auto_encoder_aryan.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "auto_encoder_aryan.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    }
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/aryansingh12/tensorflow-keras/blob/master/auto_encoder_aryan.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "H2tLHOYjt_F8",
        "colab_type": "code",
        "outputId": "b9bf58d7-961e-4615-92cd-c751da325a06",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "print(5)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "5\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "9eRII-2FuIWa",
        "colab_type": "code",
        "outputId": "275fde10-35d9-4ed9-b8db-6acf24ceaa33",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 451
        }
      },
      "source": [
        "from keras.models import Model\n",
        "from keras.layers import Input\n",
        "from keras.layers import LSTM\n",
        "from keras.layers import Dense\n",
        "from keras.utils.vis_utils import plot_model\n",
        "\n",
        "\n",
        "# configure\n",
        "num_encoder_tokens = 71\n",
        "num_decoder_tokens = 93\n",
        "latent_dim = 256\n",
        "\n",
        "# Define an input sequence and process it.\n",
        "encoder_inputs = Input(shape=(None, num_encoder_tokens))\n",
        "encoder = LSTM(latent_dim, return_state=True)\n",
        "encoder_outputs, state_h, state_c = encoder(encoder_inputs)\n",
        "\n",
        "# We discard `encoder_outputs` and only keep the states.\n",
        "encoder_states = [state_h, state_c]\n",
        "\n",
        "# Set up the decoder, using `encoder_states` as initial state.\n",
        "decoder_inputs = Input(shape=(None, num_decoder_tokens))\n",
        "\n",
        "# We set up our decoder to return full output sequences,\n",
        "# and to return internal states as well. We don't use the\n",
        "# return states in the training model, but we will use them in inference.\n",
        "decoder_lstm = LSTM(latent_dim, return_sequences=True, return_state=True)\n",
        "decoder_outputs, _, _ = decoder_lstm(decoder_inputs, initial_state=encoder_states)\n",
        "decoder_dense = Dense(num_decoder_tokens, activation='softmax')\n",
        "decoder_outputs = decoder_dense(decoder_outputs)\n",
        "\n",
        "# Define the model that will turn\n",
        "# `encoder_input_data` & `decoder_input_data` into `decoder_target_data`\n",
        "model = Model([encoder_inputs, decoder_inputs], decoder_outputs)\n",
        "\n",
        "# plot the model\n",
        "plot_model(model, to_file='model.png', show_shapes=True)\n",
        "\n",
        "# define encoder inference model\n",
        "encoder_model = Model(encoder_inputs, encoder_states)\n",
        "\n",
        "# define decoder inference model\n",
        "decoder_state_input_h = Input(shape=(latent_dim,))\n",
        "decoder_state_input_c = Input(shape=(latent_dim,))\n",
        "decoder_states_inputs = [decoder_state_input_h, decoder_state_input_c]\n",
        "decoder_outputs, state_h, state_c = decoder_lstm(decoder_inputs, initial_state=decoder_states_inputs)\n",
        "decoder_states = [state_h, state_c]\n",
        "decoder_outputs = decoder_dense(decoder_outputs)\n",
        "decoder_model = Model([decoder_inputs] + decoder_states_inputs, [decoder_outputs] + decoder_states)\n",
        "\n",
        "# summarize model\n",
        "plot_model(encoder_model, to_file='encoder_model.png', show_shapes=True)\n",
        "plot_model(decoder_model, to_file='decoder_model.png', show_shapes=True)"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Using TensorFlow backend.\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:66: The name tf.get_default_graph is deprecated. Please use tf.compat.v1.get_default_graph instead.\n",
            "\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:541: The name tf.placeholder is deprecated. Please use tf.compat.v1.placeholder instead.\n",
            "\n",
            "WARNING:tensorflow:From /usr/local/lib/python3.6/dist-packages/keras/backend/tensorflow_backend.py:4432: The name tf.random_uniform is deprecated. Please use tf.random.uniform instead.\n",
            "\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAABL0AAAEnCAYAAABMhDEjAAAABmJLR0QA/wD/AP+gvaeTAAAgAElE\nQVR4nOzdeVRTd/4//meAkEBIWBSFsomgpS6t2kWhWrRFtCIqdaPLdJlpx3baUqu21nWsC621Yz1u\n7Wnr+J2xLriNChUXtNY6RccpODJoreMKbigV2WV7/f7oj3yMQSVAuCQ8H+dwevrO+977et8b7yv3\nlZv3VYmIgIiIiIiIiIiIyI44KB0AERERERERERFRU2PRi4iIiIiIiIiI7A6LXkREREREREREZHdY\n9CIiIiIiIiIiIrvjdHtDeno6Fi5cqEQsRETUSk2YMAHh4eFWWffo0aOtsl4iImpeGzZssMp6ef1D\nRGQfwsPDMWHCBJM2szu9cnJysHHjxmYLisgaNm7ciNzcXKXDsCkHDx7EwYMHlQ6DWqGNGzciJyfH\nquvn+YBsGc/PlsvNzeXnWTti7ePJ6x+yB/y8YznmV/ty8OBBpKenm7Wb3elVy1rfpBA1B5VKhXff\nfRdjxoxROhSbUXs3DP/tU3NTqVRW3wbPB2TLeH623Pr16zF27FjuMztRezytje8XsmW8/rEc86t9\nudOvOzinFxERERERERER2R0WvYiIiIiIiIiIyO6w6EVERERERERERHaHRS8iIiIiIiIiIrI7LHoR\nEREREREREZHdYdGL6C62b98Od3d3JCcnKx1Ki5eWloYpU6Zg06ZN6NixI1QqFVQqFX73u9+Z9Y2O\njoZer4ejoyO6du2KjIwMBSKuH3sbDwBUVlZi5syZ6NixI5ydneHn54dJkyahrKzMpN/8+fMRFhYG\nFxcX6HQ6hIWFYcaMGSgsLDT22bZtG+bPn4/q6urmHgYRWYg5rf7sNacBwOzZs9GlSxcYDAZoNBqE\nhobi/fffR3FxsUm/uXPnGsd961+3bt3M1llZWYnExESEhobC2dkZHh4e6NatG86ePQuAuYLIljBX\n1B9zhW3kCha9iO5CRJQOwSb8+c9/xuLFizF16lSMHDkSp0+fRkhICNq0aYNvvvkG3377rUn/Xbt2\nYcOGDYiNjUV2djZ69eqlUOT3Zm/jAYDx48djwYIFSExMRH5+PlavXo2vvvoKr776qkm/H374Aa+9\n9hrOnz+PK1euYM6cOZg/fz5GjRpl7DNs2DBotVo89dRTKCgoaO6hEJEFmNPqx55zGgDs3bsXb731\nFs6ePYtr164hMTERixYtuuOj3utj7Nix+Pvf/47Vq1ejtLQUx48fR0hIiPHiiLmCyHYwV9QPc4Xl\nlMoVLHoR3UVMTAxu3LiB2NhYpUNBWVkZIiIilA7DzMcff4x169Zh/fr10Ov1Jq8tXrwYDg4OGDdu\nHG7cuKFQhE3HHsZz+vRpfPHFF3jxxRcRHx8PvV6P/v37IyEhAWvWrMHx48eNfZ2dnfHmm2/C29sb\nbm5uGD16NEaMGIHdu3fj0qVLxn7vvPMOHnroIQwZMgRVVVVKDIuI6oE57d5aQ05zc3PDuHHj4OXl\nBb1ejzFjxiAuLg47duxATk6OSd9Vq1ZBREz+/vvf/5r0WbduHbZs2YINGzagd+/ecHJygq+vL7Zu\n3WryTT9zBZFtYK64N+YK28oVLHoR2YgVK1YgLy9P6TBM/O9//8OMGTPw4YcfQqvVmr0eERGB8ePH\n48KFC5g0aZICETYtexjP4cOHUVNTg969e5u0Dx48GACwc+dOY9vmzZvNjqufnx8AmN3aPGvWLBw5\ncgSLFi2yRthEZGeY05STkpICR0dHk7a2bdsCAEpLSy1e3+eff45evXqhe/fu9+zLXEFElmCuUI49\n5QoWvYju4MCBAwgMDIRKpcLSpUsBAMuXL4dOp4Orqyu2bt2Kp59+GgaDAf7+/li7dq1x2cWLF0Or\n1aJdu3Z4/fXX4evrC61Wi4iICBw6dMjYLyEhAc7OzvDx8TG2vfnmm9DpdFCpVLh27RqA336ONnHi\nRJw6dQoqlQqhoaEAgB07dsBgMGDevHnNsUvMLF68GCKCYcOG3bHP3Llz0blzZ3z99ddIS0u76/pE\nBAsXLsQDDzwAjUYDT09PjBgxAj///LOxT32PAQBUV1dj5syZCAwMhIuLCx588EEkJSU1asy2Ph4H\nh99O+y4uLibtnTp1AgCTO73qcvLkSXh4eCAoKMik3dPTE5GRkVi0aBFviydqgZjT7q015rRaFy5c\ngIuLC4KDgy1arqKiAgcPHkSPHj3q1Z+5gqhlY664N+YKG8wVcpukpCSpo5nIpgCQpKSkRq8nJydH\nAMiSJUuMbdOmTRMAsmfPHrlx44bk5eVJv379RKfTSUVFhbHfuHHjRKfTybFjx6S8vFyys7Pl0Ucf\nFb1eL+fPnzf2e/7556V9+/Ym212wYIEAkKtXrxrbRo4cKSEhISb9UlJSRK/Xy+zZsxs91lGjRsmo\nUaMsWqZjx47SpUuXOl8LCQmRM2fOiIjIjz/+KA4ODtKhQwcpLi4WEZHU1FQZPny4yTIzZ84UZ2dn\nWbVqlRQUFMjRo0elV69e0rZtW7l8+bKxX32PwaRJk0Sj0cjGjRvl+vXrMnXqVHFwcJDDhw9bNE57\nGs/Ro0cFgMyYMcOkvaqqSgBIXFyc2TIVFRWSm5srS5YsEY1GI6tWrapz3VOmTBEAkpmZWe94RJru\n36tS6yeytoacn+vSmnJaQz7PtqacdquSkhLR6/WSkJBg0j5nzhzx9/cXDw8PUavV0qFDBxk+fLj8\n61//MvY5c+aMAJAePXpI//79xcfHRzQajYSFhcnSpUulpqbGbHsNyRXWvj7h9Q/ZA17/WI7XP/Vn\nC7niTseTd3oRNVBERAQMBgO8vb0RHx+PkpISnD9/3qSPk5OTsWrfpUsXLF++HEVFRVi5cmWTxBAT\nE4PCwkLMmDGjSdZniZKSEpw5cwYhISH37BseHo53330XZ8+exQcffFBnn7KyMixcuBDPPPMMXnjh\nBbi7u6N79+744osvcO3aNXz55Zdmy9ztGJSXl2P58uWIi4vDyJEj4eHhgenTp0OtVjd6/9vyeLp3\n747Bgwdj2bJl2Lt3L8rLy3H58mVs3rwZKpUKlZWVZssEBATA398fs2bNwieffIKxY8fWue7au8Wy\nsrLqHQ8RtQzMaa03pyUmJsLX1xdz5841aX/ppZewbds25OTkoLi4GGvXrsX58+cRGRmJ7OxsAP/3\nU3dvb2/MmzcP2dnZuHLlCkaMGIG33noLa9asMdsecwWR7WKuYK6wxVzBohdRE3B2dgaAOgsGt3rk\nkUfg6upqcruqrcrLy4OIwNXVtV79586di/vvvx/Lli3DgQMHzF7Pzs5GcXExHnnkEZP2Rx99FM7O\nzia3Rdfl9mNw4sQJlJaWmkyM6OLiAh8fnybZ/7Y8nnXr1mH06NF48cUX4eXlhccffxz/+Mc/ICJo\n06aNWf+cnBzk5eVhzZo1+Nvf/oaePXvWOb9C7XvhypUrFsVDRC0Lc9q92XIOuNXmzZuxfv167Ny5\n02wy5oCAAPTs2RNubm5wdnZGnz59sHLlSpSVlWHZsmUAAI1GAwDo2rUrIiIi4OXlBXd3d3z44Ydw\nd3ev84KNuYLIPjBX3BtzRcvIFSx6ETUzjUaDq1evKh1Go5WXlwP4v5PYvWi1WqxcuRIqlQq///3v\nUVZWZvJ67WNp3dzczJb18PBAUVGRRfGVlJQAAKZPnw6VSmX8O3fuXIMmX7ydLY/H3d0dX3zxBXJz\nc1FaWopTp07hL3/5CwDgvvvuM+uvVqvh7e2N6OhorFu3DtnZ2UhMTDTrVztPWO17g4jsH3Oa7eWA\nWuvWrcPHH3+Mffv2oUOHDvVapnv37nB0dMQvv/wCAPD19QUA4xw8tZydnREUFIRTp06ZrYO5gqj1\nYa5grgCUyxUsehE1o8rKShQUFMDf31/pUBqt9kRUXV1d72XCw8MxYcIEnDx5EnPmzDF5zcPDAwDq\nPLk3ZJ95e3sDAD777DOzR+imp6dbtK47safxHD58GAAwYMCAu/YLDQ2Fo6Oj8XblW1VUVAAwnySf\niOwTc5rt5oAlS5bgm2++wd69e+v8suNOampqUFNTY7zgc3NzQ6dOnXDs2DGzvlVVVXB3dzdrZ64g\nal2YK5grlM4VLHoRNaN9+/ZBRNCnTx9jm5OT0z1vC26J2rVrB5VKhRs3bli03Jw5cxAWFobMzEyT\n9m7dusHNzQ3//ve/TdoPHTqEiooKPPzwwxZtJyAgAFqtFkeOHLFoOUvZy3i++uorBAcHIzIyEgCQ\nn5+P5557zqzfyZMnUV1djYCAALPXat8L7du3t0qMRNSyMKfZXg4QEUyePBlZWVnYsmVLnXcX1Bo0\naJBZ2+HDhyEiCA8PN7aNHTsWmZmZOH36tLGttLQU586dq/PR9MwVRK0LcwVzBaBsrmDRi8iKampq\ncP36dVRVVeHo0aMYP348AgMD8fLLLxv7hIaG4tdff8WWLVtQWVmJq1ev4ty5c2br8vLywsWLF3H2\n7FkUFRWhsrISqampij2y19XVFR07dkRubq5Fy9Xe5uvo6GjWPnHiRGzevBnffPMNCgsLkZWVhTfe\neAO+vr4YN26cxdt55ZVXsHbtWixfvhyFhYWorq5Gbm4uLl26BACIj49H+/btkZGRYdG6bX08jz32\nGM6dO4eqqiqcPXsWkyZNQlpaGlasWGGcG0Cn02HXrl3Yu3cvCgsLUVlZiczMTLz00kvQ6XSYMGGC\n2Xpr3wt1JS4isn3MaeZsLQccO3YMn3zyCb766iuo1WqTn7+oVCp8+umnxr4XLlzAunXrUFBQgMrK\nSqSnp+PVV19FYGAg3njjDWO/CRMmICgoCC+//DLOnz+P/Px8TJ48GWVlZXVO3sxcQWTfmCvMMVco\nnCtuf5wjH9lL9gBN8MjeJUuWiI+PjwAQV1dXGTZsmCxbtkxcXV0FgHTq1ElOnTolX375pRgMBgEg\nQUFB8ssvv4jIb4/sVavV4ufnJ05OTmIwGGTEiBFy6tQpk+3k5+fLgAEDRKvVSnBwsLz99tvy3nvv\nCQAJDQ01Pt43IyNDgoKCxMXFRfr27SuXL1+W7du3i16vl7lz5zZqrCINe2RvQkKCqNVqKS0tNbZt\n3rxZQkJCBIC0bdtW3nrrrTqXfe+998we2VtTUyMLFiyQTp06iVqtFk9PT4mLi5MTJ04Y+1hyDG7e\nvCmTJ0+WwMBAcXJyEm9vbxk5cqRkZ2eLiEhcXJwAkJkzZ95xjPY2HhGRgQMHioeHhzg5OYmnp6fE\nxMTU+RjjYcOGSXBwsLi5uYlGo5GQkBCJj4+XrKysOtcbExMjfn5+dT52+G6a4t+rkusnsraGnJ9v\n19pyWkM+z7aGnJaVlSUA7vi3YMECY9+JEydKSEiI6HQ6cXJyEn9/f3nttdfk4sWLZuvNycmRZ599\nVjw9PUWj0chjjz0mqampdcbQkFxh7esTXv+QPeD1j+V4/VM3W80VdzqeLHqRXWoJF7njxo0TLy8v\nRWOwRENO+idPnhQnJydZtWqVlaKyrurqaunXr5+sWLFC6VCahJLjuXbtmmi1Wvn0008tXpZFL6K7\na4qiV2PZWk5ryOdZ5jTra2iuYNGL6N5awucdW8sVvP5pmRqaK+50PPnzRiIrsmSSQ1sUGhqK2bNn\nY/bs2SguLlY6HItUV1djy5YtKCoqQnx8vNLhNJrS45k1axZ69OiBhISEZt82ETUP5rSWS+kcUF/M\nFUT2j7mi5WqtuYJFLyJqlClTpmD06NGIj4+3eFJHJe3btw+bNm1CamoqXF1dlQ6n0ZQcz8KFC3Hk\nyBFs374darW6WbdNRNSUmNOsh7mCiOwFc4X1WCNXNEnRa/v27XB3d0dycnJTrE4xs2fPRpcuXWAw\nGKDRaBAaGor333+/QRXcgwcP4oEHHoCDgwNUKhXat2+PuXPnWiHqhtu0aRM6duxonJDOx8cHL7zw\ngtJh2YWpU6di5cqVuHHjBoKDg7Fx40alQ7KqefPmISEhAR999JHSodTbU089hdWrV8PHx0fpUJqE\nUuPZunUrbt68iX379sHT07NZt20t9pLT5s+fj7CwMLi4uECn0yEsLAwzZsxAYWGhxetiTmvdmNNa\nvpae05grbEd5eTnCwsIwffp0i5dlrmjdmCtavtaaK5yaYiUi0hSrUdzevXvx1ltvIT4+Hmq1Gqmp\nqXjhhReQlZWF1NRUi9bVp08fHD9+HIMHD8bOnTtx4sQJeHh4WCnyhhk5ciRGjhyJ0NBQXLt2DZcv\nX1Y6JLuRmJiIxMREpcNoVtHR0YiOjlY6DGpmw4cPx/Dhw5UOo0nZS0774Ycf8Nprr+HFF1+Ei4sL\nUlNT8fzzz+PQoUPYtWuXRetiTmvdmNOosZgrbMe0adNw4sSJBi3LXNG6MVdQY1krVzTJnV4xMTG4\nceMGYmNjm2J1jVJWVoaIiIgGLevm5oZx48bBy8sLer0eY8aMQVxcHHbs2IGcnJwmjrT5NWbfEBG1\nFvaS05ydnfHmm2/C29sbbm5uGD16NEaMGIHdu3cbH1tty5jTiEhJ9pIrbvXjjz/iv//9bxNE1HIw\nVxCR3c3ptWLFCuTl5TVo2ZSUFDg6Opq0tW3bFgBQWlra6NiU1ph9Q0REza8x5+3NmzdDq9WatPn5\n+QGAzU28WhfmNCKi3zTF+bCsrAzvvfceFi1a1ERRtQzMFUTU6KLXgQMHEBgYCJVKhaVLlwIAli9f\nDp1OB1dXV2zduhVPP/00DAYD/P39sXbtWuOyixcvhlarRbt27fD666/D19cXWq0WEREROHTokLFf\nQkICnJ2dTX57+uabb0Kn00GlUuHatWsAgPHjx2PixIk4deoUVCoVQkNDGzs8XLhwAS4uLggODja2\n7dixAwaDAfPmzbN4fba+b3744Qd06dIF7u7u0Gq16N69O3bu3AkAePXVV42/jw8JCUFmZiYA4JVX\nXoGrqyvc3d2xbds2AL89OWLmzJkIDAyEi4sLHnzwQSQlJQEAPvnkE7i6ukKv1yMvLw8TJ06En59f\ng2+1JiKqL3vPaSdPnoSHhweCgoKMbcxpzGlEZBl7zBXTpk0z3h1cF+YK5goimyW3SUpKkjqa7yon\nJ0cAyJIlS4xt06ZNEwCyZ88euXHjhuTl5Um/fv1Ep9NJRUWFsd+4ceNEp9PJsWPHpLy8XLKzs+XR\nRx8VvV4v58+fN/Z7/vnnpX379ibbXbBggQCQq1evGttGjhwpISEhFsV/JyUlJaLX6yUhIcGkPSUl\nRfR6vcyePfue6xg0aJAAkOvXrxvbWtq+CQkJEXd393vvEBHZsGGDzJo1S3799VfJz8+XPn36SJs2\nbUy24ejoKBcuXDBZ7rnnnpNt27YZ/3/SpEmi0Whk48aNcv36dZk6dao4ODjI4cOHTfbRO++8I0uW\nLJFnnnlGjh8/Xq8YRUQASFJSUr37k8ioUaNk1KhRSodBrZC1/71aun57y2kVFRWSm5srS5YsEY1G\nI6tWrTJ5nTmt5ec0np8t15DPs9RyWft4tvbrnwMHDsiwYcNEROTq1asCQKZNm2bSh7mi5ecKXv9Y\njvnVvtzpeFr9540REREwGAzw9vZGfHw8SkpKcP78eZM+Tk5OeOCBB6DRaNClSxcsX74cRUVFWLly\npbXDu6vExET4+vqaPXUkJiYGhYWFmDFjRqPWb4v7ZtSoUfjzn/8MT09PeHl5YdiwYcjPz8fVq1cB\nAG+88Qaqq6tN4issLMThw4cxZMgQAL89FWb58uWIi4vDyJEj4eHhgenTp0OtVpuN6+OPP8Zbb72F\nTZs2ISwsrPkGSkRUB1s8bwcEBMDf3x+zZs3CJ598grFjx5q8zpzGnEZETcuWzodlZWUYP348li9f\nftd+zBXMFUS2qkme3lhfzs7OAIDKysq79nvkkUfg6uqKn3/+uTnCqtPmzZuxfv167Nq1C3q93urb\ns6V9cyu1Wg3gt9t1AeDJJ59E586d8de//hVTp06FSqXCunXrEB8fb5wv7cSJEygtLUW3bt2M63Fx\ncYGPj0+Tjmvs2LFmF3d0byqVSukQiGyCrZy3c3JyUFBQgMzMTEyZMgVffvkl9u7di3bt2lltm7ay\nb27XknPaxo0beX5uAO4zUlpLPx9OnToVf/zjH41zPjanlr5v7qQl5wpe/zQMc4X9GDVqlFlbsxa9\nLKHRaIzV8+a2bt06LFy4EPv27cN9992nSAx3o+S++fbbb7FgwQJkZ2ejsLDQLEmpVCq8/vrrmDBh\nAvbs2YOoqCj8/e9/x+rVq419SkpKAADTp0/H9OnTTZb39fVtsljHjx+P8PDwJlufvfvss88AAO++\n+67CkVBr0xo+nCl53lar1fD29kZ0dDSCg4PRuXNnJCYmtpjJipnT6qdPnz48P1sgPT0dixYtMs6X\nQ7at9njau+Y+Hx44cABZWVlYuHBhs22zoZgr6ofXP5bh9Y99qT2et2uRRa/KykoUFBTA39+/2be9\nZMkS7Ny5E3v37oWbm1uzb/9emnvf7N+/Hz/99BPeffddnD9/HnFxcXjmmWfw17/+Fffddx+WLFmC\n999/32SZl19+GVOnTsXXX3+NgIAAGAwGk0mTayfI/OyzzzB+/HirxR4eHo4xY8ZYbf32ZsOGDQDA\nfUbNzt6LXkrmtNuFhobC0dER2dnZSocCgDnNEv7+/jw/W2jRokXcZ3bE3oteSuSKFStWYM+ePXBw\nMJ/xZt68eZg3bx4OHz6MRx55pNliqgtzRf3x+scyvP6xL7XH83Ytsui1b98+iAj69OljbHNycrrn\nra+NISL44IMPcP36dWzZsgVOTi1y1zT7vvnpp5+g0+kAAFlZWaisrMSf/vQndOzYEUDdt4J6enpi\n7NixWLduHfR6PV577TWT1wMCAqDVanHkyBGrxExE1JIokdPy8/Px9ttvY82aNSbtJ0+eRHV1NQIC\nAqy2bUswpxER/UaJXLFy5Uqz+aSuXbsGb29vTJs2zWxeY6UwVxBRY1h9Ivv6qKmpwfXr11FVVYWj\nR49i/PjxCAwMxMsvv2zsExoail9//RVbtmxBZWUlrl69inPnzpmty8vLCxcvXsTZs2dRVFRU75Ph\nsWPH8Mknn+Crr76CWq02Pnq29u/TTz819k1NTW3wI3stpdS+qaysxJUrV7Bv3z7jST8wMBAAkJaW\nhvLycpw8edLk8cG3euONN3Dz5k2kpKQgNjbW5DWtVotXXnkFa9euxfLly1FYWIjq6mrk5ubi0qVL\nlu4iIqIWpSXkNJ1Oh127dmHv3r3Gn2JkZmbipZdegk6nw4QJE4x9mdOY04io+bWEXGEJ5grmCiKb\ndfvjHC19ZO+SJUvEx8dHAIirq6sMGzZMli1bJq6urgJAOnXqJKdOnZIvv/xSDAaDAJCgoCD55Zdf\nROS3x9Kq1Wrx8/MTJycnMRgMMmLECDl16pTJdvLz82XAgAGi1WolODhY3n77bXnvvfcEgISGhhof\nYZuRkSFBQUHi4uIiffv2lcuXL9drHFlZWQLgjn8LFiww9t2+fbvo9XqZO3fuHdd38OBB6dq1qzg4\nOAgA8fHxkXnz5rWoffP5559LSEjIXccNQDZv3mzc1uTJk8XLy0s8PDxk9OjRsnTpUgEgISEhJo8R\nFhHp2bOnTJkypc79c/PmTZk8ebIEBgaKk5OTeHt7y8iRIyU7O1vmz58vLi4uAkACAgJk1apV9TqG\ntwIf2WsxPrKXlGLtf6+WrN9ecpqIyLBhwyQ4OFjc3NxEo9FISEiIxMfHS1ZWlkk/5rSWn9N4frac\npZ9nqWWz9vFsrdc/dbl69aoAkGnTppm0M1e0/FzB6x/LMb/alzsdT5WIyK1FsPXr12Ps2LG4rdlq\nXn/9dWzYsAH5+fnNsj1bYuv7JiYmBkuXLkVwcHCzb1ulUiEpKYm/z7bA6NGjAdz5t9BE1mLtf6/N\neT6w9fO2Ndn6vlEyp/H8bLnm/jxL1mXt48nrn5bD1vcNr39sC/OrfbnT8WwRP2+sfdwrmbOlfXPr\n7cJHjx6FVqtV5IRPRKQkWzpvNzdb2jfMaURkTbZ0PmxutrRvmCuIWr4WUfSylp9//tlsbq66/uLj\n45UO1S5MnjwZJ0+exC+//IJXXnkFc+bMUTokakZpaWmYMmUKNm3ahI4dOxr/ff3ud78z6xsdHQ29\nXg9HR0d07doVGRkZCkRcP/Y2HuC3D2gzZ85Ex44d4ezsDD8/P0yaNAllZWUm/ebPn4+wsDC4uLhA\np9MhLCwMM2bMQGFhobHPtm3bMH/+fJv6gGqrmNOaF3Na62avOQ0AZs+ejS5dusBgMECj0SA0NBTv\nv/8+iouLTfrNnTu3znNMt27dzNZZWVmJxMREhIaGwtnZGR4eHujWrRvOnj0LgLmiOTFXNC/mitaN\nucJGcsXtv3dszjkQpkyZIs7OzgJAOnToIBs2bGiW7doCW9w306ZNEwcHBwkICJBt27YpGgv4m3aL\nNeY37TNnzpTY2FgpLCw0toWEhEibNm0EgKSkpJgtk5qaKsOHD29wvM3Nnsbzpz/9SbRaraxdu1YK\nCwvlu+++E4PBIM8995xJv5iYGPn0008lLy9PioqKZP369aJWq2XgwIEm/RYtWiSRkZFy/fr1BsVj\n7X+vzXU+sMXzdnOxxX3TknIa5xyxXGM+z9p7TouMjJRly5ZJfn6+FBYWSlJSkqjVahk8eLBJvzlz\n5tQ511HXrl3N1hkXFyf333+/HDx4UCorK+XixYsybNgwkzkEG5MrWtqcXo1hi+fD5mKL+6Yl5Qpe\n/1iO1z93Zou54k7HU9GiF5G1KH3SLy0tlfDwcJvaRkNP+h999JF07txZysrKTNpDQkJk9erV4uDg\nIH5+flJQUGDyui2d9EXsZzynTp0SBwcH+eMf/2jSPn36dAEgx44dM7bFxcWZHdfRo0cLALl48aJJ\ne0JCgoSHh0tlZaXFMdlL0YvIWlpC0cvW8lpDP8+2hpwWExMjVVVVJm1jxowRACaTcs+ZM6dek2mv\nXbtWVCqVHD169J59G5or7KnoRWQtSn/esbU8IcLrn7uxxVxxp+Np1z9vJO6+iTgAACAASURBVFLK\nihUrkJeXZ/PbuJf//e9/mDFjBj788ENotVqz1yMiIjB+/HhcuHABkyZNUiDCpmUP4zl8+DBqamrQ\nu3dvk/bBgwcDAHbu3Gls27x5s9lx9fPzAwCzW5tnzZqFI0eOYNGiRdYIm4gU1hryWmvJaSkpKXB0\ndDRpa9u2LQCgtLTU4vV9/vnn6NWrF7p3737PvswVRParNeQJgLkCsL1cwaIXEQARwcKFC/HAAw9A\no9HA09MTI0aMwM8//2zsk5CQAGdnZ/j4+Bjb3nzzTeh0OqhUKly7dg0AMH78eEycOBGnTp2CSqVC\naGgoFi9eDK1Wi3bt2uH111+Hr68vtFotIiIicOjQoSbZBgDs2LEDBoMB8+bNs+r+qrV48WKICIYN\nG3bHPnPnzkXnzp3x9ddfIy0t7a7rq89xWL58OXQ6HVxdXbF161Y8/fTTMBgM8Pf3x9q1a03WV11d\njZkzZyIwMBAuLi548MEHkZSU1Kgx2/p4HBx+O+27uLiYtHfq1AkAcPz48bsuf/LkSXh4eCAoKMik\n3dPTE5GRkVi0aBGflkbUAjCvWa415rRaFy5cgIuLi8UTcFdUVODgwYPo0aNHvfozVxC1HMwTDcNc\nYYO54vZbv3h7L9kDWHh778yZM8XZ2VlWrVolBQUFcvToUenVq5e0bdtWLl++bOz3/PPPS/v27U2W\nXbBggQCQq1evGttGjhwpISEhJv3GjRsnOp1Ojh07JuXl5ZKdnS2PPvqo6PV6k1tEG7ONlJQU0ev1\nMnv27HqPvVZDbu/t2LGjdOnSpc7XQkJC5MyZMyIi8uOPP4qDg4N06NBBiouLRaTu23vrexymTZsm\nAGTPnj1y48YNycvLk379+olOp5OKigpjv0mTJolGo5GNGzfK9evXZerUqeLg4CCHDx+2aJz2NJ6j\nR48KAJkxY4ZJe1VVlQCQuLg4s2UqKiokNzdXlixZIhqN5o63ME+ZMkUASGZmZr3jEeHPG4nupSHn\n59ae1xryebY15bRblZSUiF6vl4SEBJP2OXPmiL+/v3h4eIharZYOHTrI8OHD5V//+pexz5kzZwSA\n9OjRQ/r37y8+Pj6i0WgkLCxMli5dKjU1NWbba0iu4M8bie6N1z+8/hFp3bmCP28kuoOysjIsXLgQ\nzzzzDF544QW4u7uje/fu+OKLL3Dt2jV8+eWXTbYtJycnYxW/S5cuWL58OYqKirBy5comWX9MTAwK\nCwsxY8aMJlnf3ZSUlODMmTMICQm5Z9/w8HC8++67OHv2LD744IM6+zTkOERERMBgMMDb2xvx8fEo\nKSnB+fPnAQDl5eVYvnw54uLiMHLkSHh4eGD69OlQq9WN3t+2PJ7u3btj8ODBWLZsGfbu3Yvy8nJc\nvnwZmzdvhkqlMnn0dq2AgAD4+/tj1qxZ+OSTTzB27Ng61117t1hWVla94yGipse8ZrnWnNMSExPh\n6+uLuXPnmrS/9NJL2LZtG3JyclBcXIy1a9fi/PnziIyMRHZ2NoD/+6m7t7c35s2bh+zsbFy5cgUj\nRozAW2+9hTVr1phtj7mCSHnMEw3DXGGbuYJFL2r1srOzUVxcjEceecSk/dFHH4Wzs7PJ7bdN7ZFH\nHoGrq6vJ7au2Ii8vDyICV1fXevWfO3cu7r//fixbtgwHDhwwe72xx8HZ2RkAjEWbEydOoLS01ORx\nuS4uLvDx8WmS/W3L41m3bh1Gjx6NF198EV5eXnj88cfxj3/8AyKCNm3amPXPyclBXl4e1qxZg7/9\n7W/o2bNnnfMp1L4Xrly5YlE8RNS0mNcs11pz2ubNm7F+/Xrs3LkTer3e5LWAgAD07NkTbm5ucHZ2\nRp8+fbBy5UqUlZVh2bJlAACNRgMA6Nq1KyIiIuDl5QV3d3d8+OGHcHd3r/OCjbmCSHnMEw3DXGGb\nuYJFL2r1CgoKAABubm5mr3l4eKCoqMiq29doNLh69apVt2EN5eXlAP7vJHYvWq0WK1euhEqlwu9/\n/3uUlZWZvN7Ux6GkpAQAMH36dKhUKuPfuXPnGjT54u1seTzu7u744osvkJubi9LSUpw6dQp/+ctf\nAAD33XefWX+1Wg1vb29ER0dj3bp1yM7ORmJiolm/2nnCat8bRKQM5jXLtcactm7dOnz88cfYt28f\nOnToUK9lunfvDkdHR/zyyy8AAF9fXwAwzrlTy9nZGUFBQTh16pTZOpgriJTHPNEwzBUd6rVMS8sV\nLHpRq+fh4QEAdZ5UCgoK4O/vb7VtV1ZWWn0b1lJ7Iqqurq73MuHh4ZgwYQJOnjyJOXPmmLzW1MfB\n29sbAPDZZ59BREz+0tPTLVrXndjTeA4fPgwAGDBgwF37hYaGwtHR0Xi78q0qKioAmE+ST0TNi3nN\ncq0tpy1ZsgTffPMN9u7dW+eXHXdSU1ODmpoa4wWfm5sbOnXqhGPHjpn1raqqgru7u1k7cwWR8pgn\nGoa5on5aWq5g0YtavW7dusHNzQ3//ve/TdoPHTqEiooKPPzww8Y2JyenOuc8aqh9+/ZBRNCnTx+r\nbcNa2rVrB5VKhRs3bli03Jw5cxAWFobMzEyTdkuOQ30EBARAq9XiyJEjFi1nKXsZz1dffYXg4GBE\nRkYCAPLz8/Hcc8+Z9Tt58iSqq6sREBBg9lrte6F9+/ZWiZGI6od5zXKtJaeJCCZPnoysrCxs2bKl\nzrsLag0aNMis7fDhwxARhIeHG9vGjh2LzMxMnD592thWWlqKc+fO1floeuYKIuUxTzQMc4U5W8gV\nLHpRq6fVajFx4kRs3rwZ33zzDQoLC5GVlYU33ngDvr6+GDdunLFvaGgofv31V2zZsgWVlZW4evUq\nzp07Z7ZOLy8vXLx4EWfPnkVRUZHxJF5TU4Pr16+jqqoKR48exfjx4xEYGIiXX365SbaRmprabI/s\ndXV1RceOHZGbm2vRcrW3+To6Opq11/c41Hc7r7zyCtauXYvly5ejsLAQ1dXVyM3NxaVLlwAA8fHx\naN++PTIyMixat62P57HHHsO5c+dQVVWFs2fPYtKkSUhLS8OKFSuMcwPodDrs2rULe/fuRWFhISor\nK5GZmYmXXnoJOp0OEyZMMFtv7XuhrsRFRM2Hec1yrSWnHTt2DJ988gm++uorqNVqk5+/qFQqfPrp\np8a+Fy5cwLp161BQUIDKykqkp6fj1VdfRWBgIN544w1jvwkTJiAoKAgvv/wyzp8/j/z8fEyePBll\nZWV1Tt7MXEGkPOaJhmGusNFccfvjHPnIXrIHsPCRvTU1NbJgwQLp1KmTqNVq8fT0lLi4ODlx4oRJ\nv/z8fBkwYIBotVoJDg6Wt99+W9577z0BIKGhocZH72ZkZEhQUJC4uLhI37595fLlyzJu3DhRq9Xi\n5+cnTk5OYjAYZMSIEXLq1Kkm28b27dtFr9fL3LlzLd5nDXlkb0JCgqjVaiktLTW2bd68WUJCQgSA\ntG3bVt566606l33vvffMHtlbn+OwbNkycXV1FQDSqVMnOXXqlHz55ZdiMBgEgAQFBckvv/wiIiI3\nb96UyZMnS2BgoDg5OYm3t7eMHDlSsrOzRUQkLi5OAMjMmTPvOEZ7G4+IyMCBA8XDw0OcnJzE09NT\nYmJi6nyM8bBhwyQ4OFjc3NxEo9FISEiIxMfHS1ZWVp3rjYmJET8/vzofO3w3lv57tZS1109kbQ05\nP7f2vNaQz7OtIadlZWUJgDv+LViwwNh34sSJEhISIjqdTpycnMTf319ee+01uXjxotl6c3Jy5Nln\nnxVPT0/RaDTy2GOPSWpqap0xNCRXWPv6hNc/ZA94/cPrH5HWnSvudDxZ9CK71BIvcseNGydeXl5K\nh3FHDTnpnzx5UpycnGTVqlVWisq6qqurpV+/frJixQqlQ2kSSo7n2rVrotVq5dNPP7V4WRa9iO6u\nIefn5tCS81pDPs8yp1lfQ3MFi15E99YSP++05DwhwuuflqqhueJOx5M/byRqRpZMemgLQkNDMXv2\nbMyePRvFxcVKh2OR6upqbNmyBUVFRYiPj1c6nEZTejyzZs1Cjx49kJCQ0OzbJiLl2FNeY06zPuYK\notbHnvIEwFzRHJo6V7DoRUSNMmXKFIwePRrx8fEWT+qopH379mHTpk1ITU2Fq6ur0uE0mpLjWbhw\nIY4cOYLt27dDrVY367aJiJoSc5r1MFcQkb1grrAea+QKFr2ImsHUqVOxcuVK3LhxA8HBwdi4caPS\nITWpefPmISEhAR999JHSodTbU089hdWrV8PHx0fpUJqEUuPZunUrbt68iX379sHT07NZt01EyrHn\nvMac1vSYK4haH3vOEwBzhTVYK1c4NdmaiOiOEhMTkZiYqHQYVhUdHY3o6Gilw6BmNnz4cAwfPlzp\nMIiomdl7XmNOa1rMFUStj73nCYC5oqlZK1fwTi8iIiIiIiIiIrI7LHoREREREREREZHdYdGLiIiI\niIiIiIjsDoteRERERERERERkd+44kf369eubMw6iJpeenq50CDYlNzcXAP/t17p69Sr0ej20Wq3S\noVAT4PmAbBnPz5ar/TfPfWYfmusczvdL/Vy5cgVeXl5Qq9VKh0K34ecdyzC/2pfc3Fz4+/ubtatE\nRG5tWL9+PcaOHdtsgRERESUlJWHMmDFWWbdKpbLKeomIqHnddtnSZHj9Q0RkH0aNGoUNGzaYtJkV\nvYiI6Lc7vfbt24e0tDTs2rULZ8+ehU6nQ3h4OKKiohAVFYVevXqxoEJEiqstGPObaiJqKqdPn0Za\nWhqSk5ORlpaG8vJydOzYEUOHDkVsbCz69esHjUajdJhERPfEohcRUT3UfvhLS0vDnj178Ouvv6Jd\nu3aIjIxEVFQUBg8ejMDAQKXDJKJWiEUvImqskpIS7N27FykpKdixYwfOnz+PNm3a4Mknn0RUVBSG\nDBlS58+GiIhaOha9iIgsVF1djSNHjhiLYAcOHDB+A1p7F1hUVBQ8PT2VDpWIWgEWvYioIbKzs5GS\nkoK0tDTs378f1dXV6NGjh/FzTP/+/eHkdMcpoImIbAKLXkREjVRWVoZ//vOfxiJYZmYmVCqVyQdH\n/gyAiKyFRS8iqo9bp25ISUnBxYsXjXetDx06FEOHDoWXl5fSYRIRNSkWvYiImti1a9fw3XffIS0t\nDbt378aZM2fg6uqKiIgIzgdGRE2ORS8iqsutd6YnJycjPT0dDg4O6N27N2JjY/l5hIhaBRa9iIis\n7Nb5wPbu3Yv8/Hx4e3ujf//+iIqKwqBBgxAUFKR0mERko1j0IqJaly9fxq5du5CSkoLdu3ejoKDA\nZPqFQYMGwWAwKB0mEVGzYdGLiKgZ1dTUIDMz867zgT311FP8eQER1RuLXkSt1+1TLGRkZMDFxcV4\nd3lsbCy6dOmidJhERIph0YuISEFlZWX46aefjB9Y9+3bBxExmQ+sb9++0Gq1SodKRC0Ui15Ercut\nd5Dv2LEDRUVFxi/Phg4diujoaM4jSkT0/2PRi4ioBSkqKsKhQ4eMH2Z/+uknuLi44PHHHzcWwXr2\n7AkHBwelQyWiFoJFLyL7VlJSgvT0dCQnJ2Pbtm04e/Ys3Nzc0L9/f8TGxuLpp59GQECA0mESEbVI\nLHoREbVgly5dwoEDB0yetNS2bVsMGDAAUVFRGDhwIIKDg5UOk4gUxKIXkf3Jzs5GSkoK0tLSsH//\nflRVVaFnz57GL8AiIyOhVquVDpOIqMVj0YuIyIbc+pOGnTt3orCwkPOBEbVyLHoR2b5bn/z87bff\n4sKFCyYPvYmNjYWvr6/SYRIR2RwWvYiIbFRVVRX+85//GItg33//Paqrq02+CeZ8YET2j0UvIttT\nXV2NI0eOIC0tDcnJyUhPT4dKpUKPHj0wdOhQxMbGolevXlCpVEqHSkRk01j0IiKyE8XFxTh48CDn\nAyNqZVj0IrINV65cwf79+5GcnIyUlBRcv34dwcHBGDhwIKKiohAdHQ13d3elwyQisissehER2anL\nly/jhx9+MPmpRJs2bfDkk08iKioKjz/+OLp27ap0mETUSCx6EbVMVVVVOHjwoHFuroyMDGi1WpMv\nox5++GGlwyQismssehERtRK3zge2a9cu3LhxA76+vujbt6/xMef33Xef0mESkYVY9CJqOe429+bQ\noUMxcOBATjtARNSMWPQiImqFbp8PbP/+/aioqDCZFH/w4MHQ6/VKh0pE98CiF5FySktL8eOPPyIt\nLQ3btm3D8ePHodPpMGDAAMTGxmLw4MEIDAxUOkwiolaLRS8iIkJJSQnS09ONRbCMjAw4OjrioYce\n4uPRiVo4Fr2Imtfp06eN83L98MMPqKysNHmIzBNPPAFnZ2elwyQiIrDoRUREdaidbDctLQ2pqanI\nycmBm5sb+vTpw3lIiFoYFr2IrCs/Px979+5FWloatm/fjtzcXHh7e6N///6cHoCIqIVj0YuIiO7p\nXvOBxcTEwM/PT+kwiVolFr2ImlZ1dTWOHDlizHvff/89ampq0KNHD2ORKyIigk9DJiKyASx6ERGR\nRW6/GKhrPrBBgwbBYDAoHSpRq8CiF1Hj5eXl4fvvv0dycjK+/fZb/Prrr/Dx8cHAgQMRGxuLgQMH\nwsPDQ+kwiYjIQix6ERFRo3A+MCJlsehFZLmqqiocPHgQKSkpxtyl1Wrx+OOPG3NXr169oFKplA6V\niIgagUUvIiJqUrXflqelpWHHjh04f/682XxgvJAgajosehHVz60/1d+5cycKCwv51GIiIjvHohcR\nEVnVrRcZu3fvRkFBAXx8fNCvXz9ERUVhyJAh8Pf3VzpMIpvFohdR3UpLS/Hjjz8ac9BPP/0EnU6H\n8PBwDB06FMOHD0eHDh2UDpOIiKyIRS8iImo2t88H9sMPP+DmzZsm37RHR0fD3d1d6VCJbAaLXkT/\np/aLluTkZOzevRs3b95Ely5dEBsbi6ioKDzxxBNwdnZWOkwiImomLHoREZFibv8WPiMjAw4ODsYn\nZPEChejeWPSi1qy4uBjfffcdUlJSkJqaipycHLRt2xYDBgzg04WJiIhFLyIiajmuXr2Kffv2Gedb\nOXfunPGnKJwPjKhuLHpRa1JTU4PMzEzjlyXff/89ampqjF+WDB06FBEREXBwcFA6VCIiagFY9CIi\nohbr1vnA0tLScP36dbRv3x5PPPEEoqKi8PTTTyMgIEDpMIkUxaIX2btbvxBJTk7GpUuX0L59e0RH\nRxt/tujp6al0mERE1AKx6EVERDahPvOBDRw4EB4eHkqHStSsWPQie1N7vk9OTkZKSgoyMjLg6OiI\n3r17G4tcvOuXiIjqg0UvIiKySbfPB5aZmQmVSmUyH1i/fv2g0WiUDpXIqlj0Intw5swZ7N69G2lp\nadi1axdu3Lhh8qXGoEGDYDAYlA6TiIhsDIteRERkF279+cuuXbtw9uxZuLq6IiIigvOBkV1j0Yts\nUVlZGf75z38av7j46aefTM7Zw4YNwwMPPKB0mEREZONY9CIiIrt063xge/bswa+//op27dohMjLS\neNdAUFCQ0mESNRqLXmQras/LycnJSEtLQ3l5OTp27IihQ4ciNjaWd+cSEVGTY9GLiIjs3u1P+zpw\n4IDxYqv2LjBOhEy2ikUvaqlKSkqQnp6O5ORkbN26FefOnUObNm3w5JNPIioqCkOGDIG/v7/SYRIR\nkR1j0YuIiFqd239Ww/nAyJax6EUtSXZ2NlJSUpCWlob9+/ejurra5Nzav39/ODk5KR0mERG1Eix6\nERFRq3ft2jV89913xiLY6dOnzeYD69mzJxwcHJQOlcgMi16kpFvPnykpKbh48aLxp+RDhw7F0KFD\n4eXlpXSYRETUSrHoRUREdJtb5wPbu3cv8vPz4e3tjf79+yMqKgrR0dHo0KGD0mFSK7R69WqsWLEC\nNTU1xrYzZ84AAIKDg41tDg4O+MMf/oDnn3++2WMk+1ZdXY0jR44Y5+ZKT0+Hg4MDevfujdjYWD40\nhIiIWhQWvYiIiO7h1iLYjh07UFRUZDIf2FNPPcU7GahZHD16FA899FC9+v7nP//Bgw8+aOWIqDW4\nfPkydu3ahZSUFOzevRsFBQUm58BBgwbBYDAoHSYREZEZFr2IiIgsUFVVhf/85z/GItj333+Pmpoa\nkzlr+vbtC61Wq3SoZKfCwsJw4sSJu/YJDQ3FyZMnmykisjfl5eU4cOCA8TyXkZEBFxcX40++Y2Nj\n0aVLF6XDJCIiuicWvYiIiBqhuLgYBw8eNF4c/vTTT3BxccHjjz/O+cDIKubNm4cPP/wQlZWVdb6u\nVqsxa9YsTJ06tZkjI1t2tztahw4diujoaD7cg4iIbA6LXkRERE3o0qVLxjskvv32W1y4cAFt27bF\ngAEDEBUVhYEDB5rMvURkqdOnTyM0NBR3+wh38uRJhIaGNmNUZGtKSkqQnp6O5ORkbNu2DWfPnoWb\nmxv69++P2NhYPP300wgICFA6TCIiokZh0YuIiMiKbr17YufOnSgsLDSZC+fJJ59EmzZtGr2dEydO\n4P7772+CiMkWPPLII8jIyDArfKlUKjz88MM4fPiwQpGRNVRWVmLfvn0YOHBgo9aTnZ2NlJQUpKWl\nYf/+/aiqqkLPnj2N56PIyEio1eomipqIiEh5LHoRERE1k7rmA6uurja56GzofGD+/v7o168fli1b\nxkn1W4HFixdj4sSJqKqqMml3cnLCwoUL8fbbbysUGTW148ePY+zYsbh69SouXrxo0VMR8/PzsXfv\nXpM7T299Em1sbCx8fX2tGD0REZGyWPQiIiJSyO3zgWVkZECr1eLxxx/H448/jr59+6J///5wcnK6\n63pOnDiBsLAwODo6wsvLC//v//0/DBkypJlGQUrIy8uDr68vampqTNodHBxw8eJFtG/fXqHIqKmI\nCJYtW4aJEyeipqYGVVVVOHr0KLp3737HZaqrq3HkyBHjOWXfvn0QEfTo0QNDhw5FbGwsevXqZVHh\njIiIyJax6EVERNRCXL58GT/88APS0tKwfft25ObmQq/Xo3fv3sY7wR5++GGz5ZYuXYp3330XVVVV\ncHBwQE1NDUaOHImvvvoKnp6eCoyEmsOTTz6J/fv3o7q6GgDg6OiIyMhI7NmzR+HIqLGuXLmCV155\nBTt37jQWNtVqNRITEzFp0iSzvvv370dycjJSUlJw/fp1BAcHY+DAgYiKikJ0dDTc3d2VGAYREZHi\nWPQiIiJqoW6dD2zXrl24ceMGfH190bdvX+MT1e677z4MGzYM27dvNxY/gN8ukNu2bYtVq1bhqaee\nUnAUZC0rV67Eq6++aiyKODo64uuvv8bLL7+sbGDUKJs2bcIf/vAHlJaWmjyhU6VSITIyErt378bB\ngweNc3Pdeofo3YrjRERErRGLXkRERDagoqLC5KeQhw8fRk1NDR588EGcOHECZWVlZss4OjqipqYG\nr776KhYuXAg3NzcFIidrKSwshLe3NyoqKgD8VujMy8uDh4eHwpFRQxQWFmLixIn4+uuvjXds3s7R\n0REajQalpaXo2rUrBg8ejEGDBqFfv34NmguQiIjI3rHoRUREZIMKCwvx3XffYfXq1diwYcNd+zo5\nOcHHxwerV6/GE0880UwRUnMYPnw4tm/fDgCIiYnBli1bFI6IGiI9PR3PPvssLl68aHJ3V13efvtt\nvPfeewgICGim6IiIiGyXg9IBEBERkeUMBgOGDx+OBx98EGq1+q59q6qqcOnSJfTv3x8JCQm4efNm\nM0VJ1vbCCy+guroa1dXVeP7555UOhyxUVVWFWbNmoW/fvsjNzb1nwav23zoLXkRERPXDO72IiIhs\nWL9+/fDPf/4T9U3nTk5OCAkJwZo1a9CrVy8rR0fWVl5ejrZt20JEcO3aNbi4uCgdEtVTdnY2nn32\nWRw/fhxVVVX1Xi4wMBDnzp2zYmRERET2g0UvIqLbrF+/HmPHjlU6DCIiIiK7lJSUhDFjxigdBhG1\nAk5KB0BE1FIlJSUpHQK1QOnp6Vi0aFGLeH8cOXIEH330kVm7s7MzdDodDAYDDAYDPDw8oNfr4ebm\nBr1eD71eD4PBAL1ej/bt20Oj0Vg91rFjx2L8+PEIDw+3+rZamyNHjkClUuGhhx5SOhSqh8rKSly4\ncAGlpaUoLi5GaWkpSkpKjP8tKSlBcXExioqKjK+XlZUZH1gAAL///e8xaNAgBUdB1HD8YpGImhOL\nXkREd8BvIOlOFi1a1CLeH507d0avXr3Qtm1btGnTxvjXHEUsS40dOxbh4eEtYr/Zm2eeeQbAbz9d\nJftVXV2NgoICFBQUQK1WIzAwUOmQiBqERS8iak78dERERGSjevTogR49eigdBimMxa7WwdHR0VjY\nJiIiovrh0xuJiIiIiIiIiMjusOhFRERERERERER2h0UvIiIiIiIiIiKyOyx6ERERERERERGR3WHR\ni4iISAHbt2+Hu7s7kpOTlQ7FZrz++utQqVTGvxdeeMGsT1paGqZMmYJNmzahY8eOxr6/+93vzPpG\nR0dDr9fD0dERXbt2RUZGRnMMo0HsbTwAUFlZiZkzZ6Jjx45wdnaGn58fJk2ahLKyMpN+8+fPR1hY\nGFxcXKDT6RAWFoYZM2agsLDQ2Gfbtm2YP38+qqurmyQ2e30fAcDs2bPRpUsXGAwGaDQahIaG4v33\n30dxcbFJv7lz55r8e6v969atm9k6KysrkZiYiNDQUDg7O8PDwwPdunXD2bNnAfD4WKIlHZ8tW7aY\nrLtt27ZWGzcRkdUIERGZSEpKEp4e6U6a6v2RkpIiBoNBtm3b1gRRtXwAJCkpqVHrGDdunHh5eUlq\naqqcOHFCysvLTV6fOXOmxMbGSmFhobEtJCRE2rRpIwAkJSXFbJ2pqakyfPjwRsXVnOxpPH/6059E\nq9XK2rVrpbCwUL777jsxGAzy3HPPmfSLiYmRTz/9VPLy8qSoqEjWr18varVaBg4caNJv0aJFEhkZ\nKdevX29UXPb+PoqMjJRly5ZJfn6+FBYWSlJSkqjVahk8eLBJvzlzm6WTxAAAIABJREFU5ggAs7+u\nXbuarTMuLk7uv/9+OXjwoFRWVsrFixdl2LBhkpWVZezD41M/Len41NTUSG5uruzfv1+GDBkibdq0\naZIxNkU+ICKqL97pRUREpICYmBjcuHEDsbGxSoeCsrIyREREKB1Gvbi4uGDw4MHo3LkzNBqNsf3j\njz/GunXrsH79euj1epNlFi9eDAcHB4wbNw43btxo7pCbnD2M5/Tp0/jiiy/w4osvIj4+Hnq9Hv37\n90dCQgLWrFmD48ePG/s6OzvjzTffhLe3N9zc3DB69GiMGDECu3fvxqVLl4z93nnnHTz00EMYMmQI\nqqqqGhRXa3gfubm5Ydy4cfDy8oJer8eYMWMQFxeHHTt2ICcnx6TvqlWrICImf//9739N+qxbtw5b\ntmzBhg0b0Lt3bzg5OcHX1xdbt241ueuIx6d+WtLxUalU8PPzQ79+/dCpUyfrD56IyApY9CIiImrl\nVqxYgby8PKXDaLD//e9/mDFjBj788ENotVqz1yMiIjB+/HhcuHABkyZNUiDCpmUP4zl8+DBqamrQ\nu3dvk/bBgwcDAHbu3Gls27x5s9lx9fPzAwCzn3zNmjULR44cwaJFiyyOqbW8j1JSUuDo6GjSVvuz\ntdLSUovX9/nnn6NXr17o3r37Pfvy+NybrR4fIqKWikUvIiKiZnbgwAEEBgZCpVJh6dKlAIDly5dD\np9PB1dUVW7duxdNPPw2DwQB/f3+sXbvWuOzixYuh1WrRrl07vP766/D19YVWq0VERAQOHTpk7JeQ\nkABnZ2f4+PgY2958803odDqoVCpcu3YNADB+/HhMnDgRp06dgkqlQmhoKABgx44dMBgMmDdvXnPs\nkkZZvHgxRATDhg27Y5+5c+eic+fO+Prrr5GWlnbX9YkIFi5ciAceeAAajQaenp4YMWIEfv75Z2Of\n+h4vAKiursbMmTMRGBgIFxcXPPjgg0hKSmrUmG19PA4Ov30EdXFxMWmvvZvk1ju96nLy5El4eHgg\nKCjIpN3T0xORkZFYtGgRRMSimFrj+6jWhQsX4OLiguDgYIuWq6iowMGDB9GjR4969efxaRhbOD5E\nRC1Wc/+ekoiopeOcXnQ3TfX+yMnJEQCyZMkSY9u0adMEgOzZs0du3LgheXl50q9fP9HpdFJRUWHs\nN27cONHpdHLs2DEpLy+X7OxsefTRR0Wv18v58+eN/Z5//nlp3769yXYXLFggAOTq1avGtpEjR0pI\nSIhJv5SUFNHr9TJ79uxGj1Wk6eb08vPzM2vv2LGjdOnSpc5lQkJC5MyZMyIi8uOPP4qDg4N06NBB\niouLRaTuuX5mzpwpzs7OsmrVKikoKJCjR49Kr169pG3btnL58mVjv/oer0mTJolGo5GNGzfK9evX\nZerUqeLg4CCHDx+2eB/Yy3iOHj0qAGTGjBkm7VVVVQJA4uLizJapqKiQ3NxcWbJkiWg0Glm1alWd\n654yZYoAkMzMzHrHI9K63ke3KikpEb1eLwkJCSbtc+bMEX9/f/Hw8BC1Wi0dOnSQ4cOHy7/+9S9j\nnzNnzggA6dGjh/Tv3198fHxEo9FIWFiYLF26VGpqasy2x+NjmZZyfN555x3O6UVENol3ehEREbUw\nERERMBgM8Pb2Rnx8PEpKSnD+/HmTPk5OTsY7FLp06YLly5ejqKgIK1eubJIYYmJiUFhYiBkzZjTJ\n+qylpKQEZ86cQUhIyD37hoeH491338XZs2fxwQcf1NmnrKwMCxcuxDPPPIMXXngB7u7u6N69O774\n4gtcu3YNX375pdkydzte5eXlWP7/sXfvUU0f+f/4nxESkgDhoiiUiwp4RazXVrF+1HXrWl1va0Fq\n7RY99aDWIooVtWqVW70V+NiCnlqX3a2t4K1Yq6gfv5Z23aprKyqLqyIKoqggilyVAPP7oz+yTYNK\nIBASno9z8kfnPe+Z12Sm5eTV98w7MRHTpk3D9OnTYW9vj1WrVkEqlTZ7rkx5PL6+vhg/fjwSEhJw\n4sQJPH78GHfv3sX+/fshkUigVqt17nF3d4ebmxvWrl2LjRs3YsaMGQ22Xf+0WGZmZqPjac/rKCYm\nBi4uLoiKitIqf/vtt/HNN98gPz8f5eXl2LVrF27evIlRo0YhKysLwH+3lzo5OSE6OhpZWVm4d+8e\npk6dioULF+Krr77S6Y/zox9TmB8ioraMSS8iIqI2TCaTAUCDSYBfGzJkCJRKpdbWnPagsLAQQggo\nlcpG1Y+KikKvXr2QkJCAkydP6lzPyspCeXk5hgwZolU+dOhQyGQyrS2kDfntfF25cgWVlZVaB0Yr\nFAo4OzsbZK5MeTzJycnw9/fHn//8Zzg6OmLEiBH4+uuvIYRAx44ddern5+ejsLAQX331Ff72t79h\n4MCBDZ5FV78W7t271+hY2us62r9/P3bv3o2jR4/qHAzv7u6OgQMHwsbGBjKZDMOGDUNSUhKqqqqQ\nkJAAAJqXSfj4+MDPzw+Ojo6ws7PDunXrYGdn12DyiPPTeKYyP0REbRmTXkRERGbCysoKRUVFxg6j\nVT1+/BgAtN7k+CxyuRxJSUmQSCSYM2cOqqqqtK6XlJQA+OUNar9lb2+PsrIyveKrqKgAAKxatQoS\niUTzycvLa9Kh1L9lyuOxs7PDtm3bcOvWLVRWViInJwcff/wxAOCFF17QqS+VSuHk5IRx48YhOTkZ\nWVlZiImJ0alXf05Y/dpojPa4jpKTk7F+/Xqkp6ejW7dujbrH19cXFhYWuHr1KgDAxcUFADRnBNaT\nyWTo2rUrcnJydNrg/DSOKc0PEVFbxqQXERGRGVCr1SgpKYGbm5uxQ2lV9T/QamtrG33P8OHDsWTJ\nEmRnZyMyMlLrmr29PQA0+KO3Kd+vk5MTACAuLg5CCK3PqVOn9GrracxpPGfPngUAjBkz5pn1vL29\nYWFhodnG9WvV1dUAdA/Jf5b2to4++eQT7Ny5EydOnGgwwfg0dXV1qKur0ySfbGxs0KNHD1y6dEmn\nbk1NDezs7HTKOT/PZ2rzQ0TUljHpRUREZAbS09MhhMCwYcM0ZZaWls/dFmnqOnfuDIlEgkePHul1\nX2RkJHr37o2MjAyt8n79+sHGxgY//fSTVvmZM2dQXV2NwYMH69WPu7s75HI5zp8/r9d9+jKX8Wzf\nvh3du3fHqFGjAADFxcWYOXOmTr3s7GzU1tbC3d1d51r9WujSpUuj+20v60gIgfDwcGRmZiI1NbXB\nJ53q/eEPf9ApO3v2LIQQGD58uKZsxowZyMjIwPXr1zVllZWVyMvLg6+vr04bnJ+nM9X5ISJqy5j0\nIiIiMkF1dXV4+PAhampqcPHiRYSGhsLDwwNBQUGaOt7e3njw4AFSU1OhVqtRVFSEvLw8nbYcHR1R\nUFCA3NxclJWVQa1WIy0tDSqVCtHR0a04Kv0plUp4enri1q1bet1Xv/3JwsJCpzwsLAz79+/Hzp07\nUVpaiszMTMyfPx8uLi4IDg7Wu5/Zs2dj165dSExMRGlpKWpra3Hr1i3cuXMHABAYGIguXbrg3Llz\nerVt6uN56aWXkJeXh5qaGuTm5mLp0qU4fvw4duzYoTkzydraGseOHcOJEydQWloKtVqNjIwMvP32\n27C2tsaSJUt02q1fC/U/6BsTT3tZR5cuXcLGjRuxfft2SKVSra14EokEmzdv1tS9ffs2kpOTUVJS\nArVajVOnTuGdd96Bh4cH5s+fr6m3ZMkSdO3aFUFBQbh58yaKi4sRHh6OqqqqBg+S5/yY1vwQEZm8\nVntPJBGRiUhJSRH8zyM9jSHWxyeffCKcnZ0FAKFUKsXkyZNFQkKCUCqVAoDo0aOHyMnJEZ999plQ\nqVQCgOjatau4evWqEEKI4OBgIZVKhaurq7C0tBQqlUpMnTpV5OTkaPVTXFwsxowZI+Ryuejevbt4\n7733xPvvvy8ACG9vb3Hz5k0hhBDnzp0TXbt2FQqFQrzyyivi7t274vDhw8LW1lZERUU1a6z1YIBX\n1AcHBwtXV1ed8pCQECGVSkVlZaWmbP/+/cLLy0sAEJ06dRILFy5ssM33339fTJkyRausrq5ObNq0\nSfTo0UNIpVLh4OAgpk2bJq5cuaKpo898PXnyRISHhwsPDw9haWkpnJycxPTp00VWVpYQQohp06YJ\nAGLNmjVPHbu5jUcIIV599VVhb28vLC0thYODg5g4caI4e/asTr3JkyeL7t27CxsbG2FlZSW8vLxE\nYGCgyMzMbLDdiRMnCldXV1FXV6dXPO1hHWVmZgoAT/1s2rRJUzcsLEx4eXkJa2trYWlpKdzc3MTc\nuXNFQUGBTrv5+fnijTfeEA4ODsLKykq89NJLIi0trcEYOD+mNT/1Fi1aJDp27PjU2PVhiL8HRESN\nxV91RES/waQXPUtbWB/BwcHC0dHRqDHoqyWTXtnZ2cLS0lJ88cUXzWrfWGpra8XIkSPFjh07jB2K\nQRhzPPfv3xdyuVxs3rxZ73i4jloe58f05qcek15EZKq4vZGIiMgE6XOgszmpqqrC0aNHkZ2drTlw\n2dvbGxEREYiIiEB5ebmRI9RPbW0tUlNTUVZWhsDAQGOH02zGHs/atWsxYMAAhISE6B0P11HL4/yY\n1vwIIVBQUICTJ0/i2rVrRo6OiKhpmPQiImqmzZs3aw7Z3bZtm7HDeaaIiAj07dsXKpUKVlZW8Pb2\nxrJly5r0A2Lfvn3w9PTUnDXi7OyMWbNmPfe+CxcuIDAwEN27d4eVlRU6deqEF198EVFRUZo6gYGB\nOmeZPO3z7bff6sSyevXqZ8YQGxsLiUSCDh06oHfv3vjhhx/0Hj8Zx4MHDzB+/Hj07NkTc+bM0ZSv\nWLEC/v7+CAwM1Puwa2NKT0/Hvn37kJaWBqVSaexwms2Y44mNjcX58+dx+PBhSKXSJsXDddRyOD+m\nNz8HDhyAq6srRo4ciUOHDhk5QiKiJjL2o2ZERG1NU7avZWdnCwBi69atLRSVYYwaNUokJCSI4uJi\nUVpaKlJSUoRUKhXjx49vcpteXl7Czs6uUXUvXrwolEqlWLRokbhx44aoqqoSV65cEcuWLRNjx47V\n1JsxY4Y4duyYKCkpEWq1Wty5c0cAEJMnTxbV1dWioqJCFBYWirlz54qDBw9qxQJAODs7i+rq6gZj\nqKmpEV27dhUAtPpsLGNvb1yxYoWQyWQCgOjWrZvYs2eP0WLRB1ppO8vRo0dFeHh4i/dDbUtqaqqI\niYkRNTU1BmmP68iwOD9tm6Hn53la6+8BEZEQQlgaLdtGRNSOVVVVYezYsfjxxx9btV8bGxsEBwdr\n3mQVEBCAffv2Yffu3cjPz4e7u3uL9r9582bY29sjPj5eU9azZ09ERkZi+vTpmjKJRIIRI0bo/B9x\niUQCqVQKqVQKpVLZ4GvlBw8ejJ9//hmpqanw9/fXub5v3z64uro2+BZDUxATE4OYmBhjh9FmjRs3\nDuPGjTN2GNTKpkyZgilTphisPa4jw+L8tG2Gnh8ioraE2xuJiIxgx44dKCwsbPV+v/32W51Xt3fq\n1AkAUFlZ2eL9FxcX49GjR3jw4IFWuUwmw8GDBzX/vGvXrkZtAQkODsYf//hHrbIFCxYAALZu3drg\nPbGxsQgLC9M3dCIiIiIiMjFMehERtZDvv/8eL730EpRKJVQqFXx9fVFaWorQ0FCEhYUhJycHEokE\n3t7eiI+Ph7W1NTp06IDBgwejS5cukEqlsLa2xqBBgzBy5Ei4u7tDLpfD3t4ey5YtM1ict2/fhkKh\nQPfu3TVlR44cgUqlQnR0tMH6AYChQ4eioqICv/vd7/DPf/7ToG3X+93vfoc+ffrgu+++w5UrV7Su\n/fOf/0RlZSWfECAiIiIiageY9CIiagEVFRWYPHkyXn/9dTx48ADZ2dno2bMnqqurER8fj0mTJsHL\nywtCCFy7dg2hoaF4//33IYTA1q1bcePGDdy9exf/8z//g4yMDKxYsQIZGRl48OAB3n77bWzatAkX\nLlxodpyVlZU4ceIE5s6dC5lMpimvfzNgXV1ds/v4tWXLlmHIkCG4cOECXnnlFfj4+GDjxo06T341\n17x58wBA58UCH3/8MZYsWWLQvoiIiIiIqG1i0ouIqAXk5uaitLQUPj4+kMvl6NKlC/bt26fZSvgs\nffv2hVKpRMeOHfHGG28AADw8PNCpUycolUrNGxIvX77c7DhjYmLg4uKi9eZEAJg4cSJKS0uf+xZE\nfSkUCvz444/43//9X/Tu3RuXLl1CeHg4+vTpg++//95g/bz99tuwtrbG3/72N1RVVQEArl+/jrNn\nz2LmzJkG64eIiIiIiNouHmRPRNQCPD090blzZ8yaNQuLFi1CUFAQunXrpnc79U9f1dTUaMrqXyWu\nVqubFeP+/fuxe/duHDt2DLa2ts1qSx9SqRQhISEICQnBmTNnsH79es2h81euXIGDg0Oz+7Czs8PM\nmTOxfft2JCcnY/bs2YiLi8OCBQsgk8lQXV3d7D52797d7Dbam1OnThk7BCIiIiJqR5j0IiJqAQqF\nAidOnMDy5csRHR2NiIgIBAQEICkpCQqFwtjhITk5GbGxsUhPT8cLL7xgtDhefvllfP3111iwYAG2\nbt2K7777Dn/6058M0vaCBQuwfft2bNu2DdOmTcOePXvwn//8xyBtA8CMGTMM1lZ7ER8fr/XmTiIi\nIiKilsSkFxFRC/Hx8cHBgwdRVFSE2NhYrF+/Hj4+PgbfMqivTz75BEePHsWJEydgY2PTon398MMP\n+Pnnn7F48WIAwPTp05GSkgJLS+0/P2+99Ra2bt1q0DdIDhgwAMOGDcPp06cRHBwMf39/gzxFVk8I\nYbC22gOJRIKUlBQEBAQYOxQiIjIiiURi7BCIqB3hmV5ERC2goKAAly5dAgA4OTnho48+wqBBgzRl\nxiCEQHh4ODIzM5GamtriCS8A+Pnnn2Ftba355ydPnjT4HdS/ZbF///4G7X/BggUAgL1792oSb0RE\nRERE1D4w6UVE1AIKCgowb948XL58GdXV1cjIyEBeXh6GDRsGAHB0dERBQQFyc3NRVlbW7PO5GuPS\npUvYuHEjtm/fDqlUColEovXZvHmzpm5aWhpUKhWio6Ob1Jdarca9e/eQnp6ulfQCgGnTpmH37t0o\nKSnBo0ePcODAASxfvhxTpkwxeNIrICAAnTp1wrRp0+Dp6WnQtomIiIiIqG1j0ouIqJliY2Pxyiuv\nAACWLl2K6dOnw8nJCbW1tfDz84NSqcQf//hHzJs3DwsXLgQAzJ8/H507d0bfvn0xYcIEfPjhh9i0\naRMAwNfXFydPnsSGDRswb948AMD48ePx1VdfISUlBePHjwcAhISEIDk5udFxGno73tdffw1vb2/k\n5OTg0aNHWgk0mUwGZ2dnfPPNN1AqlZp7Fi1ahKFDh2LlypVwdnZG586dER4ejvnz5yMlJUWnj7Ky\nMowaNQo+Pj4AgIMHD6JHjx6IiYl5aixDhw7Fe++9BwCwsrLCnDlzEBYWpqm7evVq9OjRAwDw3Xff\nwcfHBydPnjTod0NERERERMYnETyUhIhIy+7duzFjxgye2UQN4vpoGp7pRUREAP8eEFHr4pNeRERE\nRERERERkdpj0IiIyYZcvX9Y5m6uhT2BgoLFDJSIiIiIialVMehERmbDevXtDCPHcjz5nfxG1BfPm\nzdNK3M6aNUunzvHjx7FixQrs27cPnp6emrpvvfWWTt1x48bB1tYWFhYW8PHxwblz51pjGE1ibuMB\ngIiICPTt2xcqlQpWVlbw9vbGsmXLUF5erlUvKiqqwcR9v379dNpUq9WIiYmBt7c3ZDIZ7O3t0a9f\nP+Tm5gIAvvnmG2zYsAG1tbUGGYO5rjeA88P5afz8pKamarXdqVOnFhs3EZFBCCIi0pKSkiL4n0d6\nGq6PpgEgUlJSGl0/ODhYODo6irS0NHHlyhXx+PFjretr1qwRkyZNEqWlpZoyLy8v0bFjRwFAfPvt\ntzptpqWliSlTpjR9EK3MnMYzatQokZCQIIqLi0VpaalISUkRUqlUjB8/XqteZGSkAKDz8fHx0Wlz\n2rRpolevXuL06dNCrVaLgoICMXnyZJGZmampEx8fL0aNGiUePnzYrPjNfb1xftq2tjQ/dXV14tat\nW+KHH34QEyZMEB07dtR7PPr+PSAiag4+6UVERGRiqqqq4OfnZ/J9PI9CocD48ePRs2dPWFlZacrX\nr1+P5ORk7N69G7a2tlr3bNmyBR06dEBwcDAePXrU2iEbnLmMx8bGBsHBwXB0dIStrS0CAgIwbdo0\nHDlyBPn5+Vp1v/jiC52nVf/9739r1UlOTkZqair27NmDl19+GZaWlnBxccGBAwe0nmpZtGgRXnzx\nRUyYMAE1NTVNir09rDfOT9vWluZHIpHA1dUVI0eO1LwJmYioLWPSi4iIyMTs2LEDhYWFJt9HU1y7\ndg2rV6/GunXrIJfLda77+fkhNDQUt2/fxtKlS40QoWGZy3i+/fZbWFhYaJXVb4uqrKzUu72tW7di\n0KBB8PX1fW7dtWvX4vz584iPj9e7n/ay3jg/bZupzg8RUVvApBcREVELE0IgNjYWffr0gZWVFRwc\nHDB16lRcvnxZUyckJAQymQzOzs6asnfffRfW1taQSCS4f/8+ACA0NBRhYWHIycmBRCKBt7c3tmzZ\nArlcjs6dO2PevHlwcXGBXC6Hn58fzpw5Y5A+AODIkSNQqVSIjo5u0e/rWbZs2QIhBCZPnvzUOlFR\nUejZsyc+//xzHD9+/JntNWZuEhMTYW1tDaVSiQMHDuC1116DSqWCm5sbdu3apdVebW0t1qxZAw8P\nDygUCvTv3x8pKSnNGrO5jafe7du3oVAo0L17d73uq66uxunTpzFgwIBG1XdwcMCoUaMQHx8PIYRe\nfbXH9VaP88P5AZo3P0REbUIrbqUkIjIJPLOJnqUp62PNmjVCJpOJL774QpSUlIiLFy+KQYMGiU6d\nOom7d+9q6r355puiS5cuWvdu2rRJABBFRUWasunTpwsvLy+tesHBwcLa2lpcunRJPH78WGRlZYmh\nQ4cKW1tbcfPmTYP08e233wpbW1sRERGh1/iFaNqZXq6urjrlnp6eom/fvg3e4+XlJW7cuCGEEOLH\nH38UHTp0EN26dRPl5eVCiIbP8Gns3HzwwQcCgPh//+//iUePHonCwkIxcuRIYW1tLaqrqzX1li5d\nKqysrMTevXvFw4cPxcqVK0WHDh3E2bNnGz12cx3Pr1VUVAhbW1sREhKiVR4ZGSnc3NyEvb29kEql\nolu3bmLKlCniX//6l6bOjRs3BAAxYMAAMXr0aOHs7CysrKxE7969xaeffirq6up0+luxYoUAIDIy\nMvSKsz2tt1/j/HB+fu1p87No0SKe6UVEbR5/1RER/QaTXvQs+q6PyspKYWNjIwIDA7XK//WvfwkA\nWgmk5ia97OzstMrOnj0rAIh169YZpI/mMETSq7y8XEgkEjFp0qQG7/n1j1whhAgLCxMAxMKFC4UQ\nuj9y9Zmb+h+5VVVVmrKEhAQBQFy7dk0IIURVVZVQKpVa7VVWVgorKyuxYMGCRo/dXMfzax988IHo\n2bOn1sHjQghx8+ZNce7cOVFWViaePHkiTp06JQYOHCgUCoX497//LYQQIjMzUwAQr776qvjnP/8p\niouLRUlJiVi+fLkAIHbu3KnT31/+8hcBQPz9739vdIztbb39Gufnvzg/T58fJr2IyBRweyMREVEL\nysrKQnl5OYYMGaJVPnToUMhkMq3th4Y2ZMgQKJVKra05pqywsBBCCCiVykbVj4qKQq9evZCQkICT\nJ0/qXG/u3MhkMgCAWq0GAFy5cgWVlZVaB0ErFAo4OzsbZA7MZTz79+/H7t27cfToUZ2Dx93d3TFw\n4EDY2NhAJpNh2LBhSEpKQlVVFRISEgBA81IDHx8f+Pn5wdHREXZ2dli3bh3s7Ozw2Wef6fRZv2bu\n3bvX6Djb63rj/HB+fqsp80NE1FYw6UVERNSCSkpKAPzy9q3fsre3R1lZWYv2b2VlhaKiohbto7U8\nfvwYALTe5PgscrkcSUlJkEgkmDNnDqqqqrSuG3puKioqAACrVq2CRCLRfPLy8pp02PRvmcN4kpOT\nsX79eqSnp6Nbt26NusfX1xcWFha4evUqAMDFxQUANGfQ1ZPJZOjatStycnJ02lAoFAD+u4Yaoz2u\nN84P58dQ80NE1FYw6UVERNSC7O3tAaDBH0wlJSVwc3Nrsb7VanWL99Ga6n941dbWNvqe4cOHY8mS\nJcjOzkZkZKTWNUPPjZOTEwAgLi4O4pcjJDSfU6dO6dXW05jyeD755BPs3LkTJ06cwAsvvNDo++rq\n6lBXV6dJbtjY2KBHjx64dOmSTt2amhrY2dnplFdXVwP47xpqjPa23jg/nB9Dzg8RUVvBpBcREVEL\n6tevH2xsbPDTTz9plZ85cwbV1dUYPHiwpszS0lKzNcYQ0tPTIYTAsGHDWqyP1tS5c2dIJBI8evRI\nr/siIyPRu3dvZGRkaJXrMzeN4e7uDrlcjvPnz+t1n75MbTxCCISHhyMzMxOpqakNPklT7w9/+INO\n2dmzZyGEwPDhwzVlM2bMQEZGBq5fv64pq6ysRF5eHnx9fXXaqF8zXbp0aXTc7WW9cX5+wfkx7PwQ\nEbUVTHoRERG1ILlcjrCwMOzfvx87d+5EaWkpMjMzMX/+fLi4uCA4OFhT19vbGw8ePEBqairUajWK\nioqQl5en06ajoyMKCgqQm5uLsrIyTRKrrq4ODx8+RE1NDS5evIjQ0FB4eHggKCjIIH2kpaVBpVIh\nOjra8F9UIyiVSnh6euLWrVt63Ve/rcnCwkKnvLFz09h+Zs+ejV27diExMRGlpaWora3FrVu3cOfO\nHQBAYGAgunTpgnPnzunVtimP59KlS9i4cSO2b98OqVSqtdVLIpFg8+bNmrq3b99GcnIySkpKoFar\ncerUKbzzzjvw8PDA/PnzNfWWLFmCrl27IigoCDdv3kRxcTHTpOaBAAAgAElEQVTCw8NRVVWF5cuX\n68RQv2bqf9A3Ju72st44P/8t5/w0bn6IiExKa52YT0RkKvj2RnqWpqyPuro6sWnTJtGjRw8hlUqF\ng4ODmDZtmrhy5YpWveLiYjFmzBghl8tF9+7dxXvvvSfef/99AUB4e3uLmzdvCiGEOHfunOjatatQ\nKBTilVdeEXfv3hXBwcFCKpUKV1dXYWlpKVQqlZg6darIyckxWB+HDx8Wtra2IioqSu/vDQZ4e6MQ\nQoSEhAipVCoqKys1Zfv37xdeXl4CgOjUqZPm7Wy/9f7772u9rU2Ixs1NQkKCUCqVAoDo0aOHyMnJ\nEZ999plQqVQCgOjatau4evWqEEKIJ0+eiPDwcOHh4SEsLS2Fk5OTmD59usjKyhJCCDFt2jQBQKxZ\ns+apYze38dS/Le5pn02bNmnqhoWFCS8vL2FtbS0sLS2Fm5ubmDt3rigoKNBpNz8/X7zxxhvCwcFB\nWFlZiZdeekmkpaU1GMPEiROFq6urqKura3TcQrSP9cb5+S/OT+Pmpx7f3khEpoC/6oiIfoNJL3qW\ntro+goODhaOjo7HDeCpDJb2ys7OFpaWl+OKLLwwZXqupra0VI0eOFDt27DB2KAZhCuO5f/++kMvl\nYvPmzZqyxsbN9dbyOD+mNz/1mPQiIlPA7Y1ERERmQp8DnU1BVVUVjh49iuzsbM1Byt7e3oiIiEBE\nRATKy8uNHKF+amtrkZqairKyMgQGBho7nGYzlfGsXbsWAwYMQEhICAD94uZ6a3mcH9OaHyEECgoK\ncPLkSVy7ds3I0RERPR+TXkRERNQmPXjwAOPHj0fPnj0xZ84cTfmKFSvg7++PwMBAvQ+xNqb09HTs\n27cPaWlpUCqVxg6n2UxhPLGxsTh//jwOHz4MqVQKQP+4ud5aDufH9ObnwIEDcHV1xciRI3Ho0CEj\nR0hE9HwSIYQwdhBERG3J7t27MWPGDPA/j9SQtrg+Vq5ciY8//hjV1dXo1q0bNm3ahNdff93YYWmR\nSCRISUlBQECAwdo8duwYTpw4gfXr1xusTTIfBw4cwKVLl7Bs2TKdQ8ubguvNsDg/bZuh5+fXWuLv\nARHR0zDpRUT0G20xqUFtB9dH0/BHDhERAfx7QESti9sbiYiIiIiIiIjI7DDpRUREREREREREZodJ\nLyIiIiIiIiIiMjtMehERERERERERkdmxNHYARERtlb+/v7FDoDbo1q1bALg+miIuLg579uwxdhhE\nRERE1E7w7Y1ERL9x6tQpxMbGGjsMIqJGycjIAAAMHDjQyJEQETXOkiVLMHz4cGOHQUTtAJNeRERE\nRCYsICAAALB7924jR0JERETUtvBMLyIiIiIiIiIiMjtMehERERERERERkdlh0ouIiIiIiIiIiMwO\nk15ERERERERERGR2mPQiIiIiIiIiIiKzw6QXERERERERERGZHSa9iIiIiIiIiIjI7DDpRURERERE\nREREZodJLyIiIiIiIiIiMjtMehERERERERERkdlh0ouIiIiIiIiIiMwOk15ERERERERERGR2mPQi\nIiIiIiIiIiKzw6QXERERERERERGZHSa9iIiIiIiIiIjI7DDpRUREREREREREZodJLyIiIiIiIiIi\nMjtMehERERERERERkdlh0ouIiIiIiIiIiMwOk15ERERERERERGR2mPQiIiIiIiIiIiKzw6QXERER\nERERERGZHSa9iIiIiIiIiIjI7DDpRUREREREREREZodJLyIiIiIiIiIiMjtMehERERERERERkdlh\n0ouIiIiIiIiIiMwOk15ERERERERERGR2mPQiIiIiIiIiIiKzw6QXERERERERERGZHSa9iIiIiIiI\niIjI7DDpRUREREREREREZsfS2AEQERERUeNUVlbiyZMnWmXV1dUAgIcPH2qVW1lZQalUtlpsRERE\nRG2NRAghjB0EERERET1fYmIi3n333UbVTUhIwIIFC1o4IiIiIqK2i0kvIiIiIhNRVFQEFxcX1NbW\nPrOehYUF7ty5Aycnp1aKjIiIiKjt4ZleRERERCbCyckJY8eOhYWFxVPrWFhY4Pe//z0TXkRERNTu\nMelFREREZEJmzZqFZz2oL4TArFmzWjEiIiIioraJ2xuJiIiITEhZWRmcnJx0DrSvJ5PJUFRUBJVK\n1cqREREREbUtfNKLiIiIyITY2tpi0qRJkEqlOtcsLS0xZcoUJryIiIiIwKQXERERkcl58803UVNT\no1NeW1uLN9980wgREREREbU93N5IREREZGKqq6vRqVMnlJWVaZXb2Njg/v37sLKyMlJkRERERG0H\nn/QiIiIiMjEymQz+/v6QyWSaMqlUihkzZjDhRURERPT/Y9KLiIiIyATNnDkT1dXVmn9Wq9WYOXOm\nESMiIiIialu4vZGIiIjIBNXV1cHZ2RlFRUUAgE6dOuHu3buwsLAwcmREREREbQOf9CIiIiIyQR06\ndMDMmTMhk8kglUrx5ptvMuFFRERE9CtMehERERGZqDfeeAPV1dXc2khERETUAEtjB0BERNRUp06d\nQn5+vrHDIDIaIQQ6duwIALhx4wZyc3ONGxCREbm7u2P48OHGDoOIiNoQnulFREQmy9/fH3v37jV2\nGERE1Aa8/vrr2LNnj7HDICKiNoRPehERkUnjjxxqColEgpSUFAQEBBg7lGa7dOkSAKBv374t2o+/\nvz8A8N83apPq1ycREdGvMelFREREZMJaOtlFREREZKp4kD0REREREREREZkdJr2IiIiIiIiIiMjs\nMOlFRERERERERERmh0kvIiIiIiIiIiIyO0x6ERERERERERGR2WHSi4iIiKiJDh8+DDs7Oxw8eNDY\nobR5x48fx4oVK7Bv3z54enpCIpFAIpHgrbfe0qk7btw42NrawsLCAj4+Pjh37pwRIm4ccxsPAERE\nRKBv375QqVSwsrKCt7c3li1bhvLycq16UVFRmnH/+tOvXz+dNtVqNWJiYuDt7Q2ZTAZ7e3v069cP\nubm5AIBvvvkGGzZsQG1tbWsMkYiI2gkmvYiIiIiaSAhh7BBMwocffogtW7Zg5cqVmD59Oq5fvw4v\nLy907NgRO3fuxKFDh7TqHzt2DHv27MGkSZOQlZWFQYMGGSny5zO38QDAiRMnsHDhQuTm5uL+/fuI\niYlBfHw8/P39m9zmjBkz8Pe//x1ffvklKisr8Z///AdeXl6aRNrkyZMhl8sxduxYlJSUGGooRETU\nzjHpRURERNREEydOxKNHjzBp0iRjh4Kqqir4+fkZOwwd69evR3JyMnbv3g1bW1uta1u2bEGHDh0Q\nHByMR48eGSlCwzGX8djY2CA4OBiOjo6wtbVFQEAApk2bhiNHjiA/P1+r7hdffAEhhNbn3//+t1ad\n5ORkpKamYs+ePXj55ZdhaWkJFxcXHDhwQOupsEWLFuHFF1/EhAkTUFNT0ypjJSIi88akFxEREZEZ\n2LFjBwoLC40dhpZr165h9erVWLduHeRyuc51Pz8/hIaG4vbt21i6dKkRIjQscxnPt99+CwsLC62y\nTp06AQAqKyv1bm/r1q0YNGgQfH19n1t37dq1OH/+POLj4/Xuh4iI6LeY9CIiIiJqgpMnT8LDwwMS\niQSffvopACAxMRHW1tZQKpU4cOAAXnvtNahUKri5uWHXrl2ae7ds2QK5XI7OnTtj3rx5cHFxgVwu\nh5+fH86cOaOpFxISAplMBmdnZ03Zu+++C2tra0gkEty/fx8AEBoairCwMOTk5EAikcDb2xsAcOTI\nEahUKkRHR7fGV6Jjy5YtEEJg8uTJT60TFRWFnj174vPPP8fx48ef2Z4QArGxsejTpw+srKzg4OCA\nqVOn4vLly5o6jZ0DAKitrcWaNWvg4eEBhUKB/v37IyUlpVljNrfx1Lt9+zYUCgW6d++u133V1dU4\nffo0BgwY0Kj6Dg4OGDVqFOLj47l9mIiImo1JLyIiIqImeOWVV/Djjz9qlS1YsACLFy9GVVUVbG1t\nkZKSgpycHHh6emLu3LlQq9UAfklmBQUFobKyEosWLUJubi7OnTuHmpoavPrqq5otZFu2bEFAQIBW\nHwkJCVi3bp1WWXx8PCZNmgQvLy8IIXDt2jUA0BwKXldX1yLfwfMcOnQIvXr1glKpfGodhUKBv/71\nr+jQoQPmzp2LioqKp9Zdu3YtVqxYgQ8++ACFhYX44YcfkJ+fj5EjR+LevXsAGj8HALB8+XJs3LgR\ncXFxuHPnDiZNmoSZM2fip59+avKYzW08wC9Pd504cQJz586FTCbTurZixQo4ODhAJpOhe/fumDp1\nKs6ePau5XlBQgOrqavz8888YM2aMJsHbp08fJCQkNJjYGjhwIG7fvo0LFy40K24iIiImvYiIiIha\ngJ+fH1QqFZycnBAYGIiKigrcvHlTq46lpaXmKZ++ffsiMTERZWVlSEpKMkgMEydORGlpKVavXm2Q\n9vRRUVGBGzduwMvL67l1hw8fjsWLFyM3NxfLly9vsE5VVRViY2Pxpz/9CbNmzYKdnR18fX2xbds2\n3L9/H5999pnOPc+ag8ePHyMxMRHTpk3D9OnTYW9vj1WrVkEqlTb7+ze38cTExMDFxQVRUVFa5W+/\n/Ta++eYb5Ofno7y8HLt27cLNmzcxatQoZGVlAYDmoHonJydER0cjKysL9+7dw9SpU7Fw4UJ89dVX\nOv316NEDAJCZmdmsuImIiJj0IiIiImph9U/H/PqpnIYMGTIESqVSa3ubqSosLIQQ4plPef1aVFQU\nevXqhYSEBJw8eVLnelZWFsrLyzFkyBCt8qFDh0Imk2ltC23Ib+fgypUrqKys1DpIXaFQwNnZ2SDf\nv7mMZ//+/di9ezeOHj2q8yICd3d3DBw4EDY2NpDJZBg2bBiSkpJQVVWFhIQEAICVlRUAwMfHB35+\nfnB0dISdnR3WrVsHOzu7BpN79Wum/mk3IiKipmLSi4iIiKgNsbKyQlFRkbHDaLbHjx8D+G/S43nk\ncjmSkpIgkUgwZ84cVFVVaV0vKSkB8MubBX/L3t4eZWVlesVXv+1w1apVkEgkmk9eXl6TDmv/LXMY\nT3JyMtavX4/09HR069atUff4+vrCwsICV69eBQC4uLgAgOb8uXoymQxdu3ZFTk6OThsKhQLAf9cQ\nERFRUzHpRURERNRGqNVqlJSUwM3NzdihNFt94qL+XLHGGD58OJYsWYLs7GxERkZqXbO3tweABpNB\nTfnOnJycAABxcXEQQmh9Tp06pVdbT2PK4/nkk0+wc+dOnDhxAi+88EKj76urq0NdXZ0m2WljY4Me\nPXrg0qVLOnVrampgZ2enU15dXQ3gv2uIiIioqZj0IiIiImoj0tPTIYTAsGHDNGWWlpbP3RbZFnXu\n3BkSiQSPHj3S677IyEj07t0bGRkZWuX9+vWDjY2NzqHsZ86cQXV1NQYPHqxXP+7u7pDL5Th//rxe\n9+nL1MYjhEB4eDgyMzORmpra4JNo9f7whz/olJ09exZCCAwfPlxTNmPGDGRkZOD69euassrKSuTl\n5cHX11enjfo106VLl+YMhYiIiEkvIiIiImOpq6vDw4cPUVNTg4sXLyI0NBQeHh4ICgrS1PH29saD\nBw+QmpoKtVqNoqIi5OXl6bTl6OiIgoIC5ObmoqysDGq1GmlpaVCpVIiOjm7FUf1CqVTC09MTt27d\n0uu++m2BFhYWOuVhYWHYv38/du7cidLSUmRmZmL+/PlwcXFBcHCw3v3Mnj0bu3btQmJiIkpLS1Fb\nW4tbt27hzp07AIDAwEB06dIF586d06ttUx7PpUuXsHHjRmzfvh1SqVRrq6REIsHmzZs1dW/fvo3k\n5GSUlJRArVbj1KlTeOedd+Dh4YH58+dr6i1ZsgRdu3ZFUFAQbt68ieLiYoSHh6OqqqrBg/7r10xD\nCTEiIiJ9MOlFRERE1ASffvophg4dCgAIDw/HlClTkJiYiLi4OABA//79cf36dWzfvh1hYWEAgPHj\nxyM7O1vTxuPHj+Hr6wuFQoGRI0eiZ8+e+O6777TOwVqwYAHGjBmDN954A7169UJkZKRm29fw4cOR\nn58PAJg/fz46d+6Mvn37YsKECXjw4EGrfA/PMnHiRGRlZWmdZ/X111/D29sbOTk5GDp0KN577z2d\n+4YNG4YlS5bolH/44YeIiYlBREQEOnXqhFGjRqFbt25IT0+HtbU1AOg1B/Hx8Vi8eDE2bNiAjh07\nwsXFBaGhoXj48CGAX7bZFRYW4sCBA08do7mNRwjx1Gu/NX78eKxatQpubm5QKpUICAjAiBEjcPr0\naXTs2FFTz8HBAf/4xz/g5uaGAQMGwNXVFf/6179w6NAhDBgwQKfds2fPwtXVFf379290LERERA2R\nCH3+shEREbUh/v7+AIA9e/YYORIyNRKJBCkpKQgICDBaDPPmzcOePXtQXFxstBj00ZR/365du4Y+\nffogKSkJs2bNaqnQWkxdXR1Gjx6NoKAgzJkzx9jhNJspjKe4uBhubm6IiorSJPYag38PiIioIXzS\ni4iIiMhI9Dnk3RR5e3sjIiICERERKC8vN3Y4eqmtrUVqairKysoQGBho7HCazVTGs3btWgwYMAAh\nISHGDoWIiMwAk15ERNSuvfPOO7C1tYVEImnxA61bWl1dHeLi4uDn59fkNvbt2wdPT0+dc3xkMhk6\nd+6M0aNHY9OmTZrtUkTPs2LFCvj7+yMwMFDvQ+2NKT09Hfv27UNaWhqUSqWxw2k2UxhPbGwszp8/\nj8OHD0MqlRo7HCIiMgNMehERUbv2+eefY/v27cYOo9mys7PxP//zP1iyZAkqKyub3M706dNx/fp1\neHl5wc7ODkII1NXVobCwELt370b37t0RHh4OHx8fnbfOUeOtXLkSSUlJePToEbp37469e/caO6QW\nFR0djZCQEHz00UfGDqXRxo4diy+//BLOzs7GDsUg2vp4Dhw4gCdPniA9PR0ODg7GDoeIiMyEpbED\nICIioua5cOECIiIiMH/+fFRUVOh1EHVjSCQS2NvbY/To0Rg9ejQmTpyIGTNmYOLEibh69Srs7OwM\n2l97EBMTg5iYGGOH0arGjRuHcePGGTsMaqOmTJmCKVOmGDsMIiIyM3zSi4iI2j2JRGLsEJrlxRdf\nxL59+/Dmm29qvfWvpbz++usICgpCYWEhtm3b1uL9ERERERE1BZNeRETUrgghsGnTJvTq1QtWVlaw\ns7PD+++/r1OvtrYWa9asgYeHBxQKBfr374+UlBQAQGJiIqytraFUKnHgwAG89tprUKlUcHNzw65d\nu7Ta+f777/HSSy9BqVRCpVLB19cXpaWlz+2jJRw5cgQqlQrR0dHNbisoKAgAkJaWpikzx++MiIiI\niEwXk15ERNSurF69GuHh4QgODsa9e/dw9+5dLF++XKfe8uXLsXHjRsTFxeHOnTuYNGkSZs6ciZ9+\n+gkLFizA4sWLUVVVBVtbW6SkpCAnJweenp6YO3cu1Go1AKCiogKTJ0/G66+/jgcPHiA7Oxs9e/ZE\ndXX1c/toCfVvCqyrq2t2WwMGDAAAXL9+XVNmjt8ZEREREZkuJr2IiKjdqKqqQlxcHH7/+99jyZIl\nsLe3h0KhgKOjo1a9x48fIzExEdOmTcP06dNhb2+PVatWQSqVIikpSauun58fVCoVnJycEBgYiIqK\nCty8eRMAkJubi9LSUvj4+EAul6NLly7Yt28fOnXqpFcfhjJx4kSUlpZi9erVzW6r/o2XZWVlAMz3\nOyMiIiIi08WD7ImIqN24du0aKisrMXbs2GfWu3LlCiorK9GvXz9NmUKhgLOzMy5fvvzU+2QyGQBo\nnlry9PRE586dMWvWLCxatAhBQUHo1q1bs/poK+oPzFepVABM8zuLi4vDnj179L6vvTp9+jQAwN/f\n38iREOk6ffo0hg0bZuwwiIiojeGTXkRE1G7cunULAODk5PTMehUVFQCAVatWQSKRaD55eXmorKxs\ndH8KhQInTpzAK6+8gujoaHh6eiIwMBBVVVUG68NYrl69CgDo3bs3AH5nRERERNT28EkvIiJqN+Ry\nOQDgyZMnz6xXnxSLi4tDaGhos/r08fHBwYMHUVRUhNjYWKxfvx4+Pj4IDAw0WB/GcOTIEQDAa6+9\nBsA0v7PFixcjICCg2e20F/VPePHpOGqL+AQiERE1hE96ERFRu9GvXz906NAB33///TPrubu7Qy6X\n4/z5883qr6CgAJcuXQLwS1Loo48+wqBBg3Dp0iWD9WEMd+/eRVxcHNzc3DBnzhwA/M6IiIiIqO1h\n0ouIiNoNJycnTJ8+HXv37sWOHTtQWlqKixcv4rPPPtOqJ5fLMXv2bOzatQuJiYkoLS1FbW0tbt26\nhTt37jS6v4KCAsybNw+XL19GdXU1MjIykJeXh2HDhhmsD32kpaVBpVIhOjq6UfWFECgvL0ddXR2E\nECgqKkJKSgpGjBgBCwsLpKamas70MtfvjIiIiIhMF5NeRETUrvzlL3/B7NmzER4eDldXV7z77rsY\nOXIkAGDSpEm4ePEiACA+Ph6LFy/Ghg0b0LFjR7i4uCA0NBQPHz5EYmIi4uLiAAD9+/fH9evXsX37\ndoSFhQEAxo8fj+zsbDg5OaG2thZ+fn5QKpX44x//iHnz5mHhwoXP7UMfp0+fxiuvvIIXXngBZ86c\nwYULF+Di4oIRI0bghx9+0KutgwcP4sUXX8SdO3fw+PFj2NnZwcLCAhYWFujZsydiY2MRFBSErKws\nDB48WOteU/rOiIiIiMj8SYQQwthBEBERNQXPGKKmkkgkSElJ4ZleeuC/b9SWcX0SEVFD+KQXERER\nERERERGZHSa9iIiI2pjLly9DIpE891P/NkMiU3T8+HGsWLEC+/btg6enp2Zdv/XWWzp1x40bB1tb\nW1hYWMDHxwfnzp0zQsSNY27jAQC1Wo01a9bA09MTMpkMrq6uWLp0KaqqqrTqbdiwAb1794ZCoYC1\ntTV69+6N1atXo7S0VFPnm2++wYYNG1BbW9vawyAionaISS8iIqI2pnfv3hBCPPeTnJxs7FCJmuTD\nDz/Eli1bsHLlSkyfPh3Xr1+Hl5cXOnbsiJ07d+LQoUNa9Y8dO4Y9e/Zg0qRJyMrKwqBBg4wU+fOZ\n23gAIDQ0FJs2bUJMTAyKi4vx5ZdfYvv27XjnnXe06v3jH//A3LlzcfPmTdy7dw+RkZHYsGEDXn/9\ndU2dyZMnQy6XY+zYsSgpKWntoRARUTvDpBcRERFRK6uqqoKfn5/J99EU69evR3JyMnbv3g1bW1ut\na1u2bEGHDh0QHByMR48eGSlCwzGH8Vy/fh3btm3Dn//8ZwQGBsLW1hajR49GSEgIvvrqK/znP//R\n1JXJZHj33Xfh5OQEGxsb+Pv7Y+rUqfi///s/rTesLlq0CC+++CImTJiAmpoaYwyLiIjaCSa9iIiI\niFrZjh07UFhYaPJ96OvatWtYvXo11q1bB7lcrnPdz88PoaGhuH37NpYuXWqECA3LHMZz9uxZ1NXV\n4eWXX9YqHz9+PADg6NGjmrL9+/frzKurqysAoLy8XKt87dq1OH/+POLj41sibCIiIgBMehERERE9\nlxACsbGx6NOnD6ysrODg4ICpU6fi8uXLmjohISGQyWRwdnbWlL377ruwtraGRCLB/fv3AfyyVSws\nLAw5OTmQSCTw9vbGli1bIJfL0blzZ8ybNw8uLi6Qy+Xw8/PDmTNnDNIHABw5cgQqlQrR0dEt+n09\nzZYtWyCEwOTJk59aJyoqCj179sTnn3+O48ePP7O9xsxLYmIirK2toVQqceDAAbz22mtQqVRwc3PD\nrl27tNqrra3FmjVr4OHhAYVCgf79+yMlJaVZYzb18XTo8MvPBYVCoVXeo0cPANB60qsh2dnZsLe3\nR9euXbXKHRwcMGrUKMTHx4MvkyciopbCpBcRERHRc6xduxYrVqzABx98gMLCQvzwww/Iz8/HyJEj\nce/ePQC/JHQCAgK07ktISMC6deu0yuLj4zFp0iR4eXlBCIFr164hJCQEQUFBqKysxKJFi5Cbm4tz\n586hpqYGr776KvLz85vdBwDN4eF1dXWG+3L0cOjQIfTq1QtKpfKpdRQKBf7617+iQ4cOmDt3Lioq\nKp5atzHzsmDBAixevBhVVVWwtbVFSkoKcnJy4Onpiblz50KtVmvaW758OTZu3Ii4uDjcuXMHkyZN\nwsyZM/HTTz81ecymPp7evXsD0E1udezYEQBQVFSkc49arcbt27fx6aef4vjx4/jkk08gk8l06g0c\nOBC3b9/GhQsXGh0PERGRPpj0IiIiInqGqqoqxMbG4k9/+hNmzZoFOzs7+Pr6Ytu2bbh//z4+++wz\ng/VlaWmpecqnb9++SExMRFlZGZKSkgzS/sSJE1FaWorVq1cbpD19VFRU4MaNG/Dy8npu3eHDh2Px\n4sXIzc3F8uXLG6zTlHnx8/ODSqWCk5MTAgMDUVFRgZs3bwIAHj9+jMTEREybNg3Tp0+Hvb09Vq1a\nBalU2uzv35TH4+vri/HjxyMhIQEnTpzA48ePcffuXezfvx8SiUQryVbP3d0dbm5uWLt2LTZu3IgZ\nM2Y02Hb902KZmZmNjoeIiEgfTHoRERERPUNWVhbKy8sxZMgQrfKhQ4dCJpNpbT80tCFDhkCpVGpt\nbzNVhYWFEEI88ymvX4uKikKvXr2QkJCAkydP6lxv7rzUP3lUn7S5cuUKKisr0a9fP00dhUIBZ2dn\ng3z/pjye5ORk+Pv7489//jMcHR0xYsQIfP311xBCaJ74+rX8/HwUFhbiq6++wt/+9jcMHDiwwfPl\n6tdC/VNsREREhsakFxEREdEzlJSUAABsbGx0rtnb26OsrKxF+7eysmpwC5mpefz4MYBfxtMYcrkc\nSUlJkEgkmDNnDqqqqrSuG3pe6rcdrlq1ChKJRPPJy8tDZWWlXm01xJTHY2dnh23btuHWrVuorKxE\nTk4OPv74YwDACy+8oFNfKpXCyckJ48aNQ3JyMrKyshCFBPAAAAWESURBVBATE6NTr/6csPq1QURE\nZGhMehERERE9g729PQA0mHQoKSmBm5tbi/WtVqtbvI/WUp/gqD9XrDGGDx+OJUuWIDs7G5GRkVrX\nDD0vTk5OAIC4uDgIIbQ+p06d0qutpzGn8Zw9exYAMGbMmGfW8/b2hoWFBbKysnSuVVdXA9A9JJ+I\niMhQmPQiIiIieoZ+/frBxsZG5/DvM2fOoLq6GoMHD9aUWVpaNnjGUVOlp6dDCIFhw4a1WB+tpXPn\nzpBIJHj06JFe90VGRqJ3797IyMjQKtdnXhrD3d0dcrkc58+f1+s+fZnLeLZv347u3btj1KhRAIDi\n4mLMnDlTp152djZqa2vh7u6uc61+LXTp0qVFYiQiImLSi4iIiOgZ5HI5wsLCsH//fuzcuROlpaXI\nzMzE/Pnz4eLiguDgYE1db29vPHjwAKmpqVCr1SgqKkJeXp5Om46OjigoKEBubi7Kyso0Say6ujo8\nfPgQNTU1uHjxIkJDQ+Hh4YGgoCCD9JGWlgaVSoXo6GjDf1HPoVQq4enpiVu3bul1X/22QAsLC53y\nxs5LY/uZPXs2du3ahcTERJSWlqK2tha3bt3CnTt3AACBgYHo0qULzp07p1fbpj6el156CXl5eaip\nqUFubi6WLl2K48ePY8eOHZqzxKytrXHs2DGcOHECpaWlUKvVyMjIwNtvvw1ra2ssWbJEp936teDr\n66vX2IiIiBqLSS8iIiKi5/jwww8RExODiIgIdOrUCaNGjUK3bt2Qnp4Oa2trTb0FCxZgzJgxeOON\nN9CrVy9ERkZqtm4NHz4c+fn5AID58+ejc+fO6Nu3LyZMmIAHDx4A+OVsI19fXygUCowcORI9e/bE\nd999p3UOVnP7MKaJEyciKytL6zyrr7/+Gt7e3sjJycHQoUPx3nvv6dw3bNiwBpMmjZmXxMRExMXF\nAQD69++P69evY/v27QgLCwMAjB8/HtnZ2QCA+Ph4LF68GBs2bEDHjh3h4uKC0NBQPHz4EMAv2/EK\nCwtx4MCBp47R3MYD/LL1csCAAVAoFBg0aBAuX76Mf/zjH1pbG+VyOUaMGIF33nkHrq6usLW1hb+/\nP7p164bTp09rHahf7+zZs3B1dUX//v2f2T8REVFTSYQQwthBEBERNYW/vz8AYM+ePUaOhEyNRCJB\nSkoKAgICjB2Kxrx587Bnzx4UFxcbO5QGGeLft2vXrqFPnz5ISkrCrFmzDBVaq6mrq8Po0aMRFBSE\nOXPmGDucZjPmeIqLi+Hm5oaoqChNwq45+PeAiIgawie9iIiIiNoIfQ55N0Xe3t6IiIhAREQEysvL\njR2OXmpra5GamoqysjIEBgYaO5xmM/Z41q5diwEDBiAkJKTV+yYiovaDSS8iIiIiajUrVqyAv78/\nAgMD9T7U3pjS09Oxb98+pKWlQalUGjucZjPmeGJjY3H+/HkcPnwYUqm0VfsmIqL2hUkvIiIiIiNb\nuXIlkpKS8OjRI3Tv3h179+41dkgtKjo6GiEhIfjoo4+MHUqjjR07Fl9++SWcnZ2NHYpBGGs8Bw4c\nwJMnT5Ceng4HB4dW7ZuIiNofS2MHQERERNTexcTEICYmxthhtKpx48Zh3Lhxxg6DWtmUKVMwZcoU\nY4dBRETtBJ/0IiIiIiIiIiIis8OkFxERERERERERmR0mvYiIiIiIiIiIyOww6UVERERERERERGaH\nSS8iIiIiIiIiIjI7EiGEMHYQRERETeHv74+9e/caOwwiImoDXn/9/2vvDqoAhkEgCnKojQjBR11G\nSIRgpDJIeTMK9vwP8Mbeu3sGABcRvQD4rXNOVFX3DAAusNaKzOyeAcBFRC8AAAAAxnHTCwAAAIBx\nRC8AAAAAxhG9AAAAABjniQgvTgAAAAAY5QOIjwgnmrgnawAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<IPython.core.display.Image object>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 2
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "SDqbJFaiuaQm",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import keras\n",
        "from keras import layers\n",
        "\n",
        "model = keras.Sequential()\n",
        "model.add(layers.Dense(20, activation='relu', input_shape=(10,)))\n",
        "model.add(layers.Dense(20, activation='relu'))\n",
        "model.add(layers.Dense(20, activation='softmax'))\n",
        "\n",
        "model.fit(x,y,epochs=10, batch_size=32)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "WNEx8JLV57D0",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import keras\n",
        "from keras import layers\n",
        "\n",
        "inputs = keras.Input(shape = (10,))\n",
        "x = layers.Dense(20, activation='relu', input_shape=(10,))(x)\n",
        "y = layers.Dense(20, activation='relu')(x)\n",
        "outputs = layers.Dense(20, activation='softmax')(x)\n",
        "\n",
        "model = keras.Model(inputs, outputs)\n",
        "model.fit(x,y,epochs=10, batch_size=32)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "pOv9b_by6jWG",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import keras\n",
        "from keras import layers\n",
        "\n",
        "class MyModel():\n",
        "    # initializing variables, idk where to store them\n",
        "    def __init__(self):\n",
        "        super(MyModel, self).__init__()\n",
        "        self.dense1 = layers.Dense(20, activation='relu')\n",
        "        self.dens2 = layers.Dense(20, activation='relu')\n",
        "        self.dense3 = layers.Dense(10, activation='softmax')\n",
        "    \n",
        "    def call(self, inputs):\n",
        "        x = self.dense1(x)\n",
        "        x = self.dense2(x)\n",
        "        return dense3(x)\n",
        "        \n",
        "        \n",
        "\n",
        "model = MyModel()\n",
        "model.fit(x,y,epochs=10, batch_size=32)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "Uipk9U5CFL1L",
        "colab_type": "text"
      },
      "source": [
        "ls -l|wc -l\n",
        "autoencoder for the system\n",
        "17000 vectors as the input\n",
        "number of patients are different\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Ywg8X-t4u-b_",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "import os\n",
        "import numpy as np\n",
        "import keras\n",
        "from keras.layers import Input, Dense\n",
        "from keras.models import Model\n",
        "from keras.callbacks import TensorBoard\n",
        "from tensorflow import set_random_seed"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "HJHabMnTvIRj",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        },
        "outputId": "a43e074d-2ba6-479f-b2d1-3c0d81b7059b"
      },
      "source": [
        "import keras  # ML library. Built on top of Tensor flow\n",
        "from keras.layers import Input, Dense\n",
        "from keras.models import Model\n",
        "from keras.callbacks import TensorBoard\n",
        "import numpy as np  # Scientific computing libarary (we will use for converting our data to NumPy arrays)\n",
        "from tensorflow import set_random_seed\n",
        "import os\n",
        "\n",
        "\n",
        "def seedy(s):  # set the seed for the RNG. For repeatable tests.\n",
        "    np.random.seed(s)\n",
        "    set_random_seed(s)\n",
        "\n",
        "\n",
        "class AutoEncoder:  # define the overall template for the model.\n",
        "    def __init__(self, encoding_dim=3):\n",
        "        self.encoding_dim = encoding_dim  # Setting the coding dimension.\n",
        "        r = lambda: np.random.randint(1, 3)\n",
        "        self.x = np.array([[r(), r(), r()] for _ in range(1000)])  # create data to pass in. The input.\n",
        "        print(self.x)\n",
        "\n",
        "    def _encoder(self):  # From starting info, compressed into the few abstract features.\n",
        "        inputs = Input(shape=(self.x[0].shape))\n",
        "        encoded = Dense(self.encoding_dim, activation='relu')(inputs)\n",
        "        model = Model(inputs, encoded)\n",
        "        self.encoder = model\n",
        "        return model\n",
        "\n",
        "    def _decoder(self):  # From the compressed abstract features, expanded back up to a full size.\n",
        "        inputs = Input(shape=(self.encoding_dim,))\n",
        "        decoded = Dense(3)(inputs)\n",
        "        model = Model(inputs, decoded)\n",
        "        self.decoder = model\n",
        "        return model\n",
        "\n",
        "    def encoder_decoder(self):  # Combine the two separate models to make the whole.\n",
        "        ec = self._encoder()\n",
        "        dc = self._decoder()\n",
        "\n",
        "        inputs = Input(shape=self.x[0].shape)\n",
        "        ec_out = ec(inputs)\n",
        "        dc_out = dc(ec_out)\n",
        "        model = Model(inputs, dc_out)\n",
        "\n",
        "        self.model = model\n",
        "        return model\n",
        "\n",
        "    def fit(self, batch_size=10, epochs=300):  # Use 'Stochastic Gradient Descent' to optimize our weights.\n",
        "        self.model.compile(optimizer='sgd', loss='mse')  # sgd = Stochastic Gradient Descent. mse = Mean Squared Error.\n",
        "        log_dir = './log/'   # Log directory. Keeps a record of training progress.\n",
        "        tbCallBack = keras.callbacks.TensorBoard(log_dir=log_dir, histogram_freq=0, write_graph=True, write_images=True)\n",
        "        self.model.fit(self.x, self.x,  # Input and target output are the same (self.x)\n",
        "                       epochs=epochs,  # 'n' epochs = 'n' forward+backward passes over the entire data set.\n",
        "                       batch_size=batch_size,  # Many batches make up the whole training set.\n",
        "                       # NOTE: batch_size=10, epochs=300 as stated on line 50.\n",
        "                       callbacks=[tbCallBack])  # Save the overall loss after each epoch (to visualize progress)\n",
        "\n",
        "    def save(self):  # To save the weights of the model. So you can reuse a trained model in the future.\n",
        "        if not os.path.exists(r'./weights'):\n",
        "            os.mkdir(r'./weights')\n",
        "        else:\n",
        "            self.encoder.save(r'./weights/encoder_weights.h5')\n",
        "            self.decoder.save(r'./weights/decoder_weights.h5')\n",
        "            self.model.save(r'./weights/ae_weights.h5')\n",
        "\n",
        "\n",
        "if __name__ == '__main__':  #\n",
        "    seedy(2)  # Seed 2 into the RNG.\n",
        "    ae = AutoEncoder(encoding_dim=2)  # create an instance of our AutoEncoder.\n",
        "    # encoding_dim=2  means that the AutoEncoder will take the input data and represent it as a 2D vector.\n",
        "    ae.encoder_decoder()  # calling the encoder_decoder function which is initializing the encoder and decoder models.\n",
        "    ae.fit(batch_size=50, epochs=300)  # fit data set (input) to the Neural Network. 50 items at a time, 300 times total\n",
        "    ae.save()  # Save the end results.\n",
        "    "
      ],
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "[[1 2 2]\n",
            " [1 1 2]\n",
            " [1 2 1]\n",
            " ...\n",
            " [1 1 2]\n",
            " [1 1 2]\n",
            " [2 1 1]]\n",
            "Epoch 1/300\n",
            "1000/1000 [==============================] - 0s 60us/step - loss: 1.7162\n",
            "Epoch 2/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.4822\n",
            "Epoch 3/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.3294\n",
            "Epoch 4/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.2907\n",
            "Epoch 5/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.2744\n",
            "Epoch 6/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.2628\n",
            "Epoch 7/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.2533\n",
            "Epoch 8/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.2451\n",
            "Epoch 9/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.2382\n",
            "Epoch 10/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.2320\n",
            "Epoch 11/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.2265\n",
            "Epoch 12/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.2217\n",
            "Epoch 13/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.2175\n",
            "Epoch 14/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.2136\n",
            "Epoch 15/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.2101\n",
            "Epoch 16/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.2068\n",
            "Epoch 17/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.2038\n",
            "Epoch 18/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.2013\n",
            "Epoch 19/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.1989\n",
            "Epoch 20/300\n",
            "1000/1000 [==============================] - 0s 25us/step - loss: 0.1966\n",
            "Epoch 21/300\n",
            "1000/1000 [==============================] - 0s 27us/step - loss: 0.1945\n",
            "Epoch 22/300\n",
            "1000/1000 [==============================] - 0s 27us/step - loss: 0.1927\n",
            "Epoch 23/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.1908\n",
            "Epoch 24/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.1892\n",
            "Epoch 25/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.1876\n",
            "Epoch 26/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.1862\n",
            "Epoch 27/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.1847\n",
            "Epoch 28/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.1834\n",
            "Epoch 29/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.1821\n",
            "Epoch 30/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.1810\n",
            "Epoch 31/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.1799\n",
            "Epoch 32/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.1789\n",
            "Epoch 33/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.1780\n",
            "Epoch 34/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.1770\n",
            "Epoch 35/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.1761\n",
            "Epoch 36/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.1753\n",
            "Epoch 37/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.1744\n",
            "Epoch 38/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.1736\n",
            "Epoch 39/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.1718\n",
            "Epoch 40/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.1656\n",
            "Epoch 41/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.1609\n",
            "Epoch 42/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.1573\n",
            "Epoch 43/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.1543\n",
            "Epoch 44/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.1512\n",
            "Epoch 45/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.1486\n",
            "Epoch 46/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.1461\n",
            "Epoch 47/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.1437\n",
            "Epoch 48/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.1413\n",
            "Epoch 49/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.1374\n",
            "Epoch 50/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.1340\n",
            "Epoch 51/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.1307\n",
            "Epoch 52/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.1278\n",
            "Epoch 53/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.1251\n",
            "Epoch 54/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.1226\n",
            "Epoch 55/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.1203\n",
            "Epoch 56/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.1182\n",
            "Epoch 57/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.1162\n",
            "Epoch 58/300\n",
            "1000/1000 [==============================] - 0s 27us/step - loss: 0.1144\n",
            "Epoch 59/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.1128\n",
            "Epoch 60/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.1112\n",
            "Epoch 61/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.1097\n",
            "Epoch 62/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.1083\n",
            "Epoch 63/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.1069\n",
            "Epoch 64/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.1059\n",
            "Epoch 65/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.1046\n",
            "Epoch 66/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.1036\n",
            "Epoch 67/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.1026\n",
            "Epoch 68/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.1017\n",
            "Epoch 69/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.1008\n",
            "Epoch 70/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.1000\n",
            "Epoch 71/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0993\n",
            "Epoch 72/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0985\n",
            "Epoch 73/300\n",
            "1000/1000 [==============================] - 0s 25us/step - loss: 0.0976\n",
            "Epoch 74/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0971\n",
            "Epoch 75/300\n",
            "1000/1000 [==============================] - 0s 25us/step - loss: 0.0964\n",
            "Epoch 76/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0959\n",
            "Epoch 77/300\n",
            "1000/1000 [==============================] - 0s 27us/step - loss: 0.0952\n",
            "Epoch 78/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0947\n",
            "Epoch 79/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0942\n",
            "Epoch 80/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0936\n",
            "Epoch 81/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0932\n",
            "Epoch 82/300\n",
            "1000/1000 [==============================] - 0s 27us/step - loss: 0.0927\n",
            "Epoch 83/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0923\n",
            "Epoch 84/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0919\n",
            "Epoch 85/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0915\n",
            "Epoch 86/300\n",
            "1000/1000 [==============================] - 0s 27us/step - loss: 0.0909\n",
            "Epoch 87/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0908\n",
            "Epoch 88/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0904\n",
            "Epoch 89/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0900\n",
            "Epoch 90/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0896\n",
            "Epoch 91/300\n",
            "1000/1000 [==============================] - 0s 29us/step - loss: 0.0893\n",
            "Epoch 92/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0891\n",
            "Epoch 93/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0888\n",
            "Epoch 94/300\n",
            "1000/1000 [==============================] - 0s 26us/step - loss: 0.0884\n",
            "Epoch 95/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0882\n",
            "Epoch 96/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0879\n",
            "Epoch 97/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0876\n",
            "Epoch 98/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0875\n",
            "Epoch 99/300\n",
            "1000/1000 [==============================] - 0s 26us/step - loss: 0.0872\n",
            "Epoch 100/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0869\n",
            "Epoch 101/300\n",
            "1000/1000 [==============================] - 0s 25us/step - loss: 0.0868\n",
            "Epoch 102/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0866\n",
            "Epoch 103/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0864\n",
            "Epoch 104/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0863\n",
            "Epoch 105/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0860\n",
            "Epoch 106/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0860\n",
            "Epoch 107/300\n",
            "1000/1000 [==============================] - 0s 28us/step - loss: 0.0857\n",
            "Epoch 108/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0856\n",
            "Epoch 109/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.0854\n",
            "Epoch 110/300\n",
            "1000/1000 [==============================] - 0s 26us/step - loss: 0.0854\n",
            "Epoch 111/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0850\n",
            "Epoch 112/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0850\n",
            "Epoch 113/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0848\n",
            "Epoch 114/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.0847\n",
            "Epoch 115/300\n",
            "1000/1000 [==============================] - 0s 25us/step - loss: 0.0846\n",
            "Epoch 116/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0844\n",
            "Epoch 117/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0844\n",
            "Epoch 118/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.0842\n",
            "Epoch 119/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0842\n",
            "Epoch 120/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.0841\n",
            "Epoch 121/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0839\n",
            "Epoch 122/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0839\n",
            "Epoch 123/300\n",
            "1000/1000 [==============================] - 0s 25us/step - loss: 0.0838\n",
            "Epoch 124/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0836\n",
            "Epoch 125/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0836\n",
            "Epoch 126/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0835\n",
            "Epoch 127/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0835\n",
            "Epoch 128/300\n",
            "1000/1000 [==============================] - 0s 27us/step - loss: 0.0834\n",
            "Epoch 129/300\n",
            "1000/1000 [==============================] - 0s 27us/step - loss: 0.0833\n",
            "Epoch 130/300\n",
            "1000/1000 [==============================] - 0s 26us/step - loss: 0.0833\n",
            "Epoch 131/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.0832\n",
            "Epoch 132/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.0832\n",
            "Epoch 133/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0831\n",
            "Epoch 134/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0831\n",
            "Epoch 135/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0830\n",
            "Epoch 136/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0829\n",
            "Epoch 137/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0829\n",
            "Epoch 138/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0828\n",
            "Epoch 139/300\n",
            "1000/1000 [==============================] - 0s 25us/step - loss: 0.0828\n",
            "Epoch 140/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0827\n",
            "Epoch 141/300\n",
            "1000/1000 [==============================] - 0s 27us/step - loss: 0.0827\n",
            "Epoch 142/300\n",
            "1000/1000 [==============================] - 0s 28us/step - loss: 0.0826\n",
            "Epoch 143/300\n",
            "1000/1000 [==============================] - 0s 25us/step - loss: 0.0827\n",
            "Epoch 144/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0825\n",
            "Epoch 145/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0825\n",
            "Epoch 146/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.0825\n",
            "Epoch 147/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0824\n",
            "Epoch 148/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0823\n",
            "Epoch 149/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0824\n",
            "Epoch 150/300\n",
            "1000/1000 [==============================] - 0s 26us/step - loss: 0.0824\n",
            "Epoch 151/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0822\n",
            "Epoch 152/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0823\n",
            "Epoch 153/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0822\n",
            "Epoch 154/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0823\n",
            "Epoch 155/300\n",
            "1000/1000 [==============================] - 0s 26us/step - loss: 0.0823\n",
            "Epoch 156/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0822\n",
            "Epoch 157/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.0821\n",
            "Epoch 158/300\n",
            "1000/1000 [==============================] - 0s 25us/step - loss: 0.0820\n",
            "Epoch 159/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0821\n",
            "Epoch 160/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0819\n",
            "Epoch 161/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0820\n",
            "Epoch 162/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0820\n",
            "Epoch 163/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0819\n",
            "Epoch 164/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0819\n",
            "Epoch 165/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0820\n",
            "Epoch 166/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0819\n",
            "Epoch 167/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0819\n",
            "Epoch 168/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0819\n",
            "Epoch 169/300\n",
            "1000/1000 [==============================] - 0s 25us/step - loss: 0.0819\n",
            "Epoch 170/300\n",
            "1000/1000 [==============================] - 0s 26us/step - loss: 0.0818\n",
            "Epoch 171/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0818\n",
            "Epoch 172/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0817\n",
            "Epoch 173/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0819\n",
            "Epoch 174/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0818\n",
            "Epoch 175/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0817\n",
            "Epoch 176/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0818\n",
            "Epoch 177/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0817\n",
            "Epoch 178/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0817\n",
            "Epoch 179/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0817\n",
            "Epoch 180/300\n",
            "1000/1000 [==============================] - 0s 27us/step - loss: 0.0817\n",
            "Epoch 181/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0816\n",
            "Epoch 182/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0817\n",
            "Epoch 183/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0816\n",
            "Epoch 184/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0816\n",
            "Epoch 185/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0817\n",
            "Epoch 186/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0816\n",
            "Epoch 187/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0816\n",
            "Epoch 188/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0817\n",
            "Epoch 189/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0816\n",
            "Epoch 190/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0815\n",
            "Epoch 191/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0816\n",
            "Epoch 192/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0815\n",
            "Epoch 193/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0816\n",
            "Epoch 194/300\n",
            "1000/1000 [==============================] - 0s 26us/step - loss: 0.0815\n",
            "Epoch 195/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.0816\n",
            "Epoch 196/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0816\n",
            "Epoch 197/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0816\n",
            "Epoch 198/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0815\n",
            "Epoch 199/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0815\n",
            "Epoch 200/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0815\n",
            "Epoch 201/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0815\n",
            "Epoch 202/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0815\n",
            "Epoch 203/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0816\n",
            "Epoch 204/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0815\n",
            "Epoch 205/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0814\n",
            "Epoch 206/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0814\n",
            "Epoch 207/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0816\n",
            "Epoch 208/300\n",
            "1000/1000 [==============================] - 0s 25us/step - loss: 0.0814\n",
            "Epoch 209/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0814\n",
            "Epoch 210/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0814\n",
            "Epoch 211/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0813\n",
            "Epoch 212/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0814\n",
            "Epoch 213/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0814\n",
            "Epoch 214/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0814\n",
            "Epoch 215/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0814\n",
            "Epoch 216/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0814\n",
            "Epoch 217/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0814\n",
            "Epoch 218/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0813\n",
            "Epoch 219/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0813\n",
            "Epoch 220/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0813\n",
            "Epoch 221/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0813\n",
            "Epoch 222/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0814\n",
            "Epoch 223/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0814\n",
            "Epoch 224/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0813\n",
            "Epoch 225/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0813\n",
            "Epoch 226/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0814\n",
            "Epoch 227/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0813\n",
            "Epoch 228/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0813\n",
            "Epoch 229/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0813\n",
            "Epoch 230/300\n",
            "1000/1000 [==============================] - 0s 28us/step - loss: 0.0814\n",
            "Epoch 231/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0813\n",
            "Epoch 232/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0814\n",
            "Epoch 233/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.0813\n",
            "Epoch 234/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0813\n",
            "Epoch 235/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0813\n",
            "Epoch 236/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0813\n",
            "Epoch 237/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0813\n",
            "Epoch 238/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0813\n",
            "Epoch 239/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0812\n",
            "Epoch 240/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.0813\n",
            "Epoch 241/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0814\n",
            "Epoch 242/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0813\n",
            "Epoch 243/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0812\n",
            "Epoch 244/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0813\n",
            "Epoch 245/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0813\n",
            "Epoch 246/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0813\n",
            "Epoch 247/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0813\n",
            "Epoch 248/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0812\n",
            "Epoch 249/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0813\n",
            "Epoch 250/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0813\n",
            "Epoch 251/300\n",
            "1000/1000 [==============================] - 0s 25us/step - loss: 0.0813\n",
            "Epoch 252/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0813\n",
            "Epoch 253/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0813\n",
            "Epoch 254/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0812\n",
            "Epoch 255/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.0813\n",
            "Epoch 256/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0813\n",
            "Epoch 257/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0812\n",
            "Epoch 258/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0812\n",
            "Epoch 259/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0812\n",
            "Epoch 260/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0813\n",
            "Epoch 261/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0812\n",
            "Epoch 262/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0813\n",
            "Epoch 263/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0813\n",
            "Epoch 264/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0812\n",
            "Epoch 265/300\n",
            "1000/1000 [==============================] - 0s 24us/step - loss: 0.0811\n",
            "Epoch 266/300\n",
            "1000/1000 [==============================] - 0s 19us/step - loss: 0.0812\n",
            "Epoch 267/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0812\n",
            "Epoch 268/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0812\n",
            "Epoch 269/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0812\n",
            "Epoch 270/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0812\n",
            "Epoch 271/300\n",
            "1000/1000 [==============================] - 0s 28us/step - loss: 0.0813\n",
            "Epoch 272/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0813\n",
            "Epoch 273/300\n",
            "1000/1000 [==============================] - 0s 26us/step - loss: 0.0812\n",
            "Epoch 274/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0812\n",
            "Epoch 275/300\n",
            "1000/1000 [==============================] - 0s 25us/step - loss: 0.0811\n",
            "Epoch 276/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0812\n",
            "Epoch 277/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0812\n",
            "Epoch 278/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0812\n",
            "Epoch 279/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0812\n",
            "Epoch 280/300\n",
            "1000/1000 [==============================] - 0s 25us/step - loss: 0.0813\n",
            "Epoch 281/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0812\n",
            "Epoch 282/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0812\n",
            "Epoch 283/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0811\n",
            "Epoch 284/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0811\n",
            "Epoch 285/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0812\n",
            "Epoch 286/300\n",
            "1000/1000 [==============================] - 0s 18us/step - loss: 0.0813\n",
            "Epoch 287/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0811\n",
            "Epoch 288/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0812\n",
            "Epoch 289/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0813\n",
            "Epoch 290/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0812\n",
            "Epoch 291/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0811\n",
            "Epoch 292/300\n",
            "1000/1000 [==============================] - 0s 23us/step - loss: 0.0811\n",
            "Epoch 293/300\n",
            "1000/1000 [==============================] - 0s 30us/step - loss: 0.0813\n",
            "Epoch 294/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0812\n",
            "Epoch 295/300\n",
            "1000/1000 [==============================] - 0s 20us/step - loss: 0.0811\n",
            "Epoch 296/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0812\n",
            "Epoch 297/300\n",
            "1000/1000 [==============================] - 0s 28us/step - loss: 0.0811\n",
            "Epoch 298/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0813\n",
            "Epoch 299/300\n",
            "1000/1000 [==============================] - 0s 21us/step - loss: 0.0812\n",
            "Epoch 300/300\n",
            "1000/1000 [==============================] - 0s 22us/step - loss: 0.0812\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "s2_TGlqhQ081",
        "colab_type": "text"
      },
      "source": [
        "Notes:\n",
        "after each epoch, we'll save the loss to visualize the loss later\n",
        "\n",
        "fit()\n",
        "input and target - same\n",
        "input is the random array and output is the same as that\n",
        "self.x = self.y\n",
        "train our model with some epoch - entire forward and backward path\n",
        "better predictions\n",
        "epoch is broken into pieces\n",
        "100 samples and 10 is the batch_size\n",
        "\n",
        "save()\n",
        "*   save the model and test it later how it performs\n",
        "\n",
        "we define the batch size and epochs and we're training on that\n",
        "check the loss, feed the loss and we're doing this certain number of times\n",
        "\n"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "CLX9ulNzveI_",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 122
        },
        "outputId": "a22aaa80-adc5-4e82-e98a-71b5e5875c0d"
      },
      "source": [
        "from keras.models import load_model  # Will be used to retrieve the saved weights of the trained NN.\n",
        "import numpy as np\n",
        "\n",
        "encoder = load_model(r'./weights/encoder_weights.h5')  # Load the TRAINED encoder.\n",
        "decoder = load_model(r'./weights/decoder_weights.h5')  # Load the TRAINED decoder.\n",
        "\n",
        "inputs = np.array([[1,2,2]])  # Feed the trained model the numbers 1,2,2\n",
        "x = encoder.predict(inputs)  # predicts the inputs (aka: compress to the abstract features.)\n",
        "y = decoder.predict(x)  # predict the outputs (aka: decompress from the abstract features to guess the original input)\n",
        "\n",
        "# print results for demonstration purposes.\n",
        "print('Input: {}'.format(inputs))\n",
        "print('Encoded: {}'.format(x))\n",
        "print('Decoded: {}'.format(y))\n"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Input: [[1 2 2]]\n",
            "Encoded: [[0.10115084 3.502789  ]]\n",
            "Decoded: [[1.1727815 1.7685553 2.1840215]]\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "stream",
          "text": [
            "/usr/local/lib/python3.6/dist-packages/keras/engine/saving.py:310: UserWarning: No training configuration found in save file: the model was *not* compiled. Compile it manually.\n",
            "  warnings.warn('No training configuration found in save file: '\n"
          ],
          "name": "stderr"
        }
      ]
    }
  ]
}