RecurrentNeuralNetwork_Batching.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "RecurrentNeuralNetwork_Batching.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "id": "WbHKXUDFNmBP",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from io import open\n",
        "import os, string, random, time, math\n",
        "import matplotlib.pyplot as plt\n",
        "import seaborn as sns\n",
        "import numpy as np"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "rBXnR_W8N98t",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from sklearn.model_selection import train_test_split"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "4P3G4j6pOAcg",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import torch\n",
        "import torch.nn as nn\n",
        "import torch.optim as optim"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DM8-SKmDOCdV",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "from IPython.display import clear_output"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DH4QW1UqOMlx",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8NVFXR5cwPIq",
        "colab_type": "text"
      },
      "source": [
        "Task: Language Classification"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "FtSqcw4l1fsd",
        "colab_type": "text"
      },
      "source": [
        "The aim of this project is to build a model which predicts the language given a name (which is essentially a sequence of characters)"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hcArIR6U16JO",
        "colab_type": "text"
      },
      "source": [
        "<h3> Data Preprocessing </h3>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "U3tTyBEcOZCX",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "languages = []\n",
        "data = []\n",
        "X = []\n",
        "y = []\n",
        "\n",
        "with open('name2lang.txt', 'r') as f: \n",
        "    for line in f:\n",
        "        line = line.split(',')\n",
        "        name = line[0].strip()\n",
        "        lang = line[1].strip()\n",
        "        if not lang in languages:\n",
        "            languages.append(lang)\n",
        "        X.append(name)\n",
        "        y.append(lang)\n",
        "        data.append((name, lang))\n",
        "\n",
        "n_languages = len(languages)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "1Qhgumvfwv35",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "l6BYye4VPDu3",
        "colab_type": "code",
        "outputId": "e5935643-c456-4c96-e407-19af03a7ab04",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        }
      },
      "source": [
        "print(languages)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "['Portuguese', 'Irish', 'Spanish', 'Vietnamese', 'Chinese', 'Greek', 'Czech', 'Dutch', 'Japanese', 'French', 'German', 'Scottish', 'English', 'Russian', 'Polish', 'Arabic', 'Korean', 'Italian']\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "s3fPMqp-w3GK",
        "colab_type": "text"
      },
      "source": [
        "There are totally 18 languages"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "c6jvNhnXPF9E",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=0, stratify=y)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "An3NsR_PxBzH",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "cRVFmOCXxCTF",
        "colab_type": "text"
      },
      "source": [
        "<h4> Creating Vector Representation for Inputs </h4>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "C1eUnNPwxRUX",
        "colab_type": "text"
      },
      "source": [
        "<p> Since the possible inputs are set of all english albhabets(capital and small) we represent each alphabet using a one hot vector. The size of each input vector is equal to the total number of characters (n_letters) </p>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "muSV4gdrPTRk",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "all_letters = string.ascii_letters + \" .,;'\"\n",
        "n_letters = len(all_letters)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "eec5_W80QD9r",
        "colab_type": "code",
        "outputId": "71649cc9-214a-46f9-8948-8e2a4468dca1",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "all_letters"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "\"abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ .,;'\""
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 14
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WTE5btexP2Ut",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Creating one hot representation for each input\n",
        "def name_rep(name):\n",
        "    rep = torch.zeros(len(name), 1, n_letters) # Tensor for word representation, which is basically combination of \n",
        "                                               # vector representation of each character in the word\n",
        "    for index, letter in enumerate(name):\n",
        "        pos = all_letters.find(letter)\n",
        "        #print(pos)\n",
        "        rep[index][0][pos] = 1 # index corresponds to number of letters\n",
        "                               # 0 corresponds to batch\n",
        "                               # pos corresponds to character position in the one hot encoding\n",
        "\n",
        "    return rep"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SYtCtlVuP5Cr",
        "colab_type": "code",
        "outputId": "9677517f-d614-4896-e571-0c5c575361df",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 252
        }
      },
      "source": [
        "name_rep('Kim') # note 'i' represented by 1 at 8th index"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "tensor([[[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., 0., 0., 0., 0., 0., 0.,\n",
              "          0., 0., 1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
              "          0., 0., 0., 0., 0., 0.]],\n",
              "\n",
              "        [[0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 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.,\n",
              "          0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
              "          0., 0., 0., 0., 0., 0.]],\n",
              "\n",
              "        [[0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0., 0.,\n",
              "          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., 0., 0., 0., 0., 0., 0.,\n",
              "          0., 0., 0., 0., 0., 0.]]])"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 16
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PNmWX1UJzOSK",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vy5F60DKzId6",
        "colab_type": "text"
      },
      "source": [
        "<h4> Creating Vector Representation for Output </h4>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "t_PKPIRgQLjX",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def lang_rep(lang):\n",
        "  return torch.tensor([languages.index(lang)],dtype = torch.long)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "AP9BbDYIzf-b",
        "colab_type": "code",
        "outputId": "82b3f268-9f2d-4373-9edb-d55b329dc383",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "languages.index('English')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "12"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 18
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "3n9JJtkhQ6_6",
        "colab_type": "code",
        "outputId": "32451a7f-639e-4aa6-cb2a-68de8cb39322",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "lang_rep('English')"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "tensor([12])"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 19
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DFcOGhi4z6V_",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mCbLsLHm0Jph",
        "colab_type": "text"
      },
      "source": [
        "<h4> Data Visualisation </h4>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pxy_euHsQ9tE",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "count = {}\n",
        "for l in languages: \n",
        "    count[l] = 0\n",
        "for d in data:\n",
        "    count[d[1]] += 1 # d[0] is name, d[1] is language"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "bG2CMB5NRDrd",
        "colab_type": "code",
        "outputId": "4d28a626-935f-4d0c-ac44-9776a308b163",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 54
        }
      },
      "source": [
        "print(count)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "{'Portuguese': 74, 'Irish': 232, 'Spanish': 298, 'Vietnamese': 73, 'Chinese': 268, 'Greek': 203, 'Czech': 519, 'Dutch': 297, 'Japanese': 991, 'French': 277, 'German': 724, 'Scottish': 100, 'English': 3668, 'Russian': 9384, 'Polish': 139, 'Arabic': 2000, 'Korean': 94, 'Italian': 709}\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "62Qh37blRFNk",
        "colab_type": "code",
        "outputId": "f7d93619-7690-46cc-aa89-3971b6517e0e",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 315
        }
      },
      "source": [
        "plt_ = sns.barplot(list(count.keys()), list(count.values()))\n",
        "plt_.set_xticklabels(plt_.get_xticklabels(), rotation=90)\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX0AAAEqCAYAAAD58oAeAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n/H/Rr6yT/hfSHAAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ecpHfX1mKCwf",
        "colab_type": "text"
      },
      "source": [
        "The number of English , Russian and Arabic datapoints are very large compared to other languages, which may cause some problems while building the model\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "k3GyexCIRLRA",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7kTDHHkHfLtA",
        "colab_type": "text"
      },
      "source": [
        "<h3> RNN Model </h3>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ut9CTGqYZt2s",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "class RNN(nn.Module):\n",
        "  def __init__(self, input_size, hidden_size, output_size):\n",
        "    super(RNN,self).__init__() #calling init of parent class\n",
        "    self.hidden_size = hidden_size\n",
        "    self.rnn_cell = nn.RNN(input_size, hidden_size)\n",
        "    self.h2o = nn.Linear(hidden_size, output_size)                                                           \n",
        "    self.softmax = nn.LogSoftmax(dim = 1) \n",
        "\n",
        "  def forward(self,input_, hidden=None, batch_size = 1):\n",
        "    out, hidden = self.rnn_cell(input_,hidden) # hidden_shape => (1,batch_size, 128)\n",
        "    output = self.h2o(hidden.view(-1, self.hidden_size)) \n",
        "    # (1,batch_size , hidden_size) => (1 x batch_size, hidden_size)\n",
        "    output = self.softmax(output)\n",
        "    return output,hidden\n",
        "\n",
        "  def init_hidden(self, batch_size=1): # We write this function here to ensure size of hidden state is uniform throughout code\n",
        "    return torch.zeros(1,batch_size, self.hidden_size)# first index corresponds to number of characters\n",
        "                                                      # second index corresponds to number of words\n",
        "\n",
        "\n",
        "\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nz7OE-3HdsrW",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "n_hidden = 128\n",
        "net = RNN(n_letters, n_hidden, n_languages) "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "DevhVNy30c6e",
        "colab_type": "text"
      },
      "source": [
        "<h3> Inference </h3>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PHMxPtv2EAzu",
        "colab_type": "text"
      },
      "source": [
        "Batching wll not used in inference and evaluation, only in training"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "udK6L7LDf1kn",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#To see output of the untrained model for given name as input\n",
        "def infer(net, name):\n",
        "    net.eval() # We are not training the model\n",
        "    name_ohe = name_rep(name) # Convert input to combination of one hot vector representation\n",
        "    hidden = net.init_hidden() # Initialise hidden state\n",
        "    \n",
        "    #print(name_ohe.size()[0]) # Basically number of letters in input\n",
        "    \n",
        "    #No need of for loop as we are using nn.RNN - for i in range(name_ohe.size()[0]):\n",
        "    output, hidden = net(name_ohe, hidden)\n",
        "    \n",
        "    if type(hidden) is tuple: # For LSTM\n",
        "        hidden = hidden[0]\n",
        "    index = torch.argmax(hidden)\n",
        "    \n",
        "\n",
        "    return output"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "p4N77jF6XeMg",
        "colab_type": "code",
        "outputId": "aeb6aa3b-7060-4468-abbf-765b3259e6fe",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "name_rep('Adam').size()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "torch.Size([4, 1, 57])"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 26
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "PTdq6wlDW6D5",
        "colab_type": "code",
        "outputId": "46cfa242-ce8b-4738-a057-6393ecf471c1",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 67
        }
      },
      "source": [
        "output = infer(net, 'Adam')\n",
        "#index = torch.argmax(output)\n",
        "print(output)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "tensor([[-2.8581, -2.8505, -2.9604, -2.8565, -2.8383, -2.9333, -2.9288, -2.9208,\n",
            "         -2.9898, -2.7748, -2.8653, -2.8930, -2.9909, -3.0008, -2.8967, -2.9034,\n",
            "         -2.9315, -2.6877]], grad_fn=<LogSoftmaxBackward>)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Qi4rMVLV8Jr-",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "g6ojrXkvJRzt",
        "colab_type": "text"
      },
      "source": [
        "<h3> Creating Vector Representation for Batched Input </h3>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FtmK3lzP8KyT",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def batched_name_rep(names, max_word_size): # list of names and size of largest name\n",
        "    rep = torch.zeros(max_word_size, len(names), n_letters) # len(names) is the batch size , n_letters is 57\n",
        "    for name_index, name in enumerate(names):\n",
        "        for letter_index, letter in enumerate(name):\n",
        "            pos = all_letters.find(letter)\n",
        "            rep[letter_index][name_index][pos] = 1 # pad represented by all zeros (see picture in video)\n",
        "            # as for loop runs only till length of words, everything after that will be zer\n",
        "    return rep"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "frWygROW8K-D",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def print_char(name_reps): # visualize batching\n",
        "    print(name_reps.size())\n",
        "    name_reps = name_reps.view((-1, name_reps.size()[-1])) # not clear about -1\n",
        "    print(name_reps.size())\n",
        "    for t in name_reps: \n",
        "        if torch.sum(t) == 0:\n",
        "            print('<pad>')\n",
        "        else:\n",
        "            index = t.argmax()\n",
        "            print(all_letters[index])"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "0ZGC0UUc8LIu",
        "colab_type": "code",
        "outputId": "d2dce971-d25d-4307-89b6-ef161bf87a97",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 974
        }
      },
      "source": [
        "out_ = batched_name_rep(['Shyam', 'Ram'], 5)\n",
        "print(out_)\n",
        "print(out_.shape)\n",
        "print_char(out_)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "tensor([[[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., 0., 0., 0., 0., 0., 0.,\n",
            "          0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0., 0., 0., 0.,\n",
            "          0., 0., 0., 0., 0., 0.],\n",
            "         [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., 0., 0., 0., 0., 0., 0.,\n",
            "          0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0., 0., 0., 0., 0.,\n",
            "          0., 0., 0., 0., 0., 0.]],\n",
            "\n",
            "        [[0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 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.,\n",
            "          0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
            "          0., 0., 0., 0., 0., 0.],\n",
            "         [1., 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., 0., 0., 0., 0., 0., 0.,\n",
            "          0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
            "          0., 0., 0., 0., 0., 0.]],\n",
            "\n",
            "        [[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., 1., 0., 0., 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.,\n",
            "          0., 0., 0., 0., 0., 0.],\n",
            "         [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0., 0.,\n",
            "          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., 0., 0., 0., 0., 0., 0.,\n",
            "          0., 0., 0., 0., 0., 0.]],\n",
            "\n",
            "        [[1., 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., 0., 0., 0., 0., 0., 0.,\n",
            "          0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
            "          0., 0., 0., 0., 0., 0.],\n",
            "         [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., 0., 0., 0., 0., 0., 0.,\n",
            "          0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
            "          0., 0., 0., 0., 0., 0.]],\n",
            "\n",
            "        [[0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 1., 0., 0., 0., 0.,\n",
            "          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., 0., 0., 0., 0., 0., 0.,\n",
            "          0., 0., 0., 0., 0., 0.],\n",
            "         [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., 0., 0., 0., 0., 0., 0.,\n",
            "          0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,\n",
            "          0., 0., 0., 0., 0., 0.]]])\n",
            "torch.Size([5, 2, 57])\n",
            "torch.Size([5, 2, 57])\n",
            "torch.Size([10, 57])\n",
            "S\n",
            "R\n",
            "h\n",
            "a\n",
            "y\n",
            "m\n",
            "a\n",
            "<pad>\n",
            "m\n",
            "<pad>\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "ZPkyeNllJmB7",
        "colab_type": "text"
      },
      "source": [
        "<h3> Creating Vector Representation for Batched Output </h3>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RjjfZkS-33iq",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# Instead of representing language using a single number\n",
        "# we use array of numbers, with each number corresponding to\n",
        "# one item in the batch\n",
        "def batched_lang_rep(langs):\n",
        "    rep = torch.zeros([len(langs)], dtype=torch.long)\n",
        "    for index, lang in enumerate(langs):\n",
        "        rep[index] = languages.index(lang)\n",
        "    return rep"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Bsa3-sSu33xS",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "8JEDdVdDUt7k",
        "colab_type": "text"
      },
      "source": [
        "<h3> Custom DataLoader </h3>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "t55ayrrW9rBm",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def dataloader(npoints, X_,y_, verbose = False):\n",
        "  # Input X_ and y_ are lists\n",
        "\n",
        "  to_ret =[]\n",
        "  for i in range(npoints):\n",
        "    index = np.random.randint(len(X_))\n",
        "    name , lang = X_[index], y_[index]\n",
        "    to_ret.append((name,lang,name_rep(name), lang_rep(lang)))\n",
        "\n",
        "  return to_ret\n",
        "  \n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DdF1ebM5-AoY",
        "colab_type": "code",
        "outputId": "988d46d5-0880-4692-968c-9afd47bf937a",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "len(X_train)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "16040"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 33
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "nYa0lKLG34Bk",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def batched_dataloader(npoints, X_, y_, verbose=False ):\n",
        "    names = []\n",
        "    langs = []\n",
        "    X_lengths = [] # array containing lengths of each item in batch\n",
        "    \n",
        "    for i in range(npoints):\n",
        "        index_ = np.random.randint(len(X_))\n",
        "        name, lang = X_[index_], y_[index_]\n",
        "        X_lengths.append(len(name))\n",
        "        names.append(name)\n",
        "        langs.append(lang)\n",
        "    max_length = max(X_lengths)\n",
        "    \n",
        "    names_rep = batched_name_rep(names, max_length) # this function internally\n",
        "                                                    # does padding\n",
        "    langs_rep = batched_lang_rep(langs)\n",
        "    \n",
        "    padded_names_rep = torch.nn.utils.rnn.pack_padded_sequence(names_rep, X_lengths, enforce_sorted = False) \n",
        "    # does packing\n",
        "    \n",
        "    #visualize packing\n",
        "    if verbose:\n",
        "        print(names_rep.shape, padded_names_rep.data.shape)\n",
        "        print('--')\n",
        "    \n",
        "    if verbose:\n",
        "        print(names)\n",
        "        print_char(names_rep)\n",
        "        print('--')\n",
        "    \n",
        "    if verbose:\n",
        "        print_char(padded_names_rep.data)\n",
        "        print('Lang Rep', langs_rep.data)\n",
        "        print('Batch sizes', padded_names_rep.batch_sizes)\n",
        "    \n",
        "    \n",
        "    return padded_names_rep, langs_rep"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "7gJdfKCEI4Cn",
        "colab_type": "text"
      },
      "source": [
        "<h4> Visulize Padding and Packing </h4>"
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "GXG0YeXkMKvl",
        "colab_type": "text"
      },
      "source": [
        "[10, 3, 57] => Maximum length of word is 10, 3 such words, and each word is represented by 57 dimension vector\n",
        "\n",
        "[25,57] => Totally 27 letters = 10 + 9 + 8\n",
        "\n",
        "As soon as 'Aldoworth' is over, batch size reduces from 3 to 2\n",
        "\n",
        "As soon as 'Babintsev' is over, batch size reduces from 2 to 1\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "oTLnxaQu34V9",
        "colab_type": "code",
        "outputId": "fb37d527-4389-40a1-c1eb-26d645660a8d",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 823
        }
      },
      "source": [
        "p, l = batched_dataloader(3, X_train, y_train, True)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "torch.Size([7, 3, 57]) torch.Size([17, 57])\n",
            "--\n",
            "['Ilyahin', 'Kent', 'Jepson']\n",
            "torch.Size([7, 3, 57])\n",
            "torch.Size([21, 57])\n",
            "I\n",
            "K\n",
            "J\n",
            "l\n",
            "e\n",
            "e\n",
            "y\n",
            "n\n",
            "p\n",
            "a\n",
            "t\n",
            "s\n",
            "h\n",
            "<pad>\n",
            "o\n",
            "i\n",
            "<pad>\n",
            "n\n",
            "n\n",
            "<pad>\n",
            "<pad>\n",
            "--\n",
            "torch.Size([17, 57])\n",
            "torch.Size([17, 57])\n",
            "I\n",
            "J\n",
            "K\n",
            "l\n",
            "e\n",
            "e\n",
            "y\n",
            "p\n",
            "n\n",
            "a\n",
            "s\n",
            "t\n",
            "h\n",
            "o\n",
            "i\n",
            "n\n",
            "n\n",
            "Lang Rep tensor([13, 12, 12])\n",
            "Batch sizes tensor([3, 3, 3, 3, 2, 2, 1])\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cw6bRST08ZbW",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "49mtRqKx7mHi",
        "colab_type": "text"
      },
      "source": [
        "<h3> Custom Evaluation Function </h3>\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "QBTRABzVfVHb",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "# no batching\n",
        "def eval(net, n_points,k, X_, y_):\n",
        "\n",
        "  data_ = dataloader(n_points,X_, y_)\n",
        "  correct = 0 \n",
        "\n",
        "  for name, lang, name_rep, lang_rep  in data_:\n",
        "      output = infer(net, name) # forwards pass\n",
        "      val, indices = output.topk(k) # obtaining top k values and their corresponding indices in the 18 dim output\n",
        "                                    # (extension of torch.argmax)\n",
        "        \n",
        "      if lang_rep in indices: # if correct language in top k predictions\n",
        "          correct += 1\n",
        "            \n",
        "  accuracy = correct/n_points\n",
        "  return accuracy"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lkbOyXtn7si8",
        "colab_type": "code",
        "outputId": "9aca7a77-8485-419a-f2a3-84b5a3e48b26",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "eval(net, 1000, 1, X_test, y_test)\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.028"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 37
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "aqcHw4p464LK",
        "colab_type": "text"
      },
      "source": [
        "<h3> Train </h3>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "h2-GLZF5fUv7",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def train_batch(net, opt, criterion, n_points):\n",
        "    \n",
        "    opt.zero_grad()\n",
        "    loss = 0\n",
        "    \n",
        "    batch_input, batch_groundtruth = batched_dataloader(n_points, X_train, y_train, False)\n",
        "    \n",
        "    # for name, language, name_ohe, lang_rep in data_:\n",
        "\n",
        "    #     hidden = net.init_hidden()\n",
        "\n",
        "    #     for i in range(name_ohe.size()[0]): # forward pass\n",
        "    #         output, hidden = net(name_ohe[i], hidden) # Note that same hidden variable is used on lhs and rhs\n",
        "            \n",
        "    output, hidden = net(batch_input) # Note: Input automatically passed to forward\n",
        "                                      # as class is derived from nn.Module\n",
        "    \n",
        "    loss = criterion(output, batch_groundtruth)\n",
        "    loss.backward(retain_graph=True) \n",
        "        \n",
        "    opt.step()   # Entire backpropagation through time is done just in this one line\n",
        "        \n",
        "            \n",
        "    return loss"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "hHaIYyC1fUrN",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "net = RNN(n_letters, n_hidden, n_languages)\n",
        "criterion = nn.NLLLoss() #negative log likelihood\n",
        "opt = optim.SGD(net.parameters(), lr=0.01, momentum=0.9)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XijhQWiZfUly",
        "colab_type": "code",
        "outputId": "1c24898d-65fd-430f-ddcf-659ac5fde2bd",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 67
        }
      },
      "source": [
        "%%time \n",
        "train_batch(net, opt, criterion, 256)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "CPU times: user 56.3 ms, sys: 12.6 ms, total: 68.9 ms\n",
            "Wall time: 286 ms\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "tensor(2.8224, grad_fn=<NllLossBackward>)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 40
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "D3PA3ei-SXIV",
        "colab_type": "code",
        "outputId": "6f48def1-eb07-4c15-8007-717c292bf573",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "eval(net, 1000, 1, X_test, y_test)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "0.092"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 41
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Q1abhRtH7UxY",
        "colab_type": "text"
      },
      "source": [
        "<h3> Train Setup </h3>\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "TX0lJyxwSayZ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def train_setup(net, lr = 0.01, n_batches = 100, batch_size = 10, momentum = 0.9, display_freq=5):\n",
        "    \n",
        "    # Hyperparameters\n",
        "    criterion = nn.NLLLoss()\n",
        "    opt = optim.SGD(net.parameters(), lr=lr, momentum=momentum)\n",
        "    \n",
        "    loss_arr = np.zeros(n_batches + 1) #torch.zeros?\n",
        "    \n",
        "    for i in range(n_batches):\n",
        "        loss_arr[i+1] = (loss_arr[i]*i + train_batch(net, opt, criterion, batch_size))/(i + 1) \n",
        "        # Sum of all the previous losses\n",
        "        \n",
        "        if i%display_freq == display_freq-1:\n",
        "            clear_output(wait=True)\n",
        "            \n",
        "            print('Iteration', i, 'Top-1:', eval(net, len(X_test), 1, X_test, y_test), 'Top-2:', eval(net, len(X_test), 2, X_test, y_test), 'Loss', loss_arr[i])\n",
        "            plt.figure()\n",
        "            plt.plot(loss_arr[1:i], '-*')\n",
        "            plt.xlabel('Iteration')\n",
        "            plt.ylabel('Loss')\n",
        "            plt.show()\n",
        "            print('\\n\\n')\n",
        "                      "
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "-HbHRPIpLx0l",
        "colab_type": "text"
      },
      "source": [
        "<h3> Running the Model </h3>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VhxTvm4YWu5I",
        "colab_type": "code",
        "outputId": "d34ca3af-dd9d-41d5-957c-27415c0478ad",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 346
        }
      },
      "source": [
        "n_hidden = 128\n",
        "net = RNN(n_letters, n_hidden, n_languages)\n",
        "train_setup(net, lr=0.0005, n_batches=100, batch_size = 256)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Iteration 99 Top-1: 0.4628428927680798 Top-2: 0.5476309226932669 Loss 2.7673158645629883\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYwAAAEGCAYAAAB2EqL0AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nA4aZmZXEAcPMzErigGFmZiX5/7cgVH1BlDrbAAAAAElFTkSuQmCC\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n",
            "\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sQxxRXQlLhdo",
        "colab_type": "text"
      },
      "source": [
        "When lr was kept constant and batch size increased from 100 to 1000 accuracy did not improve but kept osciallting, possibly becuase learning rate was too small"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "w2PQo12HCvuu",
        "colab_type": "code",
        "outputId": "63c457fc-e76f-4fc8-afb6-a2052c001fa7",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 350
        }
      },
      "source": [
        "n_hidden = 128\n",
        "net = RNN(n_letters, n_hidden, n_languages)\n",
        "train_setup(net, lr=0.2, n_batches=1000, batch_size = 256)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Iteration 999 Top-1: 0.6625935162094763 Top-2: 0.7870324189526184 Loss 1.45095956325531\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYIAAAEKCAYAAAAfGVI8AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\njP4AAAAASUVORK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        },
        {
          "output_type": "stream",
          "text": [
            "\n",
            "\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "mFBZcaqqJNQj",
        "colab_type": "text"
      },
      "source": [
        "<h3> Confusion Matrix </h3"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "C5f0JEiHWyW0",
        "colab_type": "code",
        "outputId": "1f4756c1-3c63-4cee-d6b7-0bdff811418b",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "len(data)"
      ],
      "execution_count": 51,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "20050"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 51
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IhzffE09_Lpm",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def get_confusion_mat(net, n_points,k, X_, y_):\n",
        "\n",
        "  cfm = np.zeros((n_languages,n_languages))\n",
        "\n",
        "  data_ = dataloader(n_points,X_, y_)\n",
        "   \n",
        "\n",
        "  for name, lang, name_rep, lang_rep  in data_:\n",
        "      output = infer(net, name) # forwards pass\n",
        "      val, index = output.topk(1) # obtaining top value and their corresponding indices in the 18 dim output\n",
        "                                    # (extension of torch.argmax)\n",
        "        \n",
        "      cfm[languages.index(lang)][index] += 1    \n",
        "\n",
        "  return cfm  \n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "qhGM8p27Axhg",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "cfm = get_confusion_mat(net, 1000, 1, X_test, y_test)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "EHhDjXSLBNfh",
        "colab_type": "code",
        "outputId": "ab47019c-417e-4bce-c2b5-595cd93e63e7",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 319
        }
      },
      "source": [
        "# Per class accuracy\n",
        "for lang_ind in range(18):\n",
        "  total = np.sum(cfm[lang_ind])\n",
        "  correct = cfm[lang_ind][lang_ind]\n",
        "  print(languages[lang_ind],\"---Total:\",total,\"--Accuracy :\", correct /total)\n"
      ],
      "execution_count": 56,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Portuguese ---Total: 3.0 --Accuracy : 0.0\n",
            "Irish ---Total: 14.0 --Accuracy : 0.0\n",
            "Spanish ---Total: 12.0 --Accuracy : 0.0\n",
            "Vietnamese ---Total: 5.0 --Accuracy : 0.0\n",
            "Chinese ---Total: 10.0 --Accuracy : 0.5\n",
            "Greek ---Total: 12.0 --Accuracy : 0.3333333333333333\n",
            "Czech ---Total: 33.0 --Accuracy : 0.030303030303030304\n",
            "Dutch ---Total: 14.0 --Accuracy : 0.0\n",
            "Japanese ---Total: 47.0 --Accuracy : 0.46808510638297873\n",
            "French ---Total: 23.0 --Accuracy : 0.0\n",
            "German ---Total: 24.0 --Accuracy : 0.16666666666666666\n",
            "Scottish ---Total: 3.0 --Accuracy : 0.0\n",
            "English ---Total: 190.0 --Accuracy : 0.7736842105263158\n",
            "Russian ---Total: 478.0 --Accuracy : 0.8765690376569037\n",
            "Polish ---Total: 3.0 --Accuracy : 0.0\n",
            "Arabic ---Total: 104.0 --Accuracy : 0.6153846153846154\n",
            "Korean ---Total: 5.0 --Accuracy : 0.0\n",
            "Italian ---Total: 20.0 --Accuracy : 0.45\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "ZuLGkeDmBlno",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "gXe8rxv1G3OJ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import seaborn as sn\n",
        "import pandas as pd"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "v0r1f7zQCFLQ",
        "colab_type": "code",
        "outputId": "6a3d1322-9af7-4474-dc88-3ce913d16431",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 431
        }
      },
      "source": [
        "df_cm = pd.DataFrame(cfm, range(18),\n",
        "                  range(18))\n",
        "plt.figure(figsize = (10,7))\n",
        "sn.set(font_scale=1)#for label size\n",
        "sn.heatmap(cfm, annot=True,annot_kws={\"size\": 5})# font size\n",
        "\n",
        "plt.show()\n"
      ],
      "execution_count": 58,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAi0AAAGeCAYAAABYXzE0AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 720x504 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "smsDuPEaG4sn",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}