LRCN_(CNN_LSTM)_two_digit_mnist.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "LRCN (CNN-LSTM) - two-digit-mnist.ipynb",
      "provenance": [],
      "collapsed_sections": []
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "SE8ipjHXO_F3",
        "colab_type": "text"
      },
      "source": [
        "### Import libraries"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lUQlzSWNIZKe",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import random\n",
        "\n",
        "import numpy as np\n",
        "from skimage.transform import resize\n",
        "\n",
        "from keras.datasets import mnist\n",
        "from keras.utils import np_utils\n",
        "\n",
        "import matplotlib.pyplot as plt\n",
        "plt.style.use('ggplot')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sBf6yylnPJaG",
        "colab_type": "text"
      },
      "source": [
        "### Load MNIST"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "J9HpF25SIj8o",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 53
        },
        "outputId": "7fc9d839-35c8-4e2f-bd97-e36c9cc9d6ee"
      },
      "source": [
        "\n",
        "mnist_img_height , mnist_img_width = 28 , 28\n",
        "(x_train,y_train), (x_test, y_test) = mnist.load_data()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Downloading data from https://s3.amazonaws.com/img-datasets/mnist.npz\n",
            "11493376/11490434 [==============================] - 2s 0us/step\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "GaaXaeD8IoWH",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 53
        },
        "outputId": "4c74e0d4-019f-4a1f-d31c-bfbb0222e139"
      },
      "source": [
        "print(f\"Number of train images:{x_train.shape[0]}\")\n",
        "print(f\"Number of test images:{x_test.shape[0]}\")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Number of train images:60000\n",
            "Number of test images:10000\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "gi4MH0t_JMPQ",
        "colab_type": "text"
      },
      "source": [
        "### Build two-digit mnist sequences "
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IEo7zjLwJOVf",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "def build_sequence_mnist(data,labels,dataset_size,IS_TRAIN=True):\n",
        "    \n",
        "    # sequence data size\n",
        "    seq_img_height = 64\n",
        "    seq_img_width = 64\n",
        "    \n",
        "    seq_data = np.ndarray(shape=(dataset_size,seq_img_height,seq_img_width),\n",
        "                           dtype=np.float32)\n",
        "    seq_labels = [] \n",
        "    \n",
        "    for i in range(0,dataset_size):\n",
        "        \n",
        "        #Pick a random number of digits to be in the dataset\n",
        "        # num_digits = random.randint(1,2)\n",
        "        \n",
        "        s_indices = [random.randint(0,len(data)-1) for p in range(0,2)]\n",
        "\n",
        "        if IS_TRAIN:\n",
        "          # concatenating images and labels together\n",
        "          new_image = np.hstack([x_train[index] for index in s_indices])\n",
        "          new_label =  [y_train[index] for index in s_indices]\n",
        "        else:\n",
        "\n",
        "          new_image = np.hstack([x_test[index] for index in s_indices])\n",
        "          new_label =  [y_test[index] for index in s_indices]\n",
        "        \n",
        "        \n",
        "        #Resize image\n",
        "        new_image = resize(new_image,(seq_img_height,seq_img_width))\n",
        "        \n",
        "        seq_data[i,:,:] = new_image\n",
        "        seq_labels.append(tuple(new_label))\n",
        "        \n",
        "    \n",
        "    #Return the synthetic dataset\n",
        "    return seq_data,seq_labels\n"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "J7-CO7tXKyOQ",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "x_seq_train,y_seq_train = build_sequence_mnist(x_train,y_train,60000)\n",
        "x_seq_test,y_seq_test = build_sequence_mnist(x_test,y_test,10000,IS_TRAIN=False)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "IOE49QEmLzCZ",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 282
        },
        "outputId": "5e3c4433-9d50-4bdc-a76a-2c1bf5df72f7"
      },
      "source": [
        "RANDOM_IMG_TO_SHOW = 2\n",
        "plt.figure()\n",
        "plt.imshow(x_seq_test[RANDOM_IMG_TO_SHOW], cmap='gray')\n",
        "plt.title(\"Labeled as: {}\".format(''.join(str(i) for i in y_seq_test[RANDOM_IMG_TO_SHOW])))\n",
        "plt.grid(b=None)\n",
        "plt.show()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAEJCAYAAABSX1EAAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3dfVCVVR4H8O+FDMT08pKhYKComaJlCb6EdlVu2mrTkLM5k2K5rdP2sma1trFbi+1WI7NKMFM2bk2DZdu22+xI2dRo1xfU1EKBNFQMF19aVF5FroAI9+wfLqfzPAgX4d7Ly/l+Zpw5cB7uPQI/nnOec87vWIQQAkTU5/l1dwOIyDcY7ESaYLATaYLBTqQJBjuRJhjsRJpgsPcwJ0+ehMViwZ49e3rE6wCAxWLBRx991OXXoe7FYPewpUuXwm63d3cz+qyKigo89dRTiIiIQEBAAEaMGIH33ntP1hcWFuLhhx/G6NGj4efnh2XLlnVja3uWG7q7AUQd5XQ6ce+99yIyMhL/+Mc/EB0djbNnz6K5uVleU1dXh6ioKDz44IN48803u7G1PQ/v7D728ccfY8qUKbBarbj55psxf/58HD9+vNV1J0+eRGJiIvr374+YmBh88sknhvrz589j6dKlGDx4MAYOHIiEhATs2rWr3ffuyNfs2LEDd9xxBwIDA3HHHXdgx44dbv9PJSUlWLBgASIiIhAUFIQJEyZg48aNhmv27NmDhIQEDBw4EAMHDsSdd96JLVu2uH1t1Zo1a1BXV4cvvvgCNpsNw4cPx7Rp0zB9+nR5TXx8PNLT07FkyRJYrdbrev2+jsHuY5cvX8Yrr7yCvLw8fP311/D398f8+fPR2NhouO73v/89Hn/8cRQUFGDRokVYvHgx8vPzAQD19fWYNWsWamtr8dVXXyE/Px/z5s3Dfffdh6NHj17zfTvyNaWlpXjggQcwadIk5OXlIT09HStWrHD7f3I6nZg9eza++uorHD58GE888QR+9atfyT8UTU1NePDBBzFlyhTk5eUhLy8Pr776KoKCguRrzJw5EzNnzmz3ff79739j+vTpeP755zF06FDcfvvtePHFF1FXV+e2jQRAkEc99thjIjExscPXV1ZWCgBiz549QgghSkpKBADxyiuvGK6bNm2aSE5OFkIIkZWVJSIjI8WVK1cM18yaNUusWLHC8Dq7d+/u8Ne8/PLLIioqynDN5s2bBQCxcePGDv+fhBDiwQcfFMuWLRNCCFFVVSUAiB07drR5/ZIlS8SSJUvafc3AwEAREBAgFi9eLHJzc8Vnn30mbr31VrFo0aJrXm+z2cSvf/3r62p3X8Yxu48VFBTgz3/+MwoKClBRUQHx/31Ip06dQkJCgrxu2rRphq9LSEjAtm3bAAC5ubk4d+4cgoODDddcvnwZ/fv3v+b7duRrjhw5gsmTJ+OGG37+tVC7yG2pq6vDX/7yF2zevBlnz55FY2MjLl++jFmzZgEAQkJCsGzZMsydOxezZ8+GzWbDQw89hDFjxsjX+PDDD92+j8vlQlhYGLKystCvXz8AQGNjIx5++GG89dZbCA0NdfsaOmOw+1BdXR3mzJmD6dOnIysrC+Hh4QCA2NjYVt349rhcLowdOxabNm1qVad2jbv6NR314osv4rPPPsObb76JMWPGYMCAAfjd736Hmpoaec17772HFStWYOvWrfj666/xpz/9CW+//TZ+85vfdPh9hg4diuHDh8tAB65+74CrfywZ7O1jsPvQ0aNHUV5ejjfeeANjx44FAOzdu1fe3VX79+/HvHnz5Md79+7FuHHjAABxcXH48MMPMWjQINxyyy0deu+OfM24ceOwceNGNDc3w9/fHwDwzTffuH3tXbt2YfHixVi4cCGAq39Yjh8/Lv+YtRg/fjzGjx+PF154AU8++STefffd6wr2GTNmYMeOHWhqapK9j6KiIgDA8OHDO/w6uuIDOi9wOp0oKCgw/Dt27Biio6MREBCAt956CydOnMC2bduwYsUKWCyWVq/x/vvv4+OPP8bx48eRmpqKffv24YUXXgAALF68GCNGjMD8+fOxdetWnDx5Et9++y1Wr16N7Ozsa7apI1/z1FNPoby8HE888QSOHj2Kbdu24eWXX3b7/x0zZgw+++wzfPfddzhy5AieeOIJlJaWyvri4mK89NJL2LNnD06dOoV9+/Zh9+7d8o8XADz66KN49NFH232flStXory8HE8//TSOHTuGHTt2YOXKlXj00UcREhIC4Gq3vuV77nQ6UVVVhYKCAhw5csTt/6PP6+6HBn3NY489JgC0+jdmzBghhBCffvqpGDVqlAgICBATJ04UO3fuFP7+/iIrK0sI8fODtQ8//FDYbDYREBAghg8fLv7+978b3qeiokI8+eSTIiIiQvTr109ERESIpKQkkZeXZ3idlgd0HfkaIYRwOBxi/Pjx4sYbbxSxsbFi27Ztbh/QnT59WsyZM0cEBQWJIUOGiNTUVPH4448Lm80mhBCitLRUPPTQQyIyMlLceOONYujQoWLZsmXiwoUL8jVsNpu8vj0Oh0PExcWJgIAAER0dLVauXCkuXbok61v+3+Z/0dHRbl+7r7MIwUw1RDpgN55IEwx2Ik0w2Ik00aWpt4KCAmRlZcHlciExMRFJSUmeahcReVin7+wulwvvv/8+/vjHPyIjIwPffPMNfvrpJ0+2jYg8qNN39uLiYgwZMkQunLjnnnuQm5uLYcOGtft115pTJiLPaG9yrdPBXlVVhbCwMPlxWFgYfvzxx1bXORwOOBwOAEBaWlpn346Iusjry2XtdjsztxD1AJ0es4eGhqKyslJ+XFlZyY0IRD1Yp4N95MiROHv2LMrKytDU1IS9e/ciLi7Ok20jIg/q0nLZvLw8fPDBB3C5XJg1axYWLFjg/g35gI7Ia9oLZ5+vjWewE3lPe+HMFXREmmCwE2mCmWq8QM3h1pLxpYWa47ypqclnbSLinZ1IEwx2Ik0w2Ik0wTG7B6ipjQEYlgffdttthroffvhBlnNzcw11Fy9e9ELriK7inZ1IEwx2Ik2wG+8BN954o+HjuXPnyvKcOXMMdZ9++qksHzt2zFDHbjx5E+/sRJpgsBNpgt34TlJXxo0aNcpQp55OOnToUENdYGCgdxtG1Abe2Yk0wWAn0gSDnUgTHLN3kjpmHz9+vKEuOjr6mtcBQENDgyxz1xv5Eu/sRJpgsBNpgt3466Dmz1On0GJjYw3XqSm1z549a6hTP758+bKnm0g9hDnXojoFa65TU7KrwzxP452dSBMMdiJNMNiJNMEx+3Xw8/v5b+PAgQNl2TxmV3fBfffdd4Y6NXlFfX29p5tIPYSadBSA4bQkc7KTffv2yXJpaanX2sQ7O5Em3N7Z33nnHeTl5cFqtSI9PR0A4HQ6kZGRgfLycgwePBjPP/88brrpJq83log6z22wz5w5E/fffz/WrVsnP5ednY0JEyYgKSkJ2dnZyM7ORnJyslcb2hOoXbPw8HBZHj16tOE6tZt26NAhQ92JEydkmVNvfYs6pWZOaHL33XfLsnnn49GjR2W5W7vx48aNa3XXzs3Nhc1mAwDYbLZWiROJqOfp1AO6mpoahISEAACCg4NRU1PT5rUOhwMOhwMAkJaW1pm3IyIP6PLTeIvF0u7JrHa73ZBauTdTn8Dfe++9stzyh6+FepJmeXm5oc7pdHqpddQd1BkateseHBxsuE7dLBUQEGCoGzRokJdaZ9Spp/FWqxXV1dUAgOrqap81log6r1PBHhcXh5ycHABATk4O4uPjPdooIvI8t934zMxMHDlyBLW1tXjyySexcOFCJCUlISMjA9u3b5dTb0TUs7kN9ueee+6an09NTfV4Y3o6dVaiZTYCaD1mb2xslOWKigpDXW1trZdaR92hrZ2Q5jH7hAkTZNk8ZlefBXkTV9ARaYLBTqQJboRpR1BQkOHjESNGyPLIkSNl2bzp4fDhw7LcMmtBfYN5mnnIkCGyPG3aNFk2nyWgduvNvxO+ykXIOzuRJhjsRJpgsBNpgmP2dgwYMMDwsToOU8fv6pJJwLiLqb19A9T7mH/WaiLJX/ziF7KsTrUBxiWxP/30k6HuypUrnmxim3hnJ9IEg51IE+zGm6jTaBEREYa6u+66S5bVabmysjLDdd9//70sc+qtbzEf5xUWFibL48aNk2Xz5jB1OJefn2+o89XvCO/sRJpgsBNpgt14E3Uzg/rEHTCukFJXPal55QAgLy9Pli9cuODpJlI3Mq+WVDdBqU/mT506ZbhO/T1oydzU4ty5c55sYpt4ZyfSBIOdSBMMdiJNcMxuok6ZmHcuqTvd1KQU6hgdAH788UdZvnTpkqebSF5m3tmmJpIcPHiwoS4qKkqW1aO61d8BADh9+rQsm6fefPVch3d2Ik0w2Ik0oX033rwiSp0+ue222wx16oaFAwcOyPKWLVsM13HzS+9m/p245ZZbZHny5MmGOjWzsnrsl5rUAoDMxgy0zkPocrk639jrwDs7kSYY7ESaYLATaYJjdtP4LDIyUpbbG7Or0yfm5Y8NDQ2ebCL5mPm4ZfU5jvn0o0mTJsmyOmZXd8MBxoQm5jG7ejagN/HOTqQJt3f2iooKrFu3DhcuXIDFYoHdbse8efPgdDqRkZGB8vJyeQSU+Rx3Iuo53Aa7v78/lixZgpiYGNTX1yMlJQV33HEHdu7ciQkTJiApKQnZ2dnIzs5GcnKyL9rsUeqqJ8C4Sm7YsGGGuubmZllWEw74qhtG3qPuZrv77rsNdeqR43FxcYY69eimyspKWd66davhOnVnpHo8mC+57caHhIQgJiYGANC/f39ERkaiqqoKubm58rwzm82G3Nxc77aUiLrkuh7QlZWVoaSkBKNGjUJNTY3cyxscHNzmQhKHwyEfYKWlpXWxuUTUWR0O9oaGBqSnp2Pp0qWtjkWyWCytNg+0sNvthm4QEXWPDgV7U1MT0tPTMWPGDEyZMgUAYLVaUV1djZCQEFRXV7dKsNdbmMfs6k63W2+91VB3/vx5WVZ3KnHM3vup02bmcfm8efNkeezYsW1+XUlJiSx//fXXhuuKi4tluceO2YUQWL9+PSIjI/HAAw/Iz8fFxcn1vjk5Oa3mH4moZ3F7Zy8qKsKuXbsQFRWFF198EQDwyCOPICkpCRkZGdi+fbuceiOinsttsN9+++3417/+dc261NRUjzfI18zd+PDwcFk2JxdUE1GoXTZf7VoizzGvnFSTi6r53wHjFKyakBQA/vvf/8ryzp07ZVldMQcA9fX1stxdwz6uoCPSBIOdSBPab4S5+eabDR+rGxjUFXMAcPHiRVlWc4yZr6OezzxEGzNmzDXLwNV1JC3MJ67+5z//keXPP//8mp8HesZQj3d2Ik0w2Ik0wWAn0oSWY/YBAwbIsnlnm5pc0Jxk4Pjx47Ks5o3nCrqey8/v5/uZ+nOPjo42XKfudDOvnFSXgpvPZSssLJTlQ4cOyXJPPOOPd3YiTTDYiTShZTdeXSVnzjOn1qkbXwBg165dsszptt5BnWKLiIiQ5fvvv99w3YwZM2TZvKrS6XTK8pEjRwx13333nSz39PMCeGcn0gSDnUgTDHYiTWg5ZlfHbmqCScC4XNY8Pjt48KAse3v5o7orS02QAFxNJnKtckdfDzBOSZmXgPYl6vdu+PDhstySP7HFXXfdJcvmvPFnzpyR5e+//95Qp063qWP7noh3diJNMNiJNKFlN3706NGybD5aV2Xu3qq73rxB7WpbrVZZjoqKMlynruJSV/IBbXfrWzIBt1BXk506der6G9tLqP9PNXWaefgWEBAgy+Yda+rxXrt37zbUmadnezLe2Yk0wWAn0oSW3Xh11dz1dOPNG2M8Te3GqwkTzMkU1K66egyVuU6lvh5gTNrRl7vx6hkHkydPluWWU45aqE/g1fyCgLEbr66iBIC6ujqPtNMXeGcn0gSDnUgTDHYiTWgzZld3s0VGRsqyOjUDAGfPnpVl8wo6T4/P1DYBwMSJE2VZTaZgniZSj6j65z//aahTx9/qMwdvP2/oKcznEKr54NWkFOoKQsCY/72goMBQp35PL1++bKjrTYlLeGcn0oTbO3tjYyNWrVqFpqYmNDc3Y+rUqVi4cCHKysqQmZmJ2tpaxMTEYPny5a3S8xJRz+E2Ovv164dVq1YhMDAQTU1NSE1NxcSJE/HFF19g/vz5SEhIwLvvvovt27djzpw5vmhzpwwdOlSW1Y0w/fv3N1yn5oM/fPiwoU49wsfTbQKAuXPnXrNsPiFXXTW3f/9+Q506DFG78ebECr1pyuh6mIdl6hBIzTdo7n6fPn1alvPz8w11paWlstybk5a47cZbLBZ5vlVzczOam5thsVhQWFiIqVOnAgBmzpyJ3Nxc77aUiLqkQ/1ul8uFl156CefOncPcuXMRHh6OoKAguQgkNDQUVVVV1/xah8MhFyWkpaV5qNlEdL06FOx+fn5Ys2YNLl26hLVr1xq6Ne7Y7XbY7fZON5CIPOO6nqgNGDAAsbGxOH78OOrq6tDc3Ax/f39UVVW1StLX06jjNXXpqHkMpk6zFBUVGeo6M14zT/GoeenvvPNOQ5069aaONc3JFNQdcYMHDzbUmRNdtGhsbDR83JcSVqjfYzX5CACMHz9eltVpOXWMDhiffRQXFxvq1KQUvWmqzcztmP3ixYvyXPLGxkYcOnQIkZGRiI2Nld+gnTt3Ii4uzrstJaIucXtnr66uxrp16+ByuSCEwLRp0zBp0iQMGzYMmZmZ+OSTTzBixAjMnj3bF+0lok5yG+zR0dH461//2urz4eHhWL16tVca5Q3qqjm1O2eegjp58qQsm7tznenCtXc0cEJCgqFOTarR3pqFltkRoPVuNnOXv4U5Z556pFFvp36vzNOZ6tBI/T+bp1W3bt0qy+Yufkfz/PV0XEFHpAkGO5EmtFnfqj4FV1fNmZ9Sl5eXy3JZWVmX39f8tHzChAmybH4ar3ZBzamfVeoTd/MKwPa+TtWbnyqbqcMadbgGtN5E1OLYsWOGj9XNL22tGenteGcn0gSDnUgTDHYiTWgzZh84cKAse3srrjpuVpNQAMbpNnVVH9C3psN8Sc2Jrx7xBBh3DKorIM1JNs3Pbvoi3tmJNMFgJ9KENt34m266SZbVVWbtrSwzd6s7Ol2lHiU0adIkQ92UKVNkWU2iQZ3XXjdeHb6pG1rMq+TYjSeiPoPBTqQJBjuRJrQZs7fFnGdczeVuToRgPh65hXkqz2azybJ5SayaeKKz1GcH5oQafWkZbFvUZyKAcbeguhzZTH0+Y8621JeSebSFd3YiTTDYiTShfTfenGdc3R1nzqvX0W78vffeK8vmbrw5B3xHmLvm7XXjzVOJfZG5G6/ubFNzzgHG6VP1+8ZuPBH1WQx2Ik1o3403p3pWn86rq+7M1IQJ6qo4wNiVNA8FOrMJx9yNV5MrmLujDQ0N1/36vY26Yg5oOzEJYFwZpx6N1ZtPY+0s3tmJNMFgJ9IEg51IE9qP2c3UHXHm8Z86jaOO7c1n2Y0dO/aa13WWeTypJsU8c+aMoa6vHsWsMo/Z2zt6TB2zqwkrdJiiNOOdnUgTHb6zu1wupKSkIDQ0FCkpKSgrK0NmZiZqa2sRExOD5cuXez3dExF1Xoej88svv0RkZCTq6+sBAB999BHmz5+PhIQEvPvuu9i+fTvmzJnjtYZ6izlBhdpFVLvjAOQBl4Bxk8zkyZMN16mbado6VdUd9cihyspKQ11+fr4sm3Pb95WjiszU72NUVJShznzkk0pdGXfu3DlZZje+DZWVlcjLy0NiYiKAq2PIwsJCTJ06FQAwc+ZM5Obmeq+VRNRlHbqzb9iwAcnJyfKuXltbi6CgIJlFNTQ0tM1TNBwOBxwOBwAgLS3NE20mok5wG+wHDx6E1WpFTEwMCgsLr/sN7HZ7q6fVROR7boO9qKgIBw4cQH5+PhobG1FfX48NGzagrq4Ozc3N8Pf3R1VVVbvTH72Jmst93rx5hjp1DK8mMrztttsM15l3ZXWGuuzVfC7Zl19+Kcs1NTVdfq/eQF26bP5+R0dHt/l16q7ACxcuyLIOy2PN3Ab7okWLsGjRIgBAYWEhNm/ejGeffRZvvvkm9u/fj4SEBOzcuRNxcXFebywRdV6n59kXL16ML774AsuXL4fT6cTs2bM92S4i8rDrmhiPjY1FbGwsgKvTS6tXr/ZKo7xB7ba114VTp82mT59uqFN3s6lddfMxTuaddG0xT/+oq73URBmHDx82XLd7925ZVqcD+zI16Yf5GOb28u+rU2/qNKWO3XiuoCPSBIOdSBParG9Vu3PtdeHUE1jb2wijrujq7Omr5kQTJ0+elOWcnBxZ3rJli+E69Qm8LivBBg8eLMs333yzoU7dvGTW1qyGOXefDnhnJ9IEg51IEwx2Ik1oOWZvb5yrTpuZc8qriSjaO9q5o8xJD0+cOCHLX331lSybx+zqzjZdppCGDBkiy50dsx89elSWOWYnoj6LwU6kCW268Xl5ebLcsg8fAIKDgw3XtZczrjPd9draWsPHah60goICQ93evXtlWe3S63A0kTvq8Mq8QlH9uZiP6Dpw4IAsq9uwdRn+qHhnJ9IEg51IEwx2Ik1oM2ZXc+QVFxfL8u233264zhN53lXm5BLqOP3zzz831H377beybE4yST9r79mJecy+b98+Wa6urpZljtmJqM9isBNpQptu/Pnz52X54MGDsmxOfDBu3DhZVo8CBow73dTVb+pxTIBxhdv3339vqNu/f78sm5NS/PTTT7Ksy262jrJarbJs3o2oMifzOH36tCzrPoXJOzuRJhjsRJrQphuvbojYs2ePLJvPp1Of0qp5zwBjYgu1u2juqqtdfPMmFvXpsHocEcCue3vUIZX556JqOcikBWc1fsY7O5EmGOxEmmCwE2lCmzG7Ou1SVFQky+ajmtSjmNXpHsB4BJG6g0qdygOAuro6WVbH6Ob31n0q6HqoPxfzKkf1OYs5KYXT6fRuw3oR3tmJNNGhO/szzzyDwMBA+Pn5wd/fH2lpaXA6ncjIyEB5eTkGDx6M559/3nDnI6KepcPd+FWrVhmmPLKzszFhwgQkJSUhOzsb2dnZSE5O9kojPU3tPp85c8ZQt23bNllWE0gAxlxn6hRPSUmJ4Tp1BZ26cg/QM/eZJ6j5ANtbQUdt63Q3Pjc3FzabDQBgs9kMu8qIqOfp8J39jTfeAADcd999sNvtqKmpQUhICICrqZ3aOifc4XDA4XAAANLS0rraXiLqpA4F+2uvvYbQ0FDU1NTg9ddfb7V5xGKxtLnH2G63w263d72lRNQlHQr20NBQAFenouLj41FcXAyr1Yrq6mqEhISgurq63SWMPY06pi4tLTXUmT+mnkEdp7eXJ9580zEvh9aZ2zF7Q0ODfBjV0NCAQ4cOISoqCnFxcfLwwZycHMTHx3u3pUTUJW7/7NXU1GDt2rUArj5Jnj59OiZOnIiRI0ciIyMD27dvl1NvRNRzuQ328PBwrFmzptXnBw4ciNTUVK80ishMfQCsrlAEjCsbW4acLSZNmiTLhYWFsqzj6kWuoCPSBIOdSBMMdiJNcF6CeoX2xuwq83HOd999tyx/8sknsswxOxH1WQx2Ik2wG0+9gppj33xsVnh4uCybj8j+4YcfZFn3HYe8sxNpgsFOpAmL8PFxlu2dwEnUFnWnpflYLjWPoLrJCQAuXLggy2qSkb6ao7+9cOadnUgTDHYiTTDYiTTBMTtRH8IxOxEx2Il0wWAn0gSDnUgTDHYiTTDYiTTBYCfSBIOdSBMMdiJNMNiJNMFgJ9JEh9JSXbp0CevXr8eZM2dgsVjw1FNPISIiAhkZGSgvL5fHP6kncxBRz9KhjTBvv/02xo4di8TERDQ1NeHy5cvYtGkTbrrpJiQlJSE7OxtOpxPJycnu35AbYYi8pksbYerq6nD06FHMnj0bwNUjcAcMGIDc3FzYbDYAgM1mQ25uroeaS0Te4LYbX1ZWhkGDBuGdd97BqVOnEBMTg6VLl6KmpgYhISEAgODgYEMSf5XD4YDD4QAApKWlebDpRHQ93AZ7c3MzSkpK8Pjjj2P06NHIyspCdna24RqLxdJm99xut8Nut3umtUTUaW678WFhYQgLC8Po0aMBAFOnTkVJSQmsViuqq6sBANXV1Rg0aJB3W0pEXeI22IODgxEWFobS0lIAV5P1Dxs2DHFxccjJyQEA5OTkID4+3rstJaIu6dDT+JMnT2L9+vVoamrCLbfcgqeffhpCCGRkZKCiouK6pt74NJ7Ie9oLZ+agI+pDmIOOiBjsRLpgsBNpgsFOpAkGO5EmGOxEmujQFldP8vFMHxH9X7fc2VNSUrrjbXtcGwC2w4ztMPJkO9iNJ9IEg51IE/6vvvrqq93xxjExMd3xtj2uDQDbYcZ2GHmqHT5fG09E3YPdeCJNMNiJNOHTefaCggJkZWXB5XIhMTERSUlJPnnfd955B3l5ebBarUhPTwcAOJ1On6fCrqiowLp163DhwgVYLBbY7XbMmzfP521pbGzEqlWr0NTUhObmZkydOhULFy5EWVkZMjMzUVtbi5iYGCxfvhw33OD9XxGXy4WUlBSEhoYiJSWlW9rxzDPPIDAwEH5+fvD390daWlq3/I54NW278JHm5mbx29/+Vpw7d05cuXJFrFy5Upw5c8Yn711YWChOnDghXnjhBfm5jRs3ik2bNgkhhNi0aZPYuHGj19tRVVUlTpw4IYQQoq6uTjz77LPizJkzPm+Ly+US9fX1Qgghrly5Iv7whz+IoqIikZ6eLvbs2SOEEOJvf/ub2LJli1fb0WLz5s0iMzNTrF69WgghuqUdTz/9tKipqTF8rjt+R9566y3hcDiEEFd/Nk6n02Pt8Fk3vri4GEOGDEF4eDhuuOEG3HPPPT5LPz1u3LhWfwm7IxV2SEiIfLLav39/REZGoqqqyudtsVgsCAwMBHA1oWhzczMsFgsKCwsxdepUAMDMmTN98j2prKxEXl4eEhMTAVxdYdkd7bgWX/9cvJ223Wfd+KqqKoSFhcmPw8LC8OOPP/rq7VvpaCpsbykrK0NJSQlGjRrVLW1xuVx46aWXcO7cOcydOxfh4WhGhcIAAAJdSURBVOEICgqCv78/ACA0NBRVVVVeb8eGDRuQnJyM+vp6AEBtbW23tAMA3njjDQDAfffdB7vd7vOfS1fTtrvj87XxPVF7qbC9oaGhAenp6Vi6dCmCgoK6pS1+fn5Ys2YNLl26hLVr18qEor508OBBWK1WxMTEoLCw0Ofvr3rttdcQGhqKmpoavP7664iIiDDU++Ln0tW07e74LNhDQ0NRWVkpP66srERoaKiv3r6VllTYISEhPk2F3dTUhPT0dMyYMQNTpkzp1rYAwIABAxAbG4vjx4+jrq4Ozc3N8Pf3R1VVldd/PkVFRThw4ADy8/PR2NiI+vp6bNiwweftACDfw2q1Ij4+HsXFxT7/uVwrbXt2drbH2uGzMfvIkSNx9uxZlJWVoampCXv37kVcXJyv3r6V7kiFLYTA+vXrERkZiQceeKDb2nLx4kVcunQJwNUn84cOHUJkZCRiY2Oxf/9+AMDOnTu9/vNZtGgR1q9fj3Xr1uG5557D+PHj8eyzz/q8HQ0NDXIY0dDQgEOHDiEqKsrnPxdvp2336Qq6vLw8fPDBB3C5XJg1axYWLFjgk/fNzMzEkSNHUFtbC6vVioULFyI+Pr5TqbC74tixY0hNTUVUVJTsij3yyCMYPXq0T9ty6tQprFu3Di6XC0IITJs2Db/85S9x/vx5ZGZmwul0YsSIEVi+fDn69evntXaoCgsLsXnzZqSkpPi8HefPn8fatWsBXO1KT58+HQsWLEBtba3Pf0c8mbbdjMtliTTBFXREmmCwE2mCwU6kCQY7kSYY7ESaYLATaYLBTqSJ/wEXf3Hzr+wADgAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Xcht9iebXJeT",
        "colab_type": "text"
      },
      "source": [
        "### One-Hot encoding"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "xbajoq1gXVMt",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "#Converting labels to One-hot representations of shape (set_size,digits,classes)\n",
        "possible_classes = 10\n",
        "\n",
        "def convert_labels(labels):\n",
        "        \n",
        "    #Declare output ndarrays\n",
        "    dig0_arr = np.ndarray(shape=(len(labels),possible_classes))\n",
        "    dig1_arr = np.ndarray(shape=(len(labels),possible_classes))\n",
        "\n",
        "    \n",
        "    for index,label in enumerate(labels):\n",
        "        \n",
        "        dig0_arr[index,:] = np_utils.to_categorical(label[0],possible_classes)\n",
        "        dig1_arr[index,:] = np_utils.to_categorical(label[1],possible_classes)\n",
        "        \n",
        "    return [dig0_arr,dig1_arr]"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "voq2ld3CXwha",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "train_labels = convert_labels(y_seq_train)\n",
        "test_labels = convert_labels(y_seq_test)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OLJmhQrYPpVe",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "312efd2d-dc66-4daf-c567-94620d31cd44"
      },
      "source": [
        "np.shape(test_labels[0])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(10000, 10)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 9
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "wY3NHcsto0iZ",
        "colab_type": "text"
      },
      "source": [
        "### Second Model --> LRCN (CNN-LSTM)"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XhwNcNBAo5ik",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 707
        },
        "outputId": "afdc31b5-1dc4-4b56-f4d9-4c314683d9dd"
      },
      "source": [
        "from keras import  applications\n",
        "from keras.models import Sequential\n",
        "from keras.layers import Conv2D, MaxPooling2D, Flatten, BatchNormalization, Lambda\n",
        "\n",
        "nb_classes = 10\n",
        "\n",
        "img_rows = 64\n",
        "img_cols = 64\n",
        "img_channels = 1\n",
        "\n",
        "#number of convulation filters to use\n",
        "nb_filters = 32\n",
        "# size of pooling area for max pooling\n",
        "pool_size = (2, 2)\n",
        "# convolution kernel size\n",
        "kernel_size = (3, 3)\n",
        "\n",
        "#defining the input\n",
        "inputs = Input(shape=(None,img_rows,img_cols,img_channels))\n",
        "\n",
        "#Model taken from keras example. Worked well for a digit, dunno for multiple\n",
        "cov = TimeDistributed(Conv2D(nb_filters,kernel_size,padding='same'))(inputs)\n",
        "cov = TimeDistributed(Activation('relu'))(cov)\n",
        "cov = TimeDistributed(BatchNormalization())(cov)\n",
        "cov = TimeDistributed(Conv2D(nb_filters,kernel_size))(cov)\n",
        "cov = TimeDistributed(Activation('relu'))(cov)\n",
        "cov = TimeDistributed(BatchNormalization())(cov)\n",
        "cov = TimeDistributed(MaxPooling2D(pool_size=pool_size))(cov)\n",
        "cov = TimeDistributed(Dropout(0.25))(cov)\n",
        "cov_out = TimeDistributed(Flatten())(cov)\n",
        "\n",
        "# return_sequences = true -> to share output of lstm to all dense layers\n",
        "encoded_vid = LSTM(64,return_sequences=True)(cov_out)\n",
        "encoded_vid = TimeDistributed(Dense(32, activation='relu'))(encoded_vid)\n",
        "encoded_vid = Dropout(0.25)(encoded_vid)\n",
        "\n",
        "#Prediction layers\n",
        "c0 = TimeDistributed(Dense(nb_classes, activation='softmax'),name='digit1')(encoded_vid)\n",
        "c1 = TimeDistributed(Dense(nb_classes, activation='softmax'),name='digit2')(encoded_vid)\n",
        "\n",
        "model = Model(inputs=[inputs],outputs=[c0,c1], name='custom-cnn-lstm')\n",
        "model.compile(optimizer='adam',loss='categorical_crossentropy',metrics=['acc'])\n",
        "model.summary()"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Model: \"custom-cnn-lstm\"\n",
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "input_2 (InputLayer)            (None, None, 64, 64, 0                                            \n",
            "__________________________________________________________________________________________________\n",
            "time_distributed_1 (TimeDistrib (None, None, 64, 64, 320         input_2[0][0]                    \n",
            "__________________________________________________________________________________________________\n",
            "time_distributed_2 (TimeDistrib (None, None, 64, 64, 0           time_distributed_1[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "time_distributed_3 (TimeDistrib (None, None, 64, 64, 128         time_distributed_2[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "time_distributed_4 (TimeDistrib (None, None, 62, 62, 9248        time_distributed_3[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "time_distributed_5 (TimeDistrib (None, None, 62, 62, 0           time_distributed_4[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "time_distributed_6 (TimeDistrib (None, None, 62, 62, 128         time_distributed_5[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "time_distributed_7 (TimeDistrib (None, None, 31, 31, 0           time_distributed_6[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "time_distributed_8 (TimeDistrib (None, None, 31, 31, 0           time_distributed_7[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "time_distributed_9 (TimeDistrib (None, None, 30752)  0           time_distributed_8[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "lstm_1 (LSTM)                   (None, None, 64)     7889152     time_distributed_9[0][0]         \n",
            "__________________________________________________________________________________________________\n",
            "time_distributed_10 (TimeDistri (None, None, 32)     2080        lstm_1[0][0]                     \n",
            "__________________________________________________________________________________________________\n",
            "dropout_4 (Dropout)             (None, None, 32)     0           time_distributed_10[0][0]        \n",
            "__________________________________________________________________________________________________\n",
            "digit1 (TimeDistributed)        (None, None, 10)     330         dropout_4[0][0]                  \n",
            "__________________________________________________________________________________________________\n",
            "digit2 (TimeDistributed)        (None, None, 10)     330         dropout_4[0][0]                  \n",
            "==================================================================================================\n",
            "Total params: 7,901,716\n",
            "Trainable params: 7,901,588\n",
            "Non-trainable params: 128\n",
            "__________________________________________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "PvZ9tiPfoSc1",
        "colab_type": "text"
      },
      "source": [
        "### Prepare Data"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "BNT6JXd4oTy1",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 35
        },
        "outputId": "3e223273-b4f0-4aec-9c05-2f2bcd2db273"
      },
      "source": [
        "def prep_data_keras(img_data):\n",
        "    \n",
        "    ### 10 refers to time_steps\n",
        "    img_data = img_data.reshape(len(img_data)//10,10,64,64,1)\n",
        "    img_data = img_data.astype('float32')\n",
        "    img_data /= 255.0\n",
        "    \n",
        "    return img_data\n",
        "\n",
        "train_images = prep_data_keras(x_seq_train)\n",
        "test_images = prep_data_keras(x_seq_test)\n",
        "print(f\"Test Image Shape: {test_images.shape}\")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train Image Shape: (1000, 10, 64, 64, 1)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Gh7G6V8D5uBP",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "train_labels[0] = train_labels[0].reshape(6000,10,10)\n",
        "train_labels[1] = train_labels[1].reshape(6000,10,10)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "RW7yO7RKW_OM",
        "colab_type": "text"
      },
      "source": [
        "### Define callbacks and call model.fit()"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "FjUIChSFpB3F",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "abb57078-c6c5-4aed-fb29-9d111e5fc532"
      },
      "source": [
        "from keras.callbacks import ModelCheckpoint, LearningRateScheduler, EarlyStopping\n",
        "import numpy as np\n",
        "\n",
        "save_dir = 'two-digit-custom-model-test3-with-cnn-lstm.h5'\n",
        "model_checkpoint = ModelCheckpoint(save_dir, \n",
        "\t\t\t\t\t\t\tmonitor='val_loss', verbose=1, \n",
        "\t\t\t\t\t\t\tsave_best_only=True, mode='min')\n",
        "\n",
        "lr_sched = LearningRateScheduler(lambda epoch: 1e-4 * (0.75 ** np.floor(epoch / 2)))\n",
        "\n",
        "early = EarlyStopping(monitor='val_loss', patience=5, mode='min', verbose=1)\n",
        "                      \n",
        "hist = model.fit(train_images,train_labels,batch_size=32,epochs=50,verbose=2,validation_split=0.2, callbacks=[model_checkpoint, lr_sched, early])"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train on 4800 samples, validate on 1200 samples\n",
            "Epoch 1/50\n",
            " - 47s - loss: 2.9499 - digit1_loss: 1.4665 - digit2_loss: 1.4833 - digit1_acc: 0.5384 - digit2_acc: 0.5351 - val_loss: 4.5822 - val_digit1_loss: 2.2911 - val_digit2_loss: 2.2911 - val_digit1_acc: 0.2533 - val_digit2_acc: 0.1524\n",
            "\n",
            "Epoch 00001: val_loss improved from inf to 4.58222, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 2/50\n",
            " - 40s - loss: 2.0619 - digit1_loss: 1.0101 - digit2_loss: 1.0518 - digit1_acc: 0.7265 - digit2_acc: 0.7059 - val_loss: 4.5332 - val_digit1_loss: 2.2676 - val_digit2_loss: 2.2655 - val_digit1_acc: 0.1525 - val_digit2_acc: 0.1780\n",
            "\n",
            "Epoch 00002: val_loss improved from 4.58222 to 4.53317, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 3/50\n",
            " - 40s - loss: 1.6838 - digit1_loss: 0.8255 - digit2_loss: 0.8583 - digit1_acc: 0.7893 - digit2_acc: 0.7754 - val_loss: 4.3148 - val_digit1_loss: 2.1575 - val_digit2_loss: 2.1573 - val_digit1_acc: 0.1667 - val_digit2_acc: 0.1777\n",
            "\n",
            "Epoch 00003: val_loss improved from 4.53317 to 4.31484, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 4/50\n",
            " - 40s - loss: 1.4573 - digit1_loss: 0.7076 - digit2_loss: 0.7497 - digit1_acc: 0.8234 - digit2_acc: 0.8082 - val_loss: 2.3554 - val_digit1_loss: 1.1861 - val_digit2_loss: 1.1695 - val_digit1_acc: 0.7442 - val_digit2_acc: 0.6925\n",
            "\n",
            "Epoch 00004: val_loss improved from 4.31484 to 2.35544, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 5/50\n",
            " - 40s - loss: 1.2772 - digit1_loss: 0.6220 - digit2_loss: 0.6552 - digit1_acc: 0.8500 - digit2_acc: 0.8370 - val_loss: 0.9798 - val_digit1_loss: 0.4844 - val_digit2_loss: 0.4955 - val_digit1_acc: 0.9351 - val_digit2_acc: 0.9284\n",
            "\n",
            "Epoch 00005: val_loss improved from 2.35544 to 0.97975, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 6/50\n",
            " - 40s - loss: 1.1648 - digit1_loss: 0.5664 - digit2_loss: 0.5984 - digit1_acc: 0.8621 - digit2_acc: 0.8518 - val_loss: 0.7610 - val_digit1_loss: 0.3640 - val_digit2_loss: 0.3975 - val_digit1_acc: 0.9463 - val_digit2_acc: 0.9433\n",
            "\n",
            "Epoch 00006: val_loss improved from 0.97975 to 0.76099, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 7/50\n",
            " - 40s - loss: 1.0739 - digit1_loss: 0.5243 - digit2_loss: 0.5495 - digit1_acc: 0.8717 - digit2_acc: 0.8663 - val_loss: 0.6552 - val_digit1_loss: 0.3095 - val_digit2_loss: 0.3461 - val_digit1_acc: 0.9573 - val_digit2_acc: 0.9492\n",
            "\n",
            "Epoch 00007: val_loss improved from 0.76099 to 0.65515, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 8/50\n",
            " - 40s - loss: 1.0068 - digit1_loss: 0.4882 - digit2_loss: 0.5186 - digit1_acc: 0.8810 - digit2_acc: 0.8732 - val_loss: 0.5976 - val_digit1_loss: 0.2848 - val_digit2_loss: 0.3133 - val_digit1_acc: 0.9588 - val_digit2_acc: 0.9554\n",
            "\n",
            "Epoch 00008: val_loss improved from 0.65515 to 0.59762, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 9/50\n",
            " - 40s - loss: 0.9526 - digit1_loss: 0.4647 - digit2_loss: 0.4880 - digit1_acc: 0.8878 - digit2_acc: 0.8829 - val_loss: 0.5591 - val_digit1_loss: 0.2654 - val_digit2_loss: 0.2941 - val_digit1_acc: 0.9617 - val_digit2_acc: 0.9583\n",
            "\n",
            "Epoch 00009: val_loss improved from 0.59762 to 0.55908, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 10/50\n",
            " - 40s - loss: 0.9135 - digit1_loss: 0.4443 - digit2_loss: 0.4692 - digit1_acc: 0.8917 - digit2_acc: 0.8867 - val_loss: 0.5227 - val_digit1_loss: 0.2486 - val_digit2_loss: 0.2746 - val_digit1_acc: 0.9636 - val_digit2_acc: 0.9599\n",
            "\n",
            "Epoch 00010: val_loss improved from 0.55908 to 0.52274, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 11/50\n",
            " - 40s - loss: 0.8787 - digit1_loss: 0.4261 - digit2_loss: 0.4526 - digit1_acc: 0.8979 - digit2_acc: 0.8919 - val_loss: 0.4962 - val_digit1_loss: 0.2357 - val_digit2_loss: 0.2610 - val_digit1_acc: 0.9643 - val_digit2_acc: 0.9601\n",
            "\n",
            "Epoch 00011: val_loss improved from 0.52274 to 0.49618, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 12/50\n",
            " - 40s - loss: 0.8561 - digit1_loss: 0.4192 - digit2_loss: 0.4369 - digit1_acc: 0.8988 - digit2_acc: 0.8955 - val_loss: 0.4758 - val_digit1_loss: 0.2261 - val_digit2_loss: 0.2501 - val_digit1_acc: 0.9646 - val_digit2_acc: 0.9623\n",
            "\n",
            "Epoch 00012: val_loss improved from 0.49618 to 0.47581, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 13/50\n",
            " - 40s - loss: 0.8222 - digit1_loss: 0.3998 - digit2_loss: 0.4224 - digit1_acc: 0.9035 - digit2_acc: 0.8993 - val_loss: 0.4595 - val_digit1_loss: 0.2176 - val_digit2_loss: 0.2423 - val_digit1_acc: 0.9656 - val_digit2_acc: 0.9629\n",
            "\n",
            "Epoch 00013: val_loss improved from 0.47581 to 0.45949, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 14/50\n",
            " - 40s - loss: 0.8007 - digit1_loss: 0.3891 - digit2_loss: 0.4117 - digit1_acc: 0.9055 - digit2_acc: 0.9023 - val_loss: 0.4486 - val_digit1_loss: 0.2123 - val_digit2_loss: 0.2365 - val_digit1_acc: 0.9675 - val_digit2_acc: 0.9645\n",
            "\n",
            "Epoch 00014: val_loss improved from 0.45949 to 0.44857, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 15/50\n",
            " - 40s - loss: 0.7878 - digit1_loss: 0.3837 - digit2_loss: 0.4042 - digit1_acc: 0.9066 - digit2_acc: 0.9023 - val_loss: 0.4361 - val_digit1_loss: 0.2070 - val_digit2_loss: 0.2294 - val_digit1_acc: 0.9678 - val_digit2_acc: 0.9642\n",
            "\n",
            "Epoch 00015: val_loss improved from 0.44857 to 0.43609, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 16/50\n",
            " - 40s - loss: 0.7686 - digit1_loss: 0.3745 - digit2_loss: 0.3941 - digit1_acc: 0.9096 - digit2_acc: 0.9081 - val_loss: 0.4185 - val_digit1_loss: 0.1997 - val_digit2_loss: 0.2192 - val_digit1_acc: 0.9680 - val_digit2_acc: 0.9650\n",
            "\n",
            "Epoch 00016: val_loss improved from 0.43609 to 0.41848, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 17/50\n",
            " - 40s - loss: 0.7656 - digit1_loss: 0.3732 - digit2_loss: 0.3924 - digit1_acc: 0.9091 - digit2_acc: 0.9054 - val_loss: 0.4142 - val_digit1_loss: 0.1962 - val_digit2_loss: 0.2183 - val_digit1_acc: 0.9679 - val_digit2_acc: 0.9653\n",
            "\n",
            "Epoch 00017: val_loss improved from 0.41848 to 0.41419, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 18/50\n",
            " - 40s - loss: 0.7438 - digit1_loss: 0.3626 - digit2_loss: 0.3812 - digit1_acc: 0.9119 - digit2_acc: 0.9092 - val_loss: 0.4066 - val_digit1_loss: 0.1933 - val_digit2_loss: 0.2137 - val_digit1_acc: 0.9688 - val_digit2_acc: 0.9663\n",
            "\n",
            "Epoch 00018: val_loss improved from 0.41419 to 0.40661, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 19/50\n",
            " - 40s - loss: 0.7417 - digit1_loss: 0.3606 - digit2_loss: 0.3811 - digit1_acc: 0.9104 - digit2_acc: 0.9086 - val_loss: 0.3992 - val_digit1_loss: 0.1888 - val_digit2_loss: 0.2108 - val_digit1_acc: 0.9684 - val_digit2_acc: 0.9663\n",
            "\n",
            "Epoch 00019: val_loss improved from 0.40661 to 0.39921, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 20/50\n",
            " - 40s - loss: 0.7324 - digit1_loss: 0.3568 - digit2_loss: 0.3756 - digit1_acc: 0.9141 - digit2_acc: 0.9105 - val_loss: 0.3948 - val_digit1_loss: 0.1872 - val_digit2_loss: 0.2080 - val_digit1_acc: 0.9687 - val_digit2_acc: 0.9663\n",
            "\n",
            "Epoch 00020: val_loss improved from 0.39921 to 0.39485, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 21/50\n",
            " - 40s - loss: 0.7291 - digit1_loss: 0.3551 - digit2_loss: 0.3740 - digit1_acc: 0.9139 - digit2_acc: 0.9110 - val_loss: 0.3902 - val_digit1_loss: 0.1849 - val_digit2_loss: 0.2056 - val_digit1_acc: 0.9696 - val_digit2_acc: 0.9667\n",
            "\n",
            "Epoch 00021: val_loss improved from 0.39485 to 0.39015, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 22/50\n",
            " - 40s - loss: 0.7216 - digit1_loss: 0.3515 - digit2_loss: 0.3701 - digit1_acc: 0.9150 - digit2_acc: 0.9122 - val_loss: 0.3866 - val_digit1_loss: 0.1831 - val_digit2_loss: 0.2038 - val_digit1_acc: 0.9689 - val_digit2_acc: 0.9667\n",
            "\n",
            "Epoch 00022: val_loss improved from 0.39015 to 0.38657, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 23/50\n",
            " - 40s - loss: 0.7159 - digit1_loss: 0.3480 - digit2_loss: 0.3679 - digit1_acc: 0.9156 - digit2_acc: 0.9106 - val_loss: 0.3822 - val_digit1_loss: 0.1808 - val_digit2_loss: 0.2017 - val_digit1_acc: 0.9696 - val_digit2_acc: 0.9669\n",
            "\n",
            "Epoch 00023: val_loss improved from 0.38657 to 0.38215, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 24/50\n",
            " - 40s - loss: 0.7103 - digit1_loss: 0.3468 - digit2_loss: 0.3635 - digit1_acc: 0.9162 - digit2_acc: 0.9151 - val_loss: 0.3800 - val_digit1_loss: 0.1801 - val_digit2_loss: 0.2002 - val_digit1_acc: 0.9696 - val_digit2_acc: 0.9676\n",
            "\n",
            "Epoch 00024: val_loss improved from 0.38215 to 0.37997, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 25/50\n",
            " - 40s - loss: 0.7105 - digit1_loss: 0.3471 - digit2_loss: 0.3634 - digit1_acc: 0.9148 - digit2_acc: 0.9135 - val_loss: 0.3773 - val_digit1_loss: 0.1790 - val_digit2_loss: 0.1986 - val_digit1_acc: 0.9692 - val_digit2_acc: 0.9671\n",
            "\n",
            "Epoch 00025: val_loss improved from 0.37997 to 0.37727, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 26/50\n",
            " - 40s - loss: 0.7018 - digit1_loss: 0.3411 - digit2_loss: 0.3607 - digit1_acc: 0.9173 - digit2_acc: 0.9134 - val_loss: 0.3749 - val_digit1_loss: 0.1781 - val_digit2_loss: 0.1972 - val_digit1_acc: 0.9696 - val_digit2_acc: 0.9672\n",
            "\n",
            "Epoch 00026: val_loss improved from 0.37727 to 0.37488, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 27/50\n",
            " - 40s - loss: 0.7072 - digit1_loss: 0.3472 - digit2_loss: 0.3600 - digit1_acc: 0.9133 - digit2_acc: 0.9136 - val_loss: 0.3732 - val_digit1_loss: 0.1768 - val_digit2_loss: 0.1967 - val_digit1_acc: 0.9701 - val_digit2_acc: 0.9675\n",
            "\n",
            "Epoch 00027: val_loss improved from 0.37488 to 0.37317, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 28/50\n",
            " - 40s - loss: 0.7017 - digit1_loss: 0.3435 - digit2_loss: 0.3582 - digit1_acc: 0.9143 - digit2_acc: 0.9150 - val_loss: 0.3720 - val_digit1_loss: 0.1764 - val_digit2_loss: 0.1960 - val_digit1_acc: 0.9699 - val_digit2_acc: 0.9678\n",
            "\n",
            "Epoch 00028: val_loss improved from 0.37317 to 0.37199, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 29/50\n",
            " - 40s - loss: 0.6945 - digit1_loss: 0.3378 - digit2_loss: 0.3567 - digit1_acc: 0.9178 - digit2_acc: 0.9158 - val_loss: 0.3702 - val_digit1_loss: 0.1756 - val_digit2_loss: 0.1949 - val_digit1_acc: 0.9695 - val_digit2_acc: 0.9677\n",
            "\n",
            "Epoch 00029: val_loss improved from 0.37199 to 0.37021, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 30/50\n",
            " - 40s - loss: 0.6991 - digit1_loss: 0.3379 - digit2_loss: 0.3612 - digit1_acc: 0.9177 - digit2_acc: 0.9136 - val_loss: 0.3690 - val_digit1_loss: 0.1749 - val_digit2_loss: 0.1944 - val_digit1_acc: 0.9700 - val_digit2_acc: 0.9682\n",
            "\n",
            "Epoch 00030: val_loss improved from 0.37021 to 0.36901, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 31/50\n",
            " - 40s - loss: 0.6925 - digit1_loss: 0.3359 - digit2_loss: 0.3565 - digit1_acc: 0.9171 - digit2_acc: 0.9150 - val_loss: 0.3686 - val_digit1_loss: 0.1745 - val_digit2_loss: 0.1943 - val_digit1_acc: 0.9703 - val_digit2_acc: 0.9680\n",
            "\n",
            "Epoch 00031: val_loss improved from 0.36901 to 0.36856, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 32/50\n",
            " - 40s - loss: 0.6949 - digit1_loss: 0.3364 - digit2_loss: 0.3585 - digit1_acc: 0.9189 - digit2_acc: 0.9140 - val_loss: 0.3670 - val_digit1_loss: 0.1741 - val_digit2_loss: 0.1933 - val_digit1_acc: 0.9699 - val_digit2_acc: 0.9680\n",
            "\n",
            "Epoch 00032: val_loss improved from 0.36856 to 0.36700, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 33/50\n",
            " - 40s - loss: 0.6933 - digit1_loss: 0.3362 - digit2_loss: 0.3571 - digit1_acc: 0.9178 - digit2_acc: 0.9137 - val_loss: 0.3668 - val_digit1_loss: 0.1739 - val_digit2_loss: 0.1932 - val_digit1_acc: 0.9699 - val_digit2_acc: 0.9678\n",
            "\n",
            "Epoch 00033: val_loss improved from 0.36700 to 0.36676, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 34/50\n",
            " - 40s - loss: 0.6933 - digit1_loss: 0.3394 - digit2_loss: 0.3539 - digit1_acc: 0.9161 - digit2_acc: 0.9149 - val_loss: 0.3659 - val_digit1_loss: 0.1734 - val_digit2_loss: 0.1928 - val_digit1_acc: 0.9701 - val_digit2_acc: 0.9678\n",
            "\n",
            "Epoch 00034: val_loss improved from 0.36676 to 0.36589, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 35/50\n",
            " - 40s - loss: 0.6891 - digit1_loss: 0.3366 - digit2_loss: 0.3525 - digit1_acc: 0.9176 - digit2_acc: 0.9163 - val_loss: 0.3654 - val_digit1_loss: 0.1731 - val_digit2_loss: 0.1926 - val_digit1_acc: 0.9701 - val_digit2_acc: 0.9680\n",
            "\n",
            "Epoch 00035: val_loss improved from 0.36589 to 0.36541, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 36/50\n",
            " - 40s - loss: 0.6914 - digit1_loss: 0.3406 - digit2_loss: 0.3508 - digit1_acc: 0.9159 - digit2_acc: 0.9184 - val_loss: 0.3649 - val_digit1_loss: 0.1728 - val_digit2_loss: 0.1924 - val_digit1_acc: 0.9703 - val_digit2_acc: 0.9682\n",
            "\n",
            "Epoch 00036: val_loss improved from 0.36541 to 0.36492, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 37/50\n",
            " - 40s - loss: 0.6949 - digit1_loss: 0.3359 - digit2_loss: 0.3589 - digit1_acc: 0.9177 - digit2_acc: 0.9147 - val_loss: 0.3645 - val_digit1_loss: 0.1727 - val_digit2_loss: 0.1921 - val_digit1_acc: 0.9703 - val_digit2_acc: 0.9682\n",
            "\n",
            "Epoch 00037: val_loss improved from 0.36492 to 0.36452, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 38/50\n",
            " - 40s - loss: 0.6888 - digit1_loss: 0.3358 - digit2_loss: 0.3529 - digit1_acc: 0.9173 - digit2_acc: 0.9156 - val_loss: 0.3642 - val_digit1_loss: 0.1725 - val_digit2_loss: 0.1920 - val_digit1_acc: 0.9703 - val_digit2_acc: 0.9681\n",
            "\n",
            "Epoch 00038: val_loss improved from 0.36452 to 0.36422, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 39/50\n",
            " - 40s - loss: 0.6850 - digit1_loss: 0.3340 - digit2_loss: 0.3510 - digit1_acc: 0.9179 - digit2_acc: 0.9165 - val_loss: 0.3638 - val_digit1_loss: 0.1724 - val_digit2_loss: 0.1918 - val_digit1_acc: 0.9702 - val_digit2_acc: 0.9680\n",
            "\n",
            "Epoch 00039: val_loss improved from 0.36422 to 0.36383, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 40/50\n",
            " - 40s - loss: 0.6895 - digit1_loss: 0.3356 - digit2_loss: 0.3539 - digit1_acc: 0.9180 - digit2_acc: 0.9164 - val_loss: 0.3637 - val_digit1_loss: 0.1723 - val_digit2_loss: 0.1917 - val_digit1_acc: 0.9704 - val_digit2_acc: 0.9683\n",
            "\n",
            "Epoch 00040: val_loss improved from 0.36383 to 0.36367, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 41/50\n",
            " - 40s - loss: 0.6884 - digit1_loss: 0.3346 - digit2_loss: 0.3538 - digit1_acc: 0.9190 - digit2_acc: 0.9155 - val_loss: 0.3635 - val_digit1_loss: 0.1723 - val_digit2_loss: 0.1915 - val_digit1_acc: 0.9703 - val_digit2_acc: 0.9681\n",
            "\n",
            "Epoch 00041: val_loss improved from 0.36367 to 0.36347, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 42/50\n",
            " - 40s - loss: 0.6871 - digit1_loss: 0.3329 - digit2_loss: 0.3542 - digit1_acc: 0.9197 - digit2_acc: 0.9156 - val_loss: 0.3632 - val_digit1_loss: 0.1721 - val_digit2_loss: 0.1914 - val_digit1_acc: 0.9703 - val_digit2_acc: 0.9682\n",
            "\n",
            "Epoch 00042: val_loss improved from 0.36347 to 0.36324, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 43/50\n",
            " - 40s - loss: 0.6854 - digit1_loss: 0.3324 - digit2_loss: 0.3530 - digit1_acc: 0.9189 - digit2_acc: 0.9162 - val_loss: 0.3630 - val_digit1_loss: 0.1720 - val_digit2_loss: 0.1913 - val_digit1_acc: 0.9701 - val_digit2_acc: 0.9680\n",
            "\n",
            "Epoch 00043: val_loss improved from 0.36324 to 0.36295, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 44/50\n",
            " - 40s - loss: 0.6832 - digit1_loss: 0.3333 - digit2_loss: 0.3498 - digit1_acc: 0.9186 - digit2_acc: 0.9159 - val_loss: 0.3627 - val_digit1_loss: 0.1719 - val_digit2_loss: 0.1911 - val_digit1_acc: 0.9701 - val_digit2_acc: 0.9680\n",
            "\n",
            "Epoch 00044: val_loss improved from 0.36295 to 0.36268, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 45/50\n",
            " - 40s - loss: 0.6888 - digit1_loss: 0.3375 - digit2_loss: 0.3512 - digit1_acc: 0.9158 - digit2_acc: 0.9174 - val_loss: 0.3625 - val_digit1_loss: 0.1718 - val_digit2_loss: 0.1911 - val_digit1_acc: 0.9701 - val_digit2_acc: 0.9680\n",
            "\n",
            "Epoch 00045: val_loss improved from 0.36268 to 0.36253, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 46/50\n",
            " - 40s - loss: 0.6871 - digit1_loss: 0.3347 - digit2_loss: 0.3524 - digit1_acc: 0.9179 - digit2_acc: 0.9156 - val_loss: 0.3624 - val_digit1_loss: 0.1717 - val_digit2_loss: 0.1910 - val_digit1_acc: 0.9702 - val_digit2_acc: 0.9681\n",
            "\n",
            "Epoch 00046: val_loss improved from 0.36253 to 0.36241, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 47/50\n",
            " - 40s - loss: 0.6905 - digit1_loss: 0.3355 - digit2_loss: 0.3550 - digit1_acc: 0.9179 - digit2_acc: 0.9143 - val_loss: 0.3623 - val_digit1_loss: 0.1717 - val_digit2_loss: 0.1910 - val_digit1_acc: 0.9703 - val_digit2_acc: 0.9682\n",
            "\n",
            "Epoch 00047: val_loss improved from 0.36241 to 0.36232, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 48/50\n",
            " - 40s - loss: 0.6909 - digit1_loss: 0.3353 - digit2_loss: 0.3556 - digit1_acc: 0.9187 - digit2_acc: 0.9143 - val_loss: 0.3623 - val_digit1_loss: 0.1717 - val_digit2_loss: 0.1909 - val_digit1_acc: 0.9703 - val_digit2_acc: 0.9681\n",
            "\n",
            "Epoch 00048: val_loss improved from 0.36232 to 0.36226, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 49/50\n",
            " - 40s - loss: 0.6869 - digit1_loss: 0.3341 - digit2_loss: 0.3527 - digit1_acc: 0.9194 - digit2_acc: 0.9162 - val_loss: 0.3622 - val_digit1_loss: 0.1716 - val_digit2_loss: 0.1909 - val_digit1_acc: 0.9702 - val_digit2_acc: 0.9681\n",
            "\n",
            "Epoch 00049: val_loss improved from 0.36226 to 0.36219, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n",
            "Epoch 50/50\n",
            " - 40s - loss: 0.6837 - digit1_loss: 0.3320 - digit2_loss: 0.3518 - digit1_acc: 0.9182 - digit2_acc: 0.9161 - val_loss: 0.3621 - val_digit1_loss: 0.1716 - val_digit2_loss: 0.1908 - val_digit1_acc: 0.9702 - val_digit2_acc: 0.9682\n",
            "\n",
            "Epoch 00050: val_loss improved from 0.36219 to 0.36209, saving model to two-digit-custom-model-test3-with-cnn-lstm.h5\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "hSs9CHhRhXY-",
        "colab_type": "text"
      },
      "source": [
        "### Model Visualization"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "8Il0BKnlhaBC",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 329
        },
        "outputId": "84b8c11d-1234-4773-bd94-b4204e3f9a75"
      },
      "source": [
        "def plot_accuracy_loss(history):\n",
        "    \"\"\"\n",
        "        Plot the accuracy and the loss during the training of the nn.\n",
        "    \"\"\"\n",
        "    fig = plt.figure(figsize=(16,10))\n",
        "\n",
        "    # Plot accuracy\n",
        "    plt.subplot(221)\n",
        "    plt.plot(hist.history['digit1_acc'],'bo--', label = \"digit1_acc\")\n",
        "    plt.plot(hist.history['val_digit1_acc'], 'ro--', label = \"val_digit1_acc\")\n",
        "    plt.title(\"digit1_acc vs val_digit1_acc\")\n",
        "    plt.ylabel(\"digit1_accuracy\")\n",
        "    plt.xlabel(\"epochs\")\n",
        "    plt.legend()\n",
        "\n",
        "    # Plot loss function\n",
        "    plt.subplot(222)\n",
        "    plt.plot(hist.history['digit2_acc'],'bo--', label = \"digit2_acc\")\n",
        "    plt.plot(hist.history['val_digit2_acc'], 'ro--', label = \"val_digit2_acc\")\n",
        "    plt.title(\"digit2_acc vs val_digit2_acc\")\n",
        "    plt.ylabel(\"digit2_accuracy\")\n",
        "    plt.xlabel(\"epochs\")\n",
        "\n",
        "    plt.legend()\n",
        "    plt.show()\n",
        "  \n",
        "plot_accuracy_loss(hist)"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": [
              "<Figure size 1152x720 with 2 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "3-8d4BKoXRjS",
        "colab_type": "text"
      },
      "source": [
        "### Save model weights"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SuCHIJgFXqZd",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "!cp two-digit-custom-model-test3-with-cnn-lstm.h5 drive/My\\ Drive/mnist-generator/."
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_f2oBOFeXbRT",
        "colab_type": "text"
      },
      "source": [
        "### Load Model"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "VMGmy9fqXiWr",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "!cp drive/My\\ Drive/mnist-generator/two-digit-custom-model-test2-with-batchnormalization.h5 ."
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7jgdWDDcXeVb",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "model.load_weights('two-digit-custom-model-test2-with-batchnormalization.h5')"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "sdEWDUgrXUqL",
        "colab_type": "text"
      },
      "source": [
        "### Evaluate model on test"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OxDhqAXtFGif",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "test_labels[0] = test_labels[0].reshape(1000,10,10)\n",
        "test_labels[1] = test_labels[1].reshape(1000,10,10)"
      ],
      "execution_count": null,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "wUsheTedYP-L",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 53
        },
        "outputId": "00e838c0-3bf4-47b6-9230-ce534085b7d8"
      },
      "source": [
        "scores = model.evaluate(test_images, test_labels)\n",
        "print(f\"First digit accuracy: {scores[3]} , Second digit accuracy: {scores[4]}\")"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "1000/1000 [==============================] - 3s 3ms/step\n",
            "First digit accuracy: 0.9678999781608582 , Second digit accuracy: 0.9670000076293945\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "vpKuFFTVZE_0",
        "colab_type": "text"
      },
      "source": [
        "### Model Inference"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CGf1rN2VZdjD",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "57014f8a-dd96-4a77-e57f-2f3593e1f4e6"
      },
      "source": [
        "#Printing some examples of real and predicted labels\n",
        "predictions = model.predict(test_images)\n",
        "predictions[0] = predictions[0].reshape(10000,10)\n",
        "predictions[1] = predictions[1].reshape(10000,10)\n",
        "test_labels[0] = test_labels[0].reshape(10000,10)\n",
        "test_labels[1] = test_labels[1].reshape(10000,10)\n",
        "\n",
        "for i in random.sample(range(0,10000),50):\n",
        "    \n",
        "    actual_labels = []\n",
        "    predicted_labels = []\n",
        "    \n",
        "    for j in range(0,2):\n",
        "        actual_labels.append(np.argmax(test_labels[j][i]))\n",
        "        predicted_labels.append(np.argmax(predictions[j][i]))\n",
        "        \n",
        "    print(\"Actual labels: {}\".format(actual_labels))\n",
        "    print(\"Predicted labels: {}\\n\".format(predicted_labels))"
      ],
      "execution_count": null,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Actual labels: [3, 3]\n",
            "Predicted labels: [3, 3]\n",
            "\n",
            "Actual labels: [9, 3]\n",
            "Predicted labels: [9, 3]\n",
            "\n",
            "Actual labels: [5, 3]\n",
            "Predicted labels: [5, 3]\n",
            "\n",
            "Actual labels: [4, 8]\n",
            "Predicted labels: [4, 8]\n",
            "\n",
            "Actual labels: [3, 1]\n",
            "Predicted labels: [3, 1]\n",
            "\n",
            "Actual labels: [4, 1]\n",
            "Predicted labels: [4, 1]\n",
            "\n",
            "Actual labels: [4, 9]\n",
            "Predicted labels: [4, 9]\n",
            "\n",
            "Actual labels: [7, 8]\n",
            "Predicted labels: [7, 8]\n",
            "\n",
            "Actual labels: [5, 4]\n",
            "Predicted labels: [5, 4]\n",
            "\n",
            "Actual labels: [7, 1]\n",
            "Predicted labels: [7, 1]\n",
            "\n",
            "Actual labels: [7, 9]\n",
            "Predicted labels: [7, 9]\n",
            "\n",
            "Actual labels: [7, 4]\n",
            "Predicted labels: [7, 4]\n",
            "\n",
            "Actual labels: [7, 7]\n",
            "Predicted labels: [7, 2]\n",
            "\n",
            "Actual labels: [9, 4]\n",
            "Predicted labels: [9, 4]\n",
            "\n",
            "Actual labels: [0, 1]\n",
            "Predicted labels: [0, 1]\n",
            "\n",
            "Actual labels: [8, 1]\n",
            "Predicted labels: [8, 1]\n",
            "\n",
            "Actual labels: [4, 8]\n",
            "Predicted labels: [4, 8]\n",
            "\n",
            "Actual labels: [7, 7]\n",
            "Predicted labels: [7, 7]\n",
            "\n",
            "Actual labels: [2, 5]\n",
            "Predicted labels: [2, 5]\n",
            "\n",
            "Actual labels: [4, 5]\n",
            "Predicted labels: [4, 5]\n",
            "\n",
            "Actual labels: [7, 0]\n",
            "Predicted labels: [7, 0]\n",
            "\n",
            "Actual labels: [5, 2]\n",
            "Predicted labels: [5, 2]\n",
            "\n",
            "Actual labels: [7, 3]\n",
            "Predicted labels: [7, 3]\n",
            "\n",
            "Actual labels: [0, 2]\n",
            "Predicted labels: [0, 2]\n",
            "\n",
            "Actual labels: [8, 2]\n",
            "Predicted labels: [8, 2]\n",
            "\n",
            "Actual labels: [2, 2]\n",
            "Predicted labels: [2, 2]\n",
            "\n",
            "Actual labels: [4, 3]\n",
            "Predicted labels: [4, 3]\n",
            "\n",
            "Actual labels: [5, 3]\n",
            "Predicted labels: [5, 3]\n",
            "\n",
            "Actual labels: [2, 1]\n",
            "Predicted labels: [2, 1]\n",
            "\n",
            "Actual labels: [3, 3]\n",
            "Predicted labels: [3, 3]\n",
            "\n",
            "Actual labels: [5, 8]\n",
            "Predicted labels: [5, 8]\n",
            "\n",
            "Actual labels: [1, 2]\n",
            "Predicted labels: [1, 2]\n",
            "\n",
            "Actual labels: [9, 3]\n",
            "Predicted labels: [9, 5]\n",
            "\n",
            "Actual labels: [9, 5]\n",
            "Predicted labels: [9, 5]\n",
            "\n",
            "Actual labels: [5, 8]\n",
            "Predicted labels: [5, 8]\n",
            "\n",
            "Actual labels: [4, 7]\n",
            "Predicted labels: [4, 7]\n",
            "\n",
            "Actual labels: [0, 1]\n",
            "Predicted labels: [0, 1]\n",
            "\n",
            "Actual labels: [3, 9]\n",
            "Predicted labels: [3, 9]\n",
            "\n",
            "Actual labels: [7, 7]\n",
            "Predicted labels: [7, 7]\n",
            "\n",
            "Actual labels: [2, 6]\n",
            "Predicted labels: [2, 6]\n",
            "\n",
            "Actual labels: [7, 1]\n",
            "Predicted labels: [7, 1]\n",
            "\n",
            "Actual labels: [3, 9]\n",
            "Predicted labels: [3, 9]\n",
            "\n",
            "Actual labels: [6, 7]\n",
            "Predicted labels: [6, 7]\n",
            "\n",
            "Actual labels: [5, 1]\n",
            "Predicted labels: [5, 1]\n",
            "\n",
            "Actual labels: [6, 0]\n",
            "Predicted labels: [6, 0]\n",
            "\n",
            "Actual labels: [1, 5]\n",
            "Predicted labels: [1, 5]\n",
            "\n",
            "Actual labels: [7, 6]\n",
            "Predicted labels: [7, 6]\n",
            "\n",
            "Actual labels: [4, 2]\n",
            "Predicted labels: [4, 2]\n",
            "\n",
            "Actual labels: [5, 8]\n",
            "Predicted labels: [5, 8]\n",
            "\n",
            "Actual labels: [9, 6]\n",
            "Predicted labels: [3, 6]\n",
            "\n"
          ],
          "name": "stdout"
        }
      ]
    }
  ]
}