        "    Concatenates a bunch of images into a big matrix since mutiple images cant be plotted together.\n",
        "    \"\"\"\n",
        "    nc, h , w, _ = X.shape\n",
        "    n = np.ceil(np.sqrt(nc)).astype(\"int8\")\n",
        "    img = np.zeros((n*w,n*h,3))\n",
        "    x = 0\n",
        "    y = 0\n",
        "    for example in range(nc):\n",
        "        img[x*w:(x+1)*w,y*h:(y+1)*h] = X[example]\n",
        "        y += 1\n",
        "        if y >= n:\n",
        "            y = 0\n",
        "            x += 1\n",
        "    return img\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4WHMoPfhznx4",
        "colab_type": "code",
        "outputId": "e06f7f22-e721-4452-9b8e-9b5a2024b675",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "\"\"\"\n",
        "The main cell the testing fuction is called for diifrent N-way One-shot classification \n",
        "\"\"\"\n",
        "N_way=[9,16,25]\n",
        "for i in N_way:\n",
        "  testing(20,i)"
      ],
      "execution_count": 30,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).\n",
            "Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAQEAAAEICAYAAABf40E1AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAQEAAAEICAYAAABf40E1AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n1ZwnAedqzpOAczXnScC5mvt/WuKoBvT8axkAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Got an average of 95.0% for 9 way one-shot learning accuracy \n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).\n",
            "Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAQEAAAEICAYAAABf40E1AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAQEAAAEICAYAAABf40E1AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjEsIGh0\ndHA6Ly9tYXRwbG90bGliLm9yZy8QZhcZAAAb3UlEQVR4nO2de/AeVXnHP1+5I0KAYAwGCGocGlQi\nEyGobRG8kKgNzqgD1RIQB7XY6khboF6KjkzVQYOMlnqBEryAiDpkGKwiFztouSSAGIhIuJVLIAgk\ngAiS8PSPPW+yefNe9n13992zu89nZue3e3Z/+z5n95zvPufs2fPIzHAcp728oGoDHMepFhcBx2k5\nLgKO03JcBByn5bgIOE7LcRFwnJbjIuBUjqSfSlpUtR1txUWgRCS9UdKvJa2T9JikX0l6XdV2dSPp\nUEn3l3TuYyVdM+gYM5tvZkvK+H1nOFtXbUBTkbQzcCnwEeAiYFvgL4Fnq7SrG0leBtqOmflSwgLM\nBdYO2H8a8N3U9kzAgK3D9tXAvwPXA08AlwC7dR17AvAgsBr4p9S5tgPODPseDOvbhX2HAvcDJwMP\nAT8E/gQ8DzwVlj172Ls98F3gUWAtcAMwLezbBTgn2PEA8HlgK+AvgGeADeG8Pa9HyOsHw/qxwK+A\nxeF37gJeH9LvA9YAi1L/+3bgpnCN7gNO6zr3McC9we5PA/cAbw77XgCcAtwZ9l/UucZtWrw5UB6/\nBzZIWiJpvqRdxzjHMcAHgOnAeuCsrv1vAmYBbwVOlvTmkP5JYB4wBzgAOAj4VOr/XgLsBuwTfmM+\n8KCZ7RSWB3vYsoiksu8F7A58mEQ8AM4L9r0CeG2w54NmtjIc97/hvFMy5vtg4JbwO98HLgReF87/\nfuBrknYKx/4x5GEKiSB8RNKRAJJmA/8BvI/kGu4CvDT1O/8AHAn8NbAn8Djw9Yw2NoeqVajJC8mT\n8DySJ+96YCmbnp6nMdwT+EJq/2zgzyRP2M6x+6X2fwk4J6zfCSxI7XsbcE9YPzScZ/vU/kOB+4fk\n5QPAr4HXdKVPI2ni7JBKOxq4KqwfC1wz5NxXs7kncEdq36tDXqel0h4F5vQ515nA4rD+GeCC1L4d\nQ947nsBK4PDU/unAc5170JbFPYESMbOVZnasmc0AXkXytDlzhFPcl1q/F9gGmDpg/55hfc+w3Wsf\nwCNm9sygH5b0VGrZG/gO8DPgQkkPSvqSpG1IvIltgNWS1kpaC3wDeHHmXG7Jw6n1PwGYWXfaTsHO\ngyVdJekRSetIPI/ONdqT1DUys6dJBKTDPsBPUnavJGm6TMthe+1wEZgQZvY7Eq/gVSHpjyRPpg4v\n6fFve6XW9yZ5Sv1hwP6OG/8gSQHvtQ+SJysDtrFNTYOdzOz/zOw5M/usmc0maaO/g8QNv4/EE5hq\nZlPCsrOZ7d/v3AXzfRIPay8z2wX4T0Bh32pgRudASTuQNDE63AfMT9k9xcy2N7MHSrY5KlwESkLS\nfpJOkjQjbO9F4iZfGw65GfgrSXtL2gU4tcdp3i9ptqQdgc8BF5vZhtT+T0vaUdL+wHHAD0L6BcCn\nJO0haSqJW/zdAeY+DOwe7OiXnzdJerWkrUg64Z4Dnjez1cDPgS9L2lnSCyS9XNJfp849Q9K2A34/\nDy8CHjOzZyQdBPxtat/FwDslvT78/mlsEghIBON0SfuEPO4haWFJdkaLi0B5PEnSwXWdpD+SVP4V\nwEkAZnY5SaW9BVhO8jqxm++QeA8PkfTO/2PX/l8Cq4ArgDPM7Och/fPAsnDu3wI3hrSeBC/lAuCu\n4Brv2eOwl5BUqidI3OZfBvsg8Qi2BW4j6Vy7mKR9DXAlcCvwkKS0F1MUfw98TtKTJGJ3UWeHmd1K\n0vl3IYlX8BTJ24XOa9qvkngRPw//fy3JPWsVCh0iTmRIupqk4/DbPfbNBO4GtjGz9ZO1rL6ENwpr\ngVlmdnfV9sSCewJOo5H0ztBkeiFwBolndE+1VsVFaSIg6QhJt0taJemUsn7HcYawkE2DpmYBR5m7\nv5tRSnMgdB79HngLyTvyG4Cjzey2wn/McZxclOUJHASsMrO7zOzPJB0zret1dZw6UNbHIy9l84Es\n9zOg13Xq1Kk2c+bMkkxxHAdg+fLlfzCzPbrTK/uCTNIJJB/AsPfee7Ns2bKqTHGcViDp3l7pZTUH\nHmDz0WwzQtpGzOybZjbXzObusccW4uQ4zoQoSwRuAGZJ2jeM1DqKZFCG4ziRUUpzwMzWS/ooyQcn\nWwHnhtFbjuNERml9AmZ2GXBZUeeThL/edZziqcWIQUkb/3bWHccphqhFoF+ldyFwnOKIVgSGVXT3\nChynGGo106yZVVrxu3/b+yicJhC9CAyraOmKWVal7Cc83lnpNIHoRWAUyhAEb3I4TSd6ERhWCfs1\nEcqovGlhSb+xcG+gP5MSUb8H4xOtCAxr/6dveq/KOUmKFoJ+efCC3h8X4/GJ9u3AuEyqIEy6/2Gc\n86QXx+lHtJ4AjF/Ruv8vT1/BMG+kyEqb9ZheeYi1ondsLaq/ZhIdwW2jcZ5AmZRV0XqdtztKTBV2\nlUXd7G06UXsCMdD9tB/0NC7q90ZJ7ycg/fZVRdVjPJz+uAhkoFelKvvtQxX/P0na0pFXhwFm3hwY\ngVGf0mVTtydrjBWgDAZ1yMbYUesiMCLpNnqW9npZjPKEianyxWRLGWSt4DEJgTcHxiSmwhybh9JW\nsgjzJPuXsuIi0CL89Vp5ZL22MXaQ5hIBSfeQBN7cAKw3s7mSdiMJtDmTJNzTe83s8XxmOnmJreC1\nmdiEoIg+gTeZ2Rwzmxu2TwGuMLNZJNFyPQRZRiY5DsFxOpTRMbgQWBLWlwBHlvAbjaL724fuSptO\n8wpdf2K7h3lFwEhiuy8PwUQAppnZ6rD+EDAt52+0kl6vmfzLxfrR/Q1HjOMG8nYMvtHMHpD0YuBy\nSb9L7zQzk9Qzl90RiCZFrBVo1HZijL3MzuZk+Qw+BnJ5Amb2QPi7BvgJSSDShyVNBwh/1/T538oi\nEMXmjnXIO+5g2LwKsRQ6J657MbYISHqhpBd11oG3AitIIg0tCoctAi7Ja2TRxHQDetH98VDWj4mK\n/HrSKZeYPvXO0xyYBvwkZGBr4Ptm9t+SbgAuknQ8cC/w3vxmFkcTKkOMHwg59WVsETCzu4ADeqQ/\nChyexygnG/2+cOx1XCw0VbjS96LT71T1jFdZ8RGDNSe2gSeDqIudRdDdAR3zRDD+AVEDGNRfEJMX\n0E0slaAoYr7Wg3BPoGHEXBAHzcswyO6miUVsuAg4E6dXE2acih7zmI9hxCRs3hxwKmHcMRExVvoO\n6Y7BUV79VZ0n9wScSqm6AhTJOB5ODPlvnQj4ABqnTEYZwxFL+WudCKSJtU3pDKYOr0XrVK5a2SdQ\npxvkOGXTShFwHGcTjReBGL/fdpyYaFWfgH940xxczIujVSLgld9xtqTxzYG6BOWYNC6ITofGiwD0\nruxtFoAOLgQOtKg54JW+Nz5WwmmFJ+A4Tn+GioCkcyWtkbQilbabpMsl3RH+7hrSJeksSask3SLp\nwDKNd0ane6IL9wKcLJ7AecARXWn9ogzNB2aF5QTg7GLMdIrEK7+TZqgImNn/AI91JfeLMrQQON8S\nrgWmdKYfdxwnTsbtE+gXZeilwH2p4+4PaVsg6QRJyyQte+SRR8Y0w3GcvOTuGLTErxzZt6wy+Ijj\nOJsYVwT6RRl6ANgrddyMkOY4TqSMKwL9ogwtBY4JbwnmAetSzQbHcSJk6GAhSRcAhwJTJd0P/Bvw\nBXpHGboMWACsAp4GjivBZsdxCmSoCJjZ0X12bRFlKPQPnJjXKMdxJoePGHScluMi4Dgtp5EfENVh\nmmfHGcakZsZunCeQ5fPYGGLCO04/epXPMstrozyBYfMJ9ruw7hk4daCs8tooEUgzbCIR9wSc2OlX\nXoueA6JxzQHIppQuCJMnHaPPm2Sj4X0CGRinQMUuBG2oLE3PXx1ojAgUQcyFMWbbstKZx6DXU83F\nYEsmdT0aIQJ5XqXE3Ck4rGOzzvQThCblsQgmcT0aIQJ5qZMQtAEXgs0p20tyEQjEPOVW7H0Xeenn\nETQxr1mYdFl0EagJTRcC6O31NDWveShaIFwEakRbhCBWj2zSdF+Hsq5NYwcLNRUz2ygATQ4c0i9f\nHmW6eGonAsOegHmfkLEXql5Dn2O3uUwm9ZFNk6lVc2ASLnAdO6TqZm8eBrnEdbx3MTBuBKLTJD0g\n6eawLEjtOzVEILpd0tuKNLbtSj/IFW5bBcgiBm26HnnI0hw4D/gacH5X+mIzOyOdIGk2cBSwP7An\n8AtJrzSzDQXYCgzvQR5XKGIuML1s6+Qz3UfQObaJYtnvC7rOdr/751+KDmfcCET9WAhcaGbPmtnd\nJBOOHpTDvsLo9fFKvydGTAVmkAD0245Z0MYhS34GDUnOeo62kqdP4KMh6Oi5nYCkRBCBKEslH0Td\nBKBfepsLvb9mHI1xReBs4OXAHGA18OVRTxBDBKL00yO2gtPLOxlmnwvB5sR6b2NjLBEws4fNbIOZ\nPQ98i00uf+URiLLe9DoVilFsdSFwRmUsEdDmkYbfBXTeHCwFjpK0naR9SUKUX5/PxPFpwqer44iV\nC0EzmNR9GzcC0aGS5pAEIr0H+BCAmd0q6SLgNmA9cGKRbwb60d1DPsoxTe1Nz3JNnHiZ5L1TDBVg\n7ty5tmzZslznGPVVUF3eCOS1qwkj6pqQh6wMqvwFlIXlZja3O71WIwazkFVBYy1MRT8BYvnoyOcW\nHE5V16VxIgD1bQOXZXcsQpAmFjtiYZy3QUXRGBGI9cmelVHGBIxD1UIwaOKQtgtC1U3TxogAjDaW\nPqZ25qSeAlULQceGOr+tKZruclhFWazdp8Sj0saClZUqx9X3G/PflrH+MZXLxonAsA9K+h0fC1W8\n46/yNemo96sJlN30G5XGiUCHLIUrNgGA6ipD1eMlBv12jPdpXKpu//eisSLQIYaLHANZvq6rWgia\nSj9hj+VaN14E6kQRo/xGLVhtmY+gKmJz/XvhIlAxedzDopoOLgTFUVTzc5JD3Bv1irBuxNQZ5h8d\nlUtR4l7G2AoXAWcjLgT5mPT1Kur3XAQiYVQ3b9icekXZ4UKQjWHXadzrOIlJUVwEIiDvDS674roQ\nxEFZguAdgxGQt8On6E4k7xTMR928KReBCoi9UDhxU3T5cRGoIVkLgb/qq55xKmz6le0kHhhZIhDt\nJekqSbdJulXSx0L6bpIul3RH+LtrSJeks5REIbpF0oFlZ6LujDIb0qiFwr2O6ij72hcl8Fk6BtcD\nJ5nZbGAecKKSSEOnAFeY2SzgirANMJ9kgtFZwAkk05M7KdI3L8uNbOtntnWl1/0aZyTnpMgSgWi1\nmd0Y1p8EVpIEFFkILAmHLQGODOsLgfMt4VpgijafndhhtG/He8VHGGVxyqeMD6AmdV9HekUoaSbw\nWuA6YJqZrQ67HgKmhfVMUYhUUgQix6mKuopwZhGQtBPwI+DjZvZEep8lOR0ptxZBBCLHcTKKgKRt\nSATge2b245D8cMfND3/XhPTKoxA5jpOdLG8HBJwDrDSzr6R2LQUWhfVFwCWp9GPCW4J5wLpUs8Fx\nnMjIMk7gDcDfAb+VdHNI+1fgC8BFko4H7gXeG/ZdBiwgCUv+NHBcoRY7jlMoQ0XAzK4B+r2fOrzH\n8QacmNMux3EmhH9A5Dgtx0XAcVqOi4DjtBwXAcdpOS4CjtNyXAQcp+W4CDhOy3ERcJyW4yLgOC3H\nRcBxWo6LgOO0HBcBx2k5LgKOUzB1mxOyMVOOpy96HaZ0cpxYaIwI5KEOAlIHG5160ormwKBwzt1p\ndXLjnLipS1lqhQikSd+YutwkxymTPBGITpP0gKSbw7Ig9T+nhghEt0t6W5kZGIe6ddw49aJuzbUs\nfQKdCEQ3SnoRsFzS5WHfYjM7I31wiE50FLA/sCfwC0mvNLMNRRpeJOnYb47TNvJEIOrHQuBCM3vW\nzO4mmXD0oCKMLYO6qbbjFE2eCEQAHw1BR8/tBCQl8ghEdYsO4zhlkycC0dnAy4E5wGrgy6P8cNER\niLxCO854jB2ByMweNrMNZvY88C02ufwegchxasTYEYi6Ig2/C1gR1pcCR0naTtK+JCHKry/OZCc9\n7qHfEjNZ7B9nccYjTwSioyXNIQlEeg/wIQAzu1XSRcBtJG8WToz5zYBTLpOsnJ3fakrTcFL5yROB\n6LIB/3M6cHoOu0pDUi0LSdNfYw67J6Pkva73uMOk73Mrvx2oayHJYnNMQlG2Ld3XI6a8j0sVeWis\nCAy7mE1zHWOn13XOWuCzfjxVZ2+pn92TKJ+N/HZgVNfRKZd+BTmd3u8+tOHrye68T3ocS2M9gV70\nK3QuBNVR5NO7bvexl72DmjhliUJjRWDYBauz69gm0vdpUF9O3TyGLAIwKRorAlnoXPS6FaC6kbXy\n9qNbCLrT6kYv979KGikCo17Uqm/CMOpa2NPkfSPTXemr7Egbl5ie/mka1zEYw0V1Enq1b9NP9FE9\nsCxNvLoQ0wdsjfIEYrmokybmfPdy2/O4w/2aAbFdg2GvMmOiUSLQdGIrPFnp1ffSva+o88VMrPfP\nRcCZGEVXglgrVd1oXJ+A4zQJHzHoOE7puAg4TstxEXCcluMi4Dgtx0XAcVpOljkGt5d0vaTfhAhE\nnw3p+0q6LkQa+oGkbUP6dmF7Vdg/s9wsNJ+YRpc5k6fs+5/FE3gWOMzMDiCZXvwISfOAL5JEIHoF\n8DhwfDj+eODxkL44HOc4TqRkiUBkZvZU2NwmLAYcBlwc0pcAR4b1hWGbsP9w1W1ol+O0iKxxB7YK\nMw2vAS4H7gTWmtn6cEg6ytDGCERh/zpg9x7nrCQCkeM4m5NJBEKQkTkkgUQOAvbL+8NFRyByHGc8\nRno7YGZrgauAQ4ApkjrfHqSjDG2MQBT27wI8Woi1juMUTpa3A3tImhLWdwDeQhKZ+Crg3eGwRcAl\nYX1p2Cbsv9K8a9txoiXLV4TTgSWStiIRjYvM7FJJtwEXSvo8cBNJqDLC3+9IWgU8BhxVgt2O4xRE\nlghEt5CEI+9Ov4tNQUjT6c8A7ynEOsdxSsdHDDpOy3ERcJyW4yLgOC3HRcBxWo6LgOO0HBcBx2k5\nLgKO03J8yvEuYosT5zhl455AoDssVjrdcZqMi0Af0h5ARyBcEJwmEmVzYJTKVmZUm35x9LyJ4Ayj\niAfGpMpZ7T2Bop/O/rR32kaUnkA3vUJcd2/nUc1BFb+7WVB3suTBPZ18jBpyvWqiF4FeFzFLuOs8\nuMvvtInaNgfKrqRNeOo71VKXB0nUnkBMF7FpohDTtXWqJWoRGOSWj9Lu6hyb5VxZ0quilz1Nrczj\nXvuqr0fa7n7lbljeJp2HPBGIzpN0t6SbwzInpEvSWSEC0S2SDsxj4CQG8HQivGSJ9DLJG5Qen1C0\nUMUmcJB/PEbVAtCPLPex+/hJksUT6EQgekrSNsA1kn4a9v2zmV3cdfx8YFZYDgbODn8z093xl7X3\nviia9kYgBsq+jrEKwLhMsnM6yxyDBvSKQNSPhcD54f+ulTRF0nQzWz2KYb3eAHTvnwSxFK5RXcq6\nk+W6x3wNBj1IBuWt++E3ifI3VgQiM7su7Do9uPyLJW0X0jZGIAqkoxOlzzk0AlG/CxBLxSybrE2U\nOtDd5Bq2jHrumOi2ZxT7qsjLWBGIJL0KOJUkEtHrgN2Ak0f54awRiPIWEKeZxOoFZHlw1a5jME0q\nAtERZrY6BCt9FvgvNk0/vjECUSAdnchxCqVOD4Vx+pomIXbjRiD6naTpIU0kEYlXhH9ZChwT3hLM\nA9aN2h/gOIOI1QuoK3kiEF0paQ9AwM3Ah8PxlwELgFXA08BxxZvtOPWk0+EdkweTJwLRYX2ON+DE\n/KY5ZTDsrUvs1O3jnF7EZndtvx1wHKcYXAQcp+W4CLSYOjcLYnOp64yLgOO0nKi/InSGk/eJGEtP\ndZ29krrjnkANyVthYqj0aVwAqsVFoGY0rcI0LT9FMOnXoN4cqBFNjI6UNQ9NGB8QKy4CNaDoGYX8\n6RsvVYidi0DkFCUA3YWrrkLQZC+gqnviIlBDRn1alDk9+ySom71FMEmxcxGInE5hKGqOwWHnc6pn\n0t6Oi0BNyDu9WJPd6KZQ1T1yEagpWUUha3PBRaK9uAg0hHHm5atDk6AONtYdHyzk1Ab3VsrBRcBx\nWk5mEQjTjt8k6dKwva+k60KkoR9I2jakbxe2V4X9M8sxvaeNI0d7cZy2M4on8DFgZWr7i8BiM3sF\n8DhwfEg/Hng8pC8Ox1WGC0Ez8KZAeWQNPjIDeDvw7bAt4DCgE4JsCcmMw5BEIFoS1i8GDtcEaqKP\nLW8mHmuifLJ6AmcC/wI8H7Z3B9aa2fqwnY4ytDECUdi/Lhy/GVkiEOWhV+Fxr2BzPKCLA9niDrwD\nWGNmy4v84awRiAr4nc22XQgcZ3OyjBN4A/A3khYA2wM7A18FpkjaOjzt01GGOhGI7pe0NbAL8Gjh\nlo9AXd6JO04VDPUEzOxUM5thZjOBo4Arzex9JOHI3h0OWwRcEtaXhm3C/ivN/U3HiZY84wROBj4h\naRVJm/+ckH4OsHtI/wRwSj4TR8ef+o6TnZGGDZvZ1cDVYf0uNgUhTR/zDPCeAmwbiW6X34XAcbLR\nqBGDWVod3jJxnM1plAjA4EruAuA4W9LIrwi9sjtOdhrnCTiOMxqK4akp6Ung9qrtKJCpwB+qNqJA\nPD9xkzU/+5jZFiPzYmkO3G5mc6s2oigkLfP8xIvnZ3O8OeA4LcdFwHFaTiwi8M2qDSgYz0/ceH5S\nRNEx6DhOdcTiCTiOUxEuAo7TcioXAUlHSLo9TEw68S8Ox0HSuZLWSFqRSttN0uWS7gh/dw3pknRW\nyN8tkg6szvItkbSXpKsk3SbpVkkfC+l1zc/2kq6X9JuQn8+G9Ogmxh2FMif6rVQEJG0FfB2YD8wG\njpY0u0qbMnIecERX2inAFWY2C7iCTZ9QzwdmheUE4OwJ2ZiV9cBJZjYbmAecGO5BXfPzLHCYmR0A\nzAGOkDSPmkyMO4DyJvpNzzM36QU4BPhZavtU4NQqbRrB9pnAitT27cD0sD6dZAAUwDeAo3sdF+NC\nMjnMW5qQH2BH4EbgYJIRdVuH9I3lDvgZcEhY3zocp6pt78rHDBIhPgy4FFCR+am6ObBxUtJAesLS\nujHNzFaH9YeAaWG9NnkMruNrgeuocX6C63wzsAa4HLiTnBPjVkzhE/2mqVoEGoklMlyrd6+SdgJ+\nBHzczJ5I76tbfsxsg5nNIXmCHgTsV7FJY1PWRL9pqhaBzqSkHdITltaNhyVNBwh/14T06PMoaRsS\nAfiemf04JNc2Px3MbC3JXJiHECbGDbt6TYxLLBPjdtGZ6Pce4EKSJsHGiX7DMbnyU7UI3ADMCj2d\n25JMZLq0YpvGJT3BavfEq8eEXvV5wLqUm105ITDMOcBKM/tKaldd87OHpClhfQeS/o2V1HRiXJvE\nRL8RdHosAH5P0m77ZNX2ZLT5AmA18BxJe+x4knbXFcAdwC+A3cKxInkDcifwW2Bu1fZ35eWNJK7+\nLcDNYVlQ4/y8Brgp5GcF8JmQ/jLgemAV8ENgu5C+fdheFfa/rOo8DMjbocClRefHhw07Tsupujng\nOE7FuAg4TstxEXCcluMi4Dgtx0XAcVqOi4DjtBwXAcdpOf8PlVD9yjwwzV0AAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Got an average of 55.0% for 16 way one-shot learning accuracy \n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).\n",
            "Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAQEAAAEICAYAAABf40E1AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nsYoAAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAQEAAAEICAYAAABf40E1AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjEsIGh0\ndHA6Ly9tYXRwbG90bGliLm9yZy8QZhcZAAAbVElEQVR4nO2dedAlVXmHn5/siDLCfI7DLAzGqTK4\nITUiLlEENYAL/KEWRsOgWFNRkmhpSoe4RC1TUcsIUjEqJZbjxiJqQVEYJSym1IAMsgiMyAdCZoaB\nGYHBFWV580ef6/Rc+t7b997u2+d0v0/Vra/7nP663z7L77zn9OnTMjMcx+kuj2vaAMdxmsVFwHE6\njouA43QcFwHH6TguAo7TcVwEHKfjuAg4jSPpu5JWN21HV3ERqBFJL5b0Y0kPSLpP0o8kPa9pu/qR\ndISkTTWd+yRJPxx2jJkdY2br6ri+M5pdmzagrUh6InAR8HbgPGB34K+APzZpVz+SvAx0HTPzXw0/\nYBWwfUj8h4Gv5fZXAAbsGvavAP4N+Anwa+ACYL++Y9cAdwFbgH/KnWsP4PQQd1fY3iPEHQFsAt4H\n3A18E/gD8Cjw2/A7oMDePYGvAfcC24GrgUUhbl/grGDHZuBjwC7AXwIPAo+E8xamR7jXt4Xtk4Af\nAaeF69wOvDCEbwS2Aqtz//sq4NqQRhuBD/ed+0TgzmD3B4E7gJeHuMcBa4HbQvx5vTTu0s+7A/Xx\nC+ARSeskHSPpSROc40TgrcBi4GHgjL74lwErgVcC75P08hD+fuBw4BDgOcBhwAdy//cUYD/gwHCN\nY4C7zGyf8LurwJbVZJV9GbA/8Hdk4gHw5WDf04DnBnveZmYbwnH/G867oOR9Px+4IVznG8A5wPPC\n+d8M/IekfcKxvwv3sIBMEN4u6XgASQcD/wm8iSwN9wWW5K7zD8DxwEuBA4D7gc+WtLE9NK1Cbf6R\ntYRfJmt5HwYuZEfr+WFGewIfz8UfDPyJrIXtHfv0XPwngbPC9m3Asbm4vwbuCNtHhPPsmYs/Atg0\n4l7eCvwYeHZf+CKyLs5eubA3ApeH7ZOAH4449xXs7Ancmot7VrjXRbmwe4FDBpzrdOC0sP0h4Oxc\n3N7h3nuewAbgqFz8YuChXh505eeeQI2Y2QYzO8nMlgLPJGttTh/jFBtz23cCuwELh8QfELYPCPtF\ncQDbzOzBYReW9NvcbznwVeB7wDmS7pL0SUm7kXkTuwFbJG2XtB34AvDk0nf5WO7Jbf8BwMz6w/YJ\ndj5f0uWStkl6gMzz6KXRAeTSyMx+TyYgPQ4EvpOzewNZ12XRFLYnh4vAjDCzn5N5Bc8MQb8ja5l6\nPKXg35bltpeTtVK/GhLfc+PvIivgRXGQtawM2cd2dA32MbP/M7OHzOwjZnYwWR/91WRu+EYyT2Ch\nmS0Ivyea2TMGnbtivkHmYS0zs32BzwMKcVuApb0DJe1F1sXosRE4Jmf3AjPb08w212xzVLgI1ISk\np0t6j6SlYX8ZmZt8ZTjkOuAlkpZL2hc4teA0b5Z0sKS9gY8C55vZI7n4D0raW9IzgLcA54bws4EP\nSJqTtJDMLf7aEHPvAfYPdgy6n5dJepakXcgG4R4CHjWzLcD3gX+X9ERJj5P0F5Jemjv3Ukm7D7n+\nNDwBuM/MHpR0GPA3ubjzgddIemG4/ofZIRCQCca/Sjow3OOcpONqsjNaXATq4zdkA1xXSfodWeW/\nEXgPgJldQlZpbwCuIXuc2M9XybyHu8lG5/+xL/4HwDxwKfApM/t+CP8YsD6c+2fAT0NYIcFLORu4\nPbjGBxQc9hSySvVrMrf5B8E+yDyC3YGbyQbXzifrXwNcBtwE3C0p78VUxTuAj0r6DZnYndeLMLOb\nyAb/ziHzCn5L9nSh95j2M2RexPfD/19JlmedQmFAxIkMSVeQDRx+sSBuBfBLYDcze3i2lqVLeKKw\nHVhpZr9s2p5YcE/AaTWSXhO6TI8HPkXmGd3RrFVxUYsISDpa0i2S5iWtreMajlOS49gxaWolcIK5\n+7sTlXcHwsDRL4BXkD0fvxp4o5ndXOmFHMephDo8gcOAeTO73cz+RDYo07kRV8dJhTpeHlnCzpNY\nNjFixHXhwoW2YsWKGkxxHKfHNddc8yszm+sPb+wNMklryF6AYfny5axfv74pUxynE0i6syi8ju7A\nZnaeybY0hO2EmZ1pZqvMbNXc3GPEyXGcGVGHCFwNrJR0UJildQLZhAzHcSKk8u6AmT0s6e/JXjbZ\nBfhSmLnlOE6E1DImYGYXAxfXcW7HcarFZww6TsdxEXAqQ9Log5zo6Owik/kC67NIpyOflpI8PROj\nsyLgTE9Ry9+0APTb1LQ9KdBZETCzqNzXlFrQWCtaUX72wmKxMUZ8TIBq+7KS/vxr4vqzJLdAZ1QU\n2ZRqGs+CJD2BtmZoKt5Az4uK2d68XW0tL1XhngDNu4pNX9/pNkl6AlUSWwWMuXV1JqOMJ9Jknicp\nAv0JNu4o9awHi8o8jswPVKYgBCl0CZxyJCkCeaYZqW6iAKdYaVKzObYxgEFpF8uTi9aMCYwzUt10\nog8qpE3b1SP/hCPvnQw73hmfXn43nX5JewJVJN4sW7lRcxOaLgyjbOhPq9jmWsQiooOoM++nufdW\neAIxZv4om8adSzALirpWsc4FcHZmmrKUrCcwbX9qVq1YUcXqn2tfZFuTjBpUbdq+PNPkYVP3UTSH\nYRpbpi3HrfAEYqS/pe93o4c9JWiKSVr9WPq1baGJdEzWE6iCph9zxdSijiK2/n+PVNJwnBZ/1unc\naRGog1QKpeP0cBFwnIhoohHxMQFnanxcYDixp4uLgOPMiFi7it4dcJyKGNXiz3Ki2DiC456A43Qc\n9wQcpyJidfdH4SLgjE3sL0A545GsCFRd4GKbDhsLsY9sO9OTrAhUhVf88fE0axedFwFnNF7p240/\nHXCG4gLQflwEHKfjuAg4TsdxEXCcjuMi4DgdZ6QISPqSpK2SbsyF7SfpEkm3hr9PCuGSdIakeUk3\nSDq0TuMdx5meMp7Al4Gj+8LWApea2Urg0rAPcAywMvzWAJ+rxkzHcepipAiY2f8A9/UFHwesC9vr\ngONz4V+xjCuBBZIWV2Ws4zjVM+mYwCIz2xK27wYWhe0lwMbccZtC2GOQtEbSeknrt23bNqEZjuNM\ny9QDg5bNJhl7RomZnWlmq8xs1dzc3LRmOI4zIZOKwD09Nz/83RrCNwPLcsctDWGO40TKpCJwIbA6\nbK8GLsiFnxieEhwOPJDrNjiOEyEjXyCSdDZwBLBQ0ibgX4CPA+dJOhm4E3hDOPxi4FhgHvg98JYa\nbHYcp0JGioCZvXFA1FEFxxpwyrRGOY4zO3zGoON0HBcBx+k4rRaBGD//7Tix0WoRcBxnNC4CjjND\nYvROfY3BiBjn89VNUVUBjvkeu4aLwIwZVNHLfKKqjooTW6vUJHUsOz8ofSdN9zrKgHcHGmJUpc//\nYqYtLXovnatM7ya/LzgO7gnMGDMrLByDwmdpz6hCNui4ov+bxONpirxNdVS0/DnH8erqtqtHK0Wg\nv6ANqnSxkkLFaQuzqmgx06ruwDjus1csJxXqFqdWegJl3Vr//mA7aYvAz6pstsoTKItX/OYws1rT\nP3YB6N17THa20hMYB/cG2kF/pRp3kLPXlayqLMRUyUfRGk8gpUR36mNaT2PW5SjfNW3qkXDrPIGU\nWnUXrumpcgZjU/kx7GnWLMpzazyBIlKYbONUS+z5nbevqAvTxOPhVngC/QmbkjdQNJEk9oIcE4PS\nr+kyMK5nkbe3f/Cw7vtphSdQpJ5l5uI3QRsmp8QqUrGm56DJavlfEbO6n1Z4AjD40cso9yrGgtPU\njMFxrhOrEExK7PdTpzfQGhHoMSyh2tAKO9VSV6NQh6jUJQStE4FBFI3ANikEZa89izUGXBAz6kqH\n/v7+OHMSZjG5qBMi0O8BND2A5JWueerOg0Hnj7Hb0YqBwWEUdQFS8ADyx7totI9J87QOEWm1CJQZ\nA4hRmR1nlnSiO1AkAE3OEBuEt/jliC3fZkGd5bW1ItDFguK0V0h9YHBM/FFgu+lantZdnls9JuA4\nsVKmZe+f+eqThSZgnGfxXWtdnOYpquCDxMEXGq2J/jkDg45xnKoYVNmbLH+dFgHHaYqitx+L4mbB\nyDEBScskXS7pZkk3SXpnCN9P0iWSbg1/nxTCJekMSfOSbpB0aN03McT2kcf0v81V5u0ux6mSpstc\nmYHBh4H3mNnBwOHAKZIOBtYCl5rZSuDSsA9wDLAy/NYAn6vc6hF45XWc8owUATPbYmY/Ddu/ATYA\nS4DjgHXhsHXA8WH7OOArlnElsEDS4sotH4G35I5TjrEeEUpaATwXuApYZGZbQtTdwKKwvQTYmPu3\nTSGs/1xrJK2XtH7btm1jmu04TlWUFgFJ+wDfAt5lZr/Ox1nW5I7V7JrZmWa2ysxWzc3NjfOvjuNU\nSCkRkLQbmQB83cy+HYLv6bn54e/WEL4ZWJb796UhzHGcCCnzdEDAWcAGM/t0LupCYHXYXg1ckAs/\nMTwlOBx4INdtcBwnMsrME3gR8LfAzyRdF8L+Gfg4cJ6kk4E7gTeEuIuBY4F54PfAWyq12HGcShkp\nAmb2Q2DQA/ejCo434JQp7XIcZ0b4C0SO03F82rDzGMZ9d93nY6SNi4CzE5MsXuEvX6WNi4BTyDgV\n2FdxShsXAacyvOVPEx8YdJyO4yLgFOKfde8OreoODCu07qqWo39pa196rf20RgS81aqOYevdNb0K\njlM9rRCBUSuyjvMByCoZ9F36VCjzMUz3GtIneREYZ0nmGAppasIwyt5BC2bGfE8xMo0nO21aJy8C\nPWIsdIO8kjJhw84xK8YR2H6vIQbBjZ1YurBJi0AsiViW1OztMawy98fF+I3HVGhKNFvxiDCFFie1\nipGavSmSL7dNluGkPYEUKOvqxzrqHosd0xJrlysGWuEJxMqwPvWo/VgY1yOI0YNwARhOkp5Aal8d\nTqXCtx1P92KSFIFxaSLzy35rLk9Mg2rTptks0jzWLtQkNPk0xbsDNTBON6BNxCJgzngkLwIxF7wy\nM+7aQhNdtP7v9/lLT5ORvAhAXJWs64NQMdxjTOVhFDGkV9IiMOwRWxPkp8xOkrkx3MMkxDA7sD/N\n8zMXY6dpjzE5ESj7MkvKNF2hJiEGIegntadITZGcCPToZWrsXx9uizANIrb7i7kslKGJ9ExWBAYR\nQ6HsDVCN0z2Iwe5xibWljcmWsjRpc+tEIHVSKcCxCkCPWOblT8KsG4TWTBaKIaOrsCGF9/FjF4Ae\nMdtWRH6yWNEYS11lI0kRSC1zJyFWMUhFAFKlXwhmQXLdgbYWvNgHOMEFIAbqSPckPYE2E2vlcgGY\nHbNO3+Q8AcdxqsU9AWco7gG0H/cEHKfjjBQBSXtK+omk6yXdJOkjIfwgSVdJmpd0rqTdQ/geYX8+\nxK+o9xacWeBeQHsp4wn8ETjSzJ4DHAIcLelw4BPAaWb2NOB+4ORw/MnA/SH8tHCckygpPLVwpmOk\nCFjGb8PubuFnwJHA+SF8HXB82D4u7BPij1KKc2IdpyOUGhOQtIuk64CtwCXAbcB2M3s4HLIJWBK2\nlwAbAUL8A8D+BedcI2m9pPXbtm2b7i4cx5mYUiJgZo+Y2SHAUuAw4OnTXtjMzjSzVWa2am5ubtrT\nOY4zIWM9HTCz7cDlwAuABZJ6jxiXApvD9mZgGUCI3xe4txJrHcepnDJPB+YkLQjbewGvADaQicHr\nwmGrgQvC9oVhnxB/mfnIkuNES5nJQouBdZJ2IRON88zsIkk3A+dI+hhwLXBWOP4s4KuS5oH7gBNq\nsNtxnIoYKQJmdgPw3ILw28nGB/rDHwReX4l1juPUjs8YdJyO4yLgOB3HRcBxOo6LgON0HBcBZyz8\nU1/tw0XAKU2+8rsQtAdfVMRxIqfuhV3cE3BK4S1/HNTRHXNPwBmL/JLYsTHKrv4Pf6Yym70ozav8\n9mNrRcDXxquGNqVjrOJVhjq/wN1aEcgzKNGaKNSptUI9UrMXHmtzGwStDiFrpQj0V7RZtgBlr9WG\nAhkbebe5KE09nYtpnQgUVcJhLULd1y5iVL86Js8lpYqTsrs/LlXmS2ufDsy68I4jAPm/45ynS4V8\nXLrgWdWV/63zBHqtbJWjp5NcP78POzIwb1e/rcMKsgtAOWIUgKrzblRZGZfWegKzZNBMunzmjMqo\nspnqYtBd6lr+vXWeQJOMGpga93+8wreDOipulfM1XASmpK5Hfi4A7WPUoHXZsuTzBHIMG0VvYlyg\nTIs+rj1F52pqvMOpnlEVv+wsyGlIVgRSGUWPyRanOYoqe3/ZGDS3ocxj72lIfmCwN1gS4zfzyghA\n0TGpziqMgZTWOxhn3KhOkhSBYZkcU8UpEqaUCmkRsdo+KJ1jtbdH2car/7gq7ytJEUiNGL2UaYn5\nftomvHXTShFouoBOO+130DsP3k0YTl5si7qIsXsHZWzzRUUCRRkbA3W+7ulMzrBu2azzKcZykaQI\nwPA+YCwJPagln+QcMdxTDDZMw6BuWUxlBmb/5CtZEYBil68/ftj/VWUD7Oyqj3JBy0wnjtmrSL07\n0sTTpFH5WdRgzCrfk50n0E8MBbPo5aB+xnnPvegcTd1nbEJUFbMWglGeXX+ez+K18taIQJMMytwq\nMioGccsTmz2pMWjQt8ijnNW6Ei4CFVHlCx2x0db7apKycwNmQdJjArHRxlayC4t1dB33BCqmrRWl\nrfcVG02kc2lPQNIukq6VdFHYP0jSVZLmJZ0rafcQvkfYnw/xK+ox3XGcKhinO/BOYENu/xPAaWb2\nNOB+4OQQfjJwfwg/LRznJEobpzw7O1NKBCQtBV4FfDHsCzgSOD8csg44PmwfF/YJ8UfJR5YcJ1rK\negKnA+8FHg37+wPbzezhsL8JWBK2lwAbAUL8A+H4nZC0RtJ6Seu3bds2ofmO40zLSBGQ9Gpgq5ld\nU+WFzexMM1tlZqvm5uaqPLXjOGNQ5unAi4DXSjoW2BN4IvAZYIGkXUNrvxTYHI7fDCwDNknaFdgX\nuLdyyx3HqYSRnoCZnWpmS81sBXACcJmZvQm4HHhdOGw1cEHYvjDsE+IvMx9ZcpxomWay0PuAd0ua\nJ+vznxXCzwL2D+HvBtZOZ6LjOHUy1mQhM7sCuCJs3w4cVnDMg8DrK7DNcZwZ4NOGHafjuAg4Tsdx\nEXCcjuMi4Dgdx0XAcTqOi4DjdBwXAcfpOL6oiNMpYlm4NSZcBJzk8TfVp8O7A07niKH1j+mDJ+4J\nOAOZtpDOqrKNWq676GtVzg5cBJzaiL3/HZs9TeEigC+rPYhRaTHqK8mzbHG7+MXmqu65EyIwqJK7\nW1gvTVbINn0Mpm6BS1IEymTuoELQ+15gkTD0Bmu61JpUQdNp1pbK3hTJPR0om+HD+qOj3H8vVGni\n4j0ZSXoCPcr0Rfs/Az5MANrkQnYFz6/pSVYE6vqgo7cmk9H0kwDPt8lJVgQmpanWvu5rxlYJZjVO\nENt9p0iyItD0YFTXGCViRWMusedR7PbNiuREoExLXjZjZ1kIprlO08/AB6X3IHvyedRURRtWRnzs\nZ2eSE4Eq8EIwOam1nKnZ2wRJi4C7c7Nh0gHWpoW2jWWjjjRNbp5AFTRdOLtC0zMG20Zd5TZJT2Cc\nqb/DCkMMBaXtghRDGqfAOOWg/63JaWm9J9CfUG2vdE56TCIAVZKkJ1DEsOm/RaPrMbRQTY/6O81T\n5g3WcZ/OjEurPQEz87cGHWcErRaBmHFBcvI0OXaVvAj0vwbcT+yVzbsCTtn3YOoqK8mLQD89MSgS\nhTZ1DVK3PxaaTscYGoFWDAyOemQy6HXi1Abmmi6wbSKGyUyxUMoTkHSHpJ9Juk7S+hC2n6RLJN0a\n/j4phEvSGZLmJd0g6dA6byBPz2Xq/xUdlyeFwpCCjU6ajNMdeJmZHWJmq8L+WuBSM1sJXBr2AY4B\nVobfGuBzVRlbJbEIQX/3ZdAvT539w67gorqDacYEjgPWhe11wPG58K9YxpXAAkmLp7hObeQrUwqV\nyit/9Xh6lhcBA74v6RpJa0LYIjPbErbvBhaF7SXAxtz/bgph0dJEQRjUdRn2c5w6KDsw+GIz2yzp\nycAlkn6ejzQzkzRWKQ1isgZg+fLl4/yr40yNi+oOSnkCZrY5/N0KfAc4DLin5+aHv1vD4ZuBZbl/\nXxrC+s95ppmtMrNVc3Nzk9+B4zhTMVIEJD1e0hN628ArgRuBC4HV4bDVwAVh+0LgxPCU4HDggVy3\nwXGcyCjTHVgEfCeMpu4KfMPM/kvS1cB5kk4G7gTeEI6/GDgWmAd+D7ylcqsdpyXEsDDOSBEws9uB\n5xSE3wscVRBuwCmVWOc4Tu20btqw4zjj0Yppw46TGk13AfK4J+A4HcdFwHE6jouA43QcxdA3kfQb\n4Jam7SjBQuBXTRsxghRshDTsbJuNB5rZY2bmxTIweEvu7cRokbQ+djtTsBHSsLMrNnp3wHE6jouA\n43ScWETgzKYNKEkKdqZgI6RhZydsjGJg0HGc5ojFE3AcpyFcBByn4zQuApKOlnRLWJ147ej/qM2O\nL0naKunGXFhUKypLWibpckk3S7pJ0jsjtXNPST+RdH2w8yMh/CBJVwV7zpW0ewjfI+zPh/gVs7Az\nXHsXSddKuihiG+td7XuSte6q+gG7ALcBTwV2B64HDm7IlpcAhwI35sI+CawN22uBT4TtY4HvAgIO\nB66akY2LgUPD9hOAXwAHR2ingH3C9m7AVeH65wEnhPDPA28P2+8APh+2TwDOnWG+vxv4BnBR2I/R\nxjuAhX1hleX5TG5iyM29APhebv9U4NQG7VnRJwK3AIvD9mKySU0AXwDeWHTcjO29AHhFzHYCewM/\nBZ5PNrNt1/68B74HvCBs7xqO0wxsW0q2XP6RwEWh4kRlY7hekQhUludNdwdiX5k42hWVgzv6XLJW\nNjo7g5t9Hdnak5eQeXzbzezhAlv+bGeIfwDYfwZmng68F3g07O8foY1Q82rfsUwbjh6z8VdUrgtJ\n+wDfAt5lZr/Of0gjFjvN7BHgEEkLyBanfXrDJu2EpFcDW83sGklHNG3PCCpf7TtP055AqZWJG2Sq\nFZXrQNJuZALwdTP7dqx29jCz7cDlZK71Akm9hidvy5/tDPH7AvfWbNqLgNdKugM4h6xL8JnIbATq\nWe07T9MicDWwMozI7k424HJhwzbliWpFZWVN/lnABjP7dMR2zgUPAEl7kY1bbCATg9cNsLNn/+uA\nyyx0aOvCzE41s6VmtoKs3F1mZm+KyUaY0WrfsxogGjLocSzZKPdtwPsbtONsYAvwEFk/6mSyPt+l\nwK3AfwP7hWMFfDbY/DNg1YxsfDFZ//AG4LrwOzZCO58NXBvsvBH4UAh/KvATspWovwnsEcL3DPvz\nIf6pM877I9jxdCAqG4M914ffTb06UmWe+7Rhx+k4TXcHHMdpGBcBx+k4LgKO03FcBByn47gIOE7H\ncRFwnI7jIuA4Hef/AbOpM/mzsFe6AAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "Got an average of 75.0% for 25 way one-shot learning accuracy \n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "31uXAsYDFAdc",
        "colab_type": "text"
      },
      "source": [
        "**Conclusions**  \n",
        "The Hyperparameters can te tuned more to get better results and avoid over-fitting.\n",
        "Future work should also use stroke data for better classification and handwriting recognition."
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "zygF3eolCxWU",
        "colab_type": "text"
      },
      "source": [
        "**Stroke Data**"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8x7aX6FVXuFN",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def get_stroke(fp):\n",
        "\n",
        "  \"\"\"\n",
        "  Opens the stroke data file and add stroke data, each start and break shows the start and end of a motor(i.e 1 movement of hand). \n",
        "  Multiple strokes make 1 motor and it is sent back to be stored in DataFrame.\n",
        "  \"\"\"\n",
        "  with open(fp,'r') as f:\n",
        "        stroke_data=f.readlines()\n",
        "        stroke_data=[l.strip() for l in stroke_data]\n",
        "        motor=[]\n",
        "        for line in stroke_data:\n",
        "            if line==\"START\":\n",
        "                stk=[]\n",
        "            elif line==\"BREAK\":\n",
        "                stk = np.array(stk)\n",
        "                motor.append(stk) # add to list of strokes\n",
        "                stk = [] \n",
        "            else:\n",
        "                arr = np.fromstring(line,dtype=float,sep=',')\n",
        "                stk.append(arr)\n",
        "  return motor"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HdI7A7gOXuFS",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def get_stroke_data(folder):\n",
        "    \"\"\"\n",
        "    Function to get stroke data and returns a dataframe in accordance to language and character \n",
        "    \"\"\"\n",
        "    lang_list=[]\n",
        "    SX={}\n",
        "    for languages in os.listdir(folder):\n",
        "        if languages.startswith('.'):\n",
        "            continue\n",
        "        else:\n",
        "            strokes={}\n",
        "            lang_list.append(languages)\n",
        "            for alphabets in os.listdir(folder+languages):\n",
        "                if alphabets.startswith('.'):\n",
        "                    continue\n",
        "                else:\n",
        "                    stk=[]\n",
        "                    for character in os.listdir(folder+languages+\"/\"+alphabets):\n",
        "                        if(character.startswith('.')):\n",
        "                            continue\n",
        "                        else:\n",
        "                            stk_data=get_stroke(folder+languages+\"/\"+alphabets+\"/\"+character)\n",
        "                            stk.append(stk_data)\n",
        "                    strokes[alphabets]=stk\n",
        "            \n",
        "        SX[languages]=strokes\n",
        "    \n",
        "    sx=pd.DataFrame.from_dict({(i,j): SX[i][j] \n",
        "                           for i in SX.keys() \n",
        "                           for j in SX[i].keys()},orient='index')\n",
        "    return sx"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CjLuFzPqXuFQ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def draw_stroke(motor):\n",
        "  \"\"\"\n",
        "  Draw the each alphabet\n",
        "  \"\"\"\n",
        "  drawing=[d[:,0:2]for d in motor]\n",
        "  for i in drawing:\n",
        "      plt.plot(i[:,0],i[:,1])"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "vb58BdckXuFV",
        "colab_type": "code",
        "outputId": "6ff5ea2a-bbbb-4786-bf76-613d2d40c27a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 51
        }
      },
      "source": [
        "print(\"Getting stroke train data\")\n",
        "stroke_train=get_stroke_data('Fellowship/strokes_background/')\n",
        "print(\"Getting stroke validation data\")\n",
        "stroke_test=get_stroke_data('Fellowship/strokes_evaluation/')"
      ],
      "execution_count": 39,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Getting stroke train data\n",
            "Getting stroke validation data\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pMwa6g9f3fHo",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 533
        },
        "outputId": "02762acb-1b16-441f-cc3c-7d27b3e8c5f5"
      },
      "source": [
        "#Example showcasing the how stroke data is stored\n",
        "draw_stroke(stroke_train.iloc[137,1])\n",
        "plt.show()\n",
        "plt.subplots()\n",
        "plt.imshow(x_train.iloc[137,1])"
      ],
      "execution_count": 40,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXkAAAD4CAYAAAAJmJb0AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.image.AxesImage at 0x7f9a8034ea90>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 40
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAQEAAAD7CAYAAABqkiE2AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "NEYreoK_U-Su",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "2eb9532a-4dcf-4114-dd15-f4a61787844e"
      },
      "source": [
        "from google.colab import drive\n",
        "drive.mount('/content/drive')"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Mounted at /content/drive\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bzcrN6Xq3pb8",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "!unzip -uq \"/content/drive/My Drive/Colab Notebooks/Fellowship\""
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}