notebooks/qm9_example.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import json\n",
    "import numpy as np\n",
    "from unittest.mock import MagicMock\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from megnet.data.qm9 import load_qm9_faber, ring_to_vector, FeatureClean, Scaler\n",
    "from megnet.data.graph import GraphBatchGenerator\n",
    "from megnet.models import MEGNetModel"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load sample qm9 molecules\n",
    "data_json = '../megnet/data/tests/qm9/qm9.json'\n",
    "with open(data_json, 'r') as f:\n",
    "    data = json.load(f)\n",
    "\n",
    "# Create a mock database and the query return value \n",
    "# of the database is the qm9 molecules loaded previously\n",
    "db = type('Mock', (), {})\n",
    "db.find = MagicMock(return_value=data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:root:Start querying...\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Query results, and get graph representation from qm9 data\n",
    "atom_feature, bond_feature, state_feature, index1, index2, targets = \\\n",
    "    load_qm9_faber(db)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Clean up the features\n",
    "# categorical arguments mean the fields that will be considered as categorical\n",
    "# feature labels mean that total features considered in the current model\n",
    "atom_fc = FeatureClean(categorical=[\"type\", \"chirality\", \"hybridization\", \"donor\", \"acceptor\", \"aromatic\"],\n",
    "                       feature_labels=[\"type\", \"chirality\", \"ring_sizes\", \"hybridization\", \"donor\", \"acceptor\", \"aromatic\"])\n",
    "bond_fc = FeatureClean(categorical=[\"bond_type\", \"same_ring\"],\n",
    "                       feature_labels=[\"bond_type\", \"graph_distance\", \"same_ring\", \"spatial_distance\"])\n",
    "state_fc = Scaler()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "processed_atom_feature = atom_fc.fit_transform(atom_feature)\n",
    "processed_bond_feature = bond_fc.fit_transform(bond_feature)\n",
    "processed_state_feature = state_fc.fit_transform(state_feature)\n",
    "u0 = np.array(targets.loc[:, \"U0\"].values, dtype='float64').reshape((-1, 1)) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# convert internal energy to internal energy per atom\n",
    "nb_atoms = np.array([len(i) for i in processed_atom_feature])\n",
    "processed_target = u0 / nb_atoms[:, None]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "n_atom_feature = len(processed_atom_feature[0][0])\n",
    "n_bond_feature = len(processed_bond_feature[0][0])\n",
    "n_state_feature = processed_state_feature[0].shape[-1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(21, 26, 2)"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# feature dimensions\n",
    "n_atom_feature, n_bond_feature, n_state_feature"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# generator for training\n",
    "BATCH_SIZE = 4\n",
    "train_generator = GraphBatchGenerator(processed_atom_feature, processed_bond_feature, processed_state_feature, \n",
    "                                     index1, index2, processed_target, batch_size=BATCH_SIZE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /Users/shyue/miniconda3/lib/python3.7/site-packages/tensorflow/python/framework/op_def_library.py:263: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Colocations handled automatically by placer.\n"
     ]
    }
   ],
   "source": [
    "# setup the model\n",
    "model = MEGNetModel(n_bond_feature, n_state_feature, nfeat_node=n_atom_feature,\n",
    "                    n1=16, n2=16, n3=8, nblocks=1, npass=1, lr=1e-3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "# number of steps for exausting the training data generator in one epoch\n",
    "n_steps = int(np.ceil(len(processed_target) / BATCH_SIZE))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /Users/shyue/miniconda3/lib/python3.7/site-packages/tensorflow/python/ops/math_ops.py:3066: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use tf.cast instead.\n",
      "WARNING:tensorflow:From /Users/shyue/miniconda3/lib/python3.7/site-packages/tensorflow/python/ops/math_grad.py:199: div (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Deprecated in favor of operator or tf.math.divide.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/40\n",
      "3/3 [==============================] - 3s 836ms/step - loss: 14.6878\n",
      "Epoch 2/40\n",
      "3/3 [==============================] - 0s 8ms/step - loss: 14.3520\n",
      "Epoch 3/40\n",
      "3/3 [==============================] - 0s 10ms/step - loss: 13.4175\n",
      "Epoch 4/40\n",
      "3/3 [==============================] - 0s 8ms/step - loss: 13.2029\n",
      "Epoch 5/40\n",
      "3/3 [==============================] - 0s 14ms/step - loss: 12.4269\n",
      "Epoch 6/40\n",
      "3/3 [==============================] - 0s 7ms/step - loss: 11.6660\n",
      "Epoch 7/40\n",
      "3/3 [==============================] - 0s 10ms/step - loss: 10.7834\n",
      "Epoch 8/40\n",
      "3/3 [==============================] - 0s 7ms/step - loss: 9.7320\n",
      "Epoch 9/40\n",
      "3/3 [==============================] - 0s 9ms/step - loss: 8.5229\n",
      "Epoch 10/40\n",
      "3/3 [==============================] - 0s 7ms/step - loss: 7.0247\n",
      "Epoch 11/40\n",
      "3/3 [==============================] - 0s 8ms/step - loss: 5.0135\n",
      "Epoch 12/40\n",
      "3/3 [==============================] - 0s 8ms/step - loss: 2.9667\n",
      "Epoch 13/40\n",
      "3/3 [==============================] - 0s 6ms/step - loss: 1.1393\n",
      "Epoch 14/40\n",
      "3/3 [==============================] - 0s 7ms/step - loss: 0.3539\n",
      "Epoch 15/40\n",
      "3/3 [==============================] - 0s 6ms/step - loss: 1.2783\n",
      "Epoch 16/40\n",
      "3/3 [==============================] - 0s 8ms/step - loss: 1.1551\n",
      "Epoch 17/40\n",
      "3/3 [==============================] - 0s 7ms/step - loss: 0.4588\n",
      "Epoch 18/40\n",
      "3/3 [==============================] - 0s 7ms/step - loss: 0.2912\n",
      "Epoch 19/40\n",
      "3/3 [==============================] - 0s 8ms/step - loss: 0.3503\n",
      "Epoch 20/40\n",
      "3/3 [==============================] - 0s 7ms/step - loss: 0.3842\n",
      "Epoch 21/40\n",
      "3/3 [==============================] - 0s 8ms/step - loss: 0.3599\n",
      "Epoch 22/40\n",
      "3/3 [==============================] - 0s 7ms/step - loss: 0.2342\n",
      "Epoch 23/40\n",
      "3/3 [==============================] - 0s 7ms/step - loss: 0.1766\n",
      "Epoch 24/40\n",
      "3/3 [==============================] - 0s 8ms/step - loss: 0.1464\n",
      "Epoch 25/40\n",
      "3/3 [==============================] - 0s 8ms/step - loss: 0.1605\n",
      "Epoch 26/40\n",
      "3/3 [==============================] - 0s 6ms/step - loss: 0.1401\n",
      "Epoch 27/40\n",
      "3/3 [==============================] - 0s 9ms/step - loss: 0.1108\n",
      "Epoch 28/40\n",
      "3/3 [==============================] - 0s 8ms/step - loss: 0.1027\n",
      "Epoch 29/40\n",
      "3/3 [==============================] - 0s 7ms/step - loss: 0.1104\n",
      "Epoch 30/40\n",
      "3/3 [==============================] - 0s 7ms/step - loss: 0.1114\n",
      "Epoch 31/40\n",
      "3/3 [==============================] - 0s 6ms/step - loss: 0.0879\n",
      "Epoch 32/40\n",
      "3/3 [==============================] - 0s 8ms/step - loss: 0.0769\n",
      "Epoch 33/40\n",
      "3/3 [==============================] - 0s 7ms/step - loss: 0.0742\n",
      "Epoch 34/40\n",
      "3/3 [==============================] - 0s 7ms/step - loss: 0.0771\n",
      "Epoch 35/40\n",
      "3/3 [==============================] - 0s 8ms/step - loss: 0.0714\n",
      "Epoch 36/40\n",
      "3/3 [==============================] - 0s 8ms/step - loss: 0.0663\n",
      "Epoch 37/40\n",
      "3/3 [==============================] - 0s 8ms/step - loss: 0.0642\n",
      "Epoch 38/40\n",
      "3/3 [==============================] - 0s 8ms/step - loss: 0.0617\n",
      "Epoch 39/40\n",
      "3/3 [==============================] - 0s 6ms/step - loss: 0.0611\n",
      "Epoch 40/40\n",
      "3/3 [==============================] - 0s 6ms/step - loss: 0.0605\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x13ab72d68>"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Fit the model for 40 epochs\n",
    "model.fit_generator(train_generator, steps_per_epoch=n_steps, verbose=1, epochs=40)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x13bd55128>"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAhsAAAGJCAYAAAAjYfFoAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4zLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvnQurowAAIABJREFUeJzs3Xl8VNX5x/HPE/YAKgoVRUMURXFXoqL4U9CqaBV3oY3WDanWXbFUQcVaWpcuLtQFrWKF1rZuVevWKiruoqLivhGoG6sIBpDl+f1xbiAkM8lMMjN3ZvJ9v173dTPnnjn3mVzJPJ577jnm7oiIiIhkS0ncAYiIiEhxU7IhIiIiWaVkQ0RERLJKyYaIiIhklZINERERySolGyIiIpJVSjZEREQkq5RsiIiISFYp2RAREZGsUrIhIiIiWdU67gCKRdeuXb28vDzuMERERHLitddem+vu3VKpq2QjQ8rLy5k6dWrcYYiIiOSEmVWlWjcvbqOY2dFmdoOZTTGzb83MzWxikrrl0fFk291NOP+eZvaImc03syVm9paZnWtmrZr/6URERFq2fOnZGA3sCCwG/gdsncJ73gQeSFA+PZ0Tm9lhwL3AUuDvwHzgUOCPQH/gmHTaExERkbXlS7JxHiHJ+BjYB5icwnumufuY5pzUzNYBbgVWAgPcfWpUfgnwFHC0mQ1197R7S0RERCTIi9so7j7Z3T9yd8/xqY8GugF31yQaUTxLCb0tAKfnOCYREZGiki89G02xsZn9DNgAmAe86O5vpdnGvtH+sQTHngWqgT3NrJ27L2t6qCIiIi1XIScb+0fbamb2NHCCu89MsY2tov2HdQ+4+woz+wzYFtgceK/poYqIiLRceXEbJU3VwBVAX6BLtNWM8xgAPGlmHVNsa91ovzDJ8Zry9RIdNLPhZjbVzKbOmTMnxVOKiIi0LAWXbLj7bHe/1N1fd/dvou1Z4ADgZWALYFiOYhnv7hXuXtGtW0rzmoiIiLQ4BZdsJOPuK4Dbopd7p/i2mp6LdZMcryn/pqlxiYiItHRFk2xEau5lpHob5YNo37vuATNrDWwGrAA+bX5oKZo0CcrLoaQk7CdNytmpRUREsqHYko1+0T7V5OCpaD8owbG9gVLghZw9iTJpEgwfDlVV4B72w4cr4RARkYJWcMmGme1iZvXiNrP9CJODAUysc2xdM9vazDaq87Z7gLnAUDOrqFW/PfDr6OVNGQu+MaNGQXX12mXV1aFcRESkQOXFo69mdjhwePSye7Tfw8wmRD/PdfcR0c9/ALY0sxcIs44C7MCaOTMucfcX6pziCOAO4E7gxJpCd//WzE4lJB1PR+uqzAcGEx6LvYcwhXluzEzyxG6ychERkQKQF8kGsBNwQp2yzaMNoAqoSTbuIiQPuwIHAW2Ar4F/AOPcfUo6J3b3B8xsH2AUcBTQnjBt+vnA9Tmd1bSsLNw6qatr15yFICIikmmW+xnCi1NFRYU3e4n5mjEbtW+lmIXxGz/+MVx/vRIPERHJC2b2mrtXNF6zAMdsFLXKShg/Hnr2DElGz55wxx0wZgzccw9ssw38/e8h+RARESkQSjbyTWUlzJgBq1aF/QknwGWXwWuvheRj6FA44gj48ks9JisikqLFixdjZhxyyCFxh9IiKdkoFNtvDy++CFdfDY8/Dr16wckn6zFZEclrZpbWNmHChLhDbrKHH35YCU0S+TJAVFLRujVceCEcdhjssAN8//3ax2sek62sjCc+EZE6Lrvssnpl1157LQsXLuScc85hvfXWXnpqp512ykocHTt25L333qNTp05ZaV8apmSjEPXuXT/RqKHHZEUkj4wZM6Ze2YQJE1i4cCHnnnsu5eXlOYnDzNh6661zci6pT7dRClVZWXrlIiIFpKKigk6dOrFkyRJGjx7NFltsQdu2bTnzzDMBmDdvHldeeSX77LMPG2+8MW3btmXDDTfkqKOO4rXXXqvXXrIxGyNGjMDMmDp1KpMmTaJv37506NCBrl27cvzxxzN79uysfcaVK1dy/fXXs8suu9CxY0c6depEv379uP322xPWf/LJJznooIPo0aMH7dq1Y6ONNqJ///5cddVVa9X74osvOOecc+jduzelpaV06dKFPn36cMoppzBr1qysfZ6GKNkoVGPHQmnp2mWlpaFcRFqeIhwwvmrVKg455BAmTJjAPvvsw7nnnkufPn0AeOONN7jsssto3749hx12GOeffz4DBgzgkUceYc899+TZZ59N61xXX301p556Kr179+aMM85gyy23ZOLEiRx44IGsXLkyK5/tqKOO4pxzzuGbb77hZz/7Gaeccgqff/45p5xyCsOHD1+r/r333ssPf/hDXn31VQ488EAuuOACDj30UMyMW265ZXW9b7/9lt13351x48bRq1cvzjzzTE488UT69OnDPffcwyeffJLxz5ISd9eWga1v376ecxMnuvfs6R6GiLqPHJn7GEQkfhMnupeWrvlbAOH1xIlxR5ZQz549HfDPPvssaZ2+ffs64LvuuqsvWLCg3vF58+b5/Pnz65V//PHHvsEGG3hFRcVa5YsWLXLAf/SjH61VfsEFFzjg66+/vn/wwQery1etWuWDBw92wP/973+n9LkeeuihhOdIZPz48Q74nnvu6dXV1avLFy5c6Ntuu60D/q9//Wt1+QEHHOCAf/TRR/XamjNnzuqf//rXvzrgo0ePrldvyZIlvmjRopQ+SyqAqZ7id6TGbBSyysqwLVsGffrAo4+Gno1WreKOTERSce65MG1a89t56aXwd6C26mo45RS49dbmtb3TTnDttc1roxl++9vf1htECrD++usnrN+rVy8GDx7MHXfcwfz585PWq+vCCy+kd+81C4CbGcOGDePBBx/klVde4eCDD27aB0ii5lbJNddcQ4cOHVaXr7POOowdO5bDDz+c2267jcGDB68VU/v27eu11TXBZI+126yR6L25otsoxaBdO/jtb+Gtt+Cuu+KORkRyrW6i0Vh5Adltt92SHps8eTJHHnkkm2yyCW3btl39+Owdd9wBwOeff57yeSoq6k+EuemmmwKwYMGCNKNu3BtvvEH79u3ZY4896h3bd999V9epUVlZibuz0047ccYZZ3DPPffw5Zdf1nvv/vvvT7du3bjkkks45JBD+NOf/sS0adNYtWpVxj9DOtSzUSyOPRZ+/3sYPRqGDIEEWa2I5JlM9RiUlydeV6lnT3j66cycIwalpaV07tw54bGJEyfy05/+lE6dOrH//vuz2Wab0bFjR8yMJ554ghdffJFlaSRbiXpPWrcOX5GZHrOxdOlSli1bRnl5OWZW73jnzp3p2LEj33zzzeqyms967bXXcsstt3DjjTcC0K9fv9UDZSH0crz88suMGTOGhx9+mH//+98AbLjhhpx99tmMHDmSVjH0fivZKBZmcM01MGBA+AN20UVxRyQiuTJ2bP11lYpgwHiiL+Iao0ePpnPnzrzxxhtsvvnmax376KOPePHFF7MdXpO1b9+etm3b8vXXXyc8vmjRIr777jt69OixVvmRRx7JkUceyaJFi3jppZd48MEHueWWWzj44IN5++23V/8eNttsM+68805WrVrF9OnTefLJJxk3bhyjRo2iVatWjBw5MuufsS7dRikm++wDhx4abqnMmRN3NCKSK4nWVRo/vmgn+FuxYgVVVVXstNNO9RKN5cuX53WiUWPnnXdmyZIlvPzyy/WOTZ48GYBddtkl4Xs7d+7M/vvvzw033MB5551HdXU1//nPf+rVKykpYYcdduC8887j4YcfBuCBBx7I4KdInZKNYnPVVeH/bq64Iu5IRCSX6q6rVKSJBoTbGz169OCdd95h7ty5q8tXrVrFRRddxGeffRZjdKk5+eSTAfjFL36x1u2eRYsWMXr0aABOOeWU1eVPP/10wts5Nb0jpdFUCG+99dZav5Nk9XJNt1GKTZ8+MGwY3HQTnHUWbLll3BGJiGTceeedx4gRI9hhhx048sgjKSkp4ZlnnmHGjBkcdNBBPProo7HF9uabb3LiiScmPNa7d28uvvhihg0bxkMPPcTDDz/Mdtttx+DBg1m5ciX33Xcfs2bN4uSTT+awww5b/b5hw4axePFi9txzT8rLyykpKeGVV15hypQp9O7dmyOOOAKABx98kF/96lf079+fLbfckq5du1JVVcW//vUvWrVqxYgRI3LxK6hHyUYxGjMGJk6Eiy+Gf/4z7mhERDLu/PPPp1OnTowbN47bb7+djh07MmDAAP7xj39w6623xpps/O9//+POO+9MeKx///5cfPHFlJSUcP/99zNu3DjuvPNObrrpJsyMbbfdlksvvXStXg0Ia8w89NBDvP766zzxxBO0atWKsrIyxowZw1lnnbV6zZfBgwczZ84cpkyZwn333cfixYvZaKONOPTQQ7ngggsSPnWTCxbm5ZDmqqio8KlTp8YdxhqXXx6SjhdegASPVomIiDSHmb3m7illLxqzUawuuAC6dw+rxCqhFBGRGCnZKFadOoXejeefh5hGH4uIiICSjeJ28slhwOgvfwnLl8cdjYiItFBKNopZ69bhUdgPP4QNNyyq1SBFRKRw6GmUYvfttyHJqJnbv6oqzDQIRf0cvoiI5A/1bBS7UaPCJD+1VVeHchERkRxQslHsZs5Mr1xERCTDlGwUu7KyxOXusN9+cN99sGJFbmMSEZEWRclGsRs7Nqz+WFuHDmFJ+o8/hqOOgs02g1//Gm68MQwg1UBSERHJIA0QLXY1g0BHjQq3TsrKQgJSWQkrV8LDD8Of/gSXXLL2+zSQVEREMkTTlWdI3k1Xnq4ePeCLL+qX9+wZVpAUERGpRdOVS/q+/DJxuQaSiohIMynZkCDZQNJk5SIiIilSsiFBooGkAEcfnftYRESkqCjZkKCyEsaPD2M0zEKPxiabwN/+tmb2URERkSZQsiFrVFaGwaCrVoWnUR54AL7+Gs4/P+7IRESkgCnZkOT69oWRI2HCBHj00bijERGRAqVkQxp26aWwzTZhzo2FC+OORkREClBeJBtmdrSZ3WBmU8zsWzNzM5uYpO6WZjbSzJ4ys1lm9r2ZfW1m/zKzgWmetzw6V7Lt7sx8wgLWrh3ccUeYg2PEiLijERGRApQvM4iOBnYEFgP/A7ZuoO4VwBDgXeARYD6wFTAYGGxm57j79Wme/03ggQTl09NspzjttltINK6+Go45Bg44IO6IRESkgOTFDKJRj8T/gI+BfYDJwCR3Py5B3ROBN939jTrl+wD/ARwod/cks1St9Z5y4DPgTnc/sTmfoeBnEG3M0qWw006wZAlMnw6dO8cdkYiIxKjgZhB198nu/pGnkPm4+4S6iUZU/gzwNNAW2DPzUbZw7duH2ymzZsEvfhF3NCIiUkDyItnIoOXRPt010zc2s5+Z2cXRfodMB1YU9tgDzjsPbr4Znnoq7mhERKRAFE2yYWY9gf2AauDZNN++P3AzMDbav2lmk81Mc3XXdcUVsMUWMGRImPhLy9GLiEgjiiLZMLN2wCSgHTDG3VOd8rKaMOC0L9Al2mrGjAwAnjSzjhkPuJCVlsKPfwxz54ZbKu5rlqNXwiEiIgkUfLJhZq2Au4D+wN+B36X6Xnef7e6Xuvvr7v5NtD0LHAC8DGwBDGvg3MPNbKqZTZ0zZ07zPkgh+ctf6pdVV8OoUbmPRURE8l5BJxtRojEROAb4B3BcKoNMG+PuK4Dbopd7N1BvvLtXuHtFt27dmnvawpFs2XktRy8iIgkUbLJhZm2AvwFDgb8CP4mShEyp6arQbZS6tBy9iIikoSCTDTNrC/yT0KPxF+B4d1+Z4dP0i/afZrjdwpdoOfp27UK5iIhIHQWXbESDQe8HDgP+DJzk7qsaec+6Zra1mW1Up3wXM6v3OzCz/YDzopcJp01v0eouR9+6NXTsCIceGndkIiKSh/JiunIzOxw4PHrZPdrvYWYTop/nunvNwhw3AwcDc4HPgUvNrG6TT7v707VeHwHcAdwJnFir/A/Almb2AmEGU4AdgH2jny9x9xea9qmKXGVl2ABefBH22gsuuABuvTXeuEREJO/kRbIB7AScUKds82gDqAJqko3Non1X4NIG2nw6hfPeRUhEdgUOAtoAXxMGm45z9ykptCF77LFm7ZSjjoJBg+KOSERE8kherI1SDIp+bZTGLF0KffuGZeinT4f11os7IhERyaKCWxtFikD79nDnnfDVV3DOOXFHIyIieUTJhmRORQVcdFGY9OvBB+OORkRE8oSSDcmsSy6BHXcM05fPmxd3NCIikgeUbEhmtW0bbqfMmwdnnhl3NCIikgeUbEjm7bgjXHop3H033HNP3NGIiEjMlGxIdvzyl+HplNNPh9mz445GRERipGRDsqNNm3A7ZcECKC+HkpKw1zL0IiItTr5M6iXFaNq0kGQsWRJeV1WFgaOwZvZREREpeurZkOwZNQqWL1+7rLo6lIuISIuhZEOyZ+bM9MpFRKQoKdmQ7CkrS69cRESKkpINyZ6xY6G0dO2ykpJQLiIiLYaSDcmeykoYPx569gQzWH99WLUqPKkiIiIthpINya7KSpgxIyQZs2eHCb9GjAgDRUVEpEVQsiG506oVXH89zJoF11wTdzQiIpIjSjYkt/beG449Fq66Sk+liIi0EEo2JPdqejUuvDDeOEREJCeUbEjulZXByJHwj3/AM8/EHY2IiGSZkg2Jx4UXhqTjnHNg5cq4oxERkSxSsiHxKC2F3/0O3nwTbr017mhERCSLlGxIfI4+GvbZB0aPDqvDiohIUVKyIfExg+uuC4nGZZfFHY2IiGSJkg2J1447hmXnb7wR3nkn7mhERCQLlGxI/K64Ajp3hnPPBfe4oxERkQxTsiHx69oVfvUr+O9/YcMNw2Jt5eUwaVLckYmISAa0jjsAEQDWWy+M4ZgzJ7yuqgq3VyCsryIiIgVLPRuSHy65pP4tlOpqGDUqnnhERCRjlGxIfki2TorWTxERKXhKNiQ/lJWlVy4iIgVDyYbkh7Fjw6yitZWWhnIRESloSjYkP1RWwvjx0LNnGCgKMGSIBoeKiBQBJRuSPyorYcaMsDBbv37w1FOwfHncUYmISDMp2ZD8YxaeTqmqgrvuijsaERFpJiUbkp8OOgj69oXf/AZWrIg7GhERaQYlG5KfzMJqsJ98AnffHXc0IiLSDEo2JH8NHgzbbx+eSFm5Mu5oRESkiWJPNszsaDO7wcymmNm3ZuZmNrGR9+xpZo+Y2XwzW2Jmb5nZuWbWqgnn38bM/mFms81sqZl9YGaXm1mHpn8qyYiSktC78f77cO+9cUcjIiJNFHuyAYwGzgR2Aj5vrLKZHQY8C+wN3A+MA9oCfwTS6m83s92BV4HDgf8C1wHfApcC/zGzdum0J1lw1FGw9dbw61/DqlVxRyMiIk2QD8nGeUBvYB3g9IYqmtk6wK3ASmCAu5/i7hcSEpUXgaPNbGgqJ416Qe4ASoGj3f0n7j4S2B24F+gfxSZxatUqrI/y9tvw4INxRyMiIk0Qe7Lh7pPd/SP3uqtwJXQ00A24292n1mpjKaGHBBpJWGrZB+gDPOvuq7/F3H0V8Ivo5WlmNTNMSWyGDoVeveCKK+ov1iYiInkv9mQjTftG+8cSHHsWqAb2TPH2R9K23P1T4EOgJ7B5E+KUTGrdGi6+GF5/HR59NO5oREQkTYWWbGwV7T+se8DdVwCfAa1JLUFI2lbko2jfO50AJUuOPz5MZa7eDRGRglNoyca60X5hkuM15evloi0zG25mU81s6pw5c1I4pTRZmzbwy1/CSy/Bk0/GHY2IiKSh0JKNvOLu4929wt0runXrFnc4xe+kk6BHj9C7ISIiBaPQko2a3oZ1kxyvKf8mx21JLrRrB7/4BTz7bNhERKQgFFqy8UG0rzeOwsxaA5sBK4BPm9NWZMton2xMh8Th1FNhnXXgwAPDpF/l5TBpUtxRiYhIAwot2Xgq2g9KcGxvwpwZL7j7sua0ZWabE5KQKlJLXCRX7rsPliyBpUvDQNGqKhg+XAmHiEgeSyvZMLP9zexqM3vJzL4ws+/NbKGZfWRm/zSz082sR7aCBe4B5gJDzayiVlztgV9HL2+qE3OpmW1tZmV12noGeA/Y28wG16pfAlwVvbw5xfk/JFdGjYLly9cuq64O5SIikpesse9SMysFzgZ+BpQBNZNcLQXmAx0IT2zUlK8AHgJ+5+4vNhqA2eGE6cIBugMHEnoTpkRlc919RJ3690TnvzuKYTDhUdZ7gGNrJwhmNgCYDDzj7gPqnHt3Qg9Hm+i9M4H9gArgeWC/FHtJqKio8KlTpzZeUZqnpCTxo69mms5cRCSHzOw1d69ovGaYk6Khhk4GrgA2At4HLid8Cb/q7t/WqmeEL/t+hGThMOBwM7sHuNDdZzZwmp2AE+qUbc6auTKqgNXJhrs/YGb7AKOAo4D2wMfA+cD16fREuPvLZrZr9LkOADpH5/sVcGWqiYbkUFlZuHWSqFxERPJSgz0bZrYKeAD4rbu/mnKjYQ2TE4BfAre4+6+aG2i+U89GjkyaFMZoVFevKevQAW69FSor44tLRKSFyVjPBlDh7q+nG0DU63GDmd0KlKf7fpGkahKKUaNg5sxwS2XIECUaIiJ5rMEBok1JNOq8f6m7v9+cNkTqqayEGTNg5UqoqIBXXtEU5iIieazQHn0VWcMMzjwT3n0Xnnqq8foiIhILJRtS2IYMga5d4YYb4o5ERESSyGiyYWabm9mnZvZJJtsVSap9+zBg9KGHwq0VERHJO5nu2WhDGBBanuF2RZI77bRwS+WmmxqvKyIiOZfpZOMTwvokmzdWUSRjNt0UDj8cbrstTGUuIiJ5JaPJhruvcPcqd08w65JIFp11FsyfD3/7W9yRiIhIHRogKsVh771hu+3CQFE9BisikleUbEhxMAu9G9OmwfPPxx2NiIjU0tgMomsxs1SXW3d379WEeESarrISRo4MvRt77RV3NCIiEkm3Z6OEsLpr3a0La55CaduEdkWar2NHOPlkuO8++PzzuKMREZFIWkmBu5e7+2YJti5Ab+AxwhMpfbIRrEijzjgjTGN+yy1xRyIiIpGM9UC4+8fAkUAP4LJMtSuSls03hx/9KCQby5bFHY2IiJD5R1+XAv8BfpzJdkXScuaZMHs23HNP3JGIiAjZGVuxAuiehXZFUrP//tC7t9ZLERHJE5leG6UrcAQwK5PtiqSlpCT0brz8Mrz6atzRiIi0eOk++nppA+1sChwGrAtc1My4RJrnhBPg4otD78Zf/hJ3NCIiLVpayQYwppHj3wK/dvermxaOSIassw7ssQfcdRdMnAhlZTB2bJiLQ0REcirdZGNgkvJVwALgfXdf0byQRDJg0iSYMiX87A5VVWEpelDCISKSY+ZaRyIjKioqfOrUqXGHITXKy0OCUVfPnjBjRq6jEREpOmb2mrtXpFJXM31KcZo5M71yERHJGiUbUpzKytIrFxGRrMn0o6+bm9mnZvZJJtsVSdvYsVBaunZZhw6hXEREcirTPRttWLMgm0h8Kith/PgwRsMslJ16qgaHiojEINPJxifAZsDmGW5XJH2VlWEw6PLlsOGGWglWRCQmmV4bZYW7V7l7gscARGLSqhUccwz8+9+waFHc0YiItDgaICotw9ChsHQpPPhg3JGIiLQ4SjakZdhjD9hkE7j77rgjERFpcdKdQRQz6wj8HDgQ6AG0S1DN3b1XM2MTyZySEhgyBK6/HhYsgC5d4o5IRKTFSKtnw8zWA14GrgIqgK2ALsCGrHkKpW267YrkxNChYbDo/ffHHYmISIuSblIwGtgGOIWQZAD8EegE7Am8TngipU+mAhTJmL59oVcv3UoREcmxdJONwcCz7n6H11pUxYOXgIOBrYFRGYxRJDPMwq2UJ5+E2bPjjkZEpMVIN9nYFHit1utV1Bqz4e6zgUeBoc0PTSQLhg6FVavg3nvjjkREpMVIN9moJiQYNRYC3evU+ZowcFQk/2y3HWyzjW6liIjkULrJxixC70aNd4G9zax2O3sBXzU3MJGsqLmVMmWKZhQVEcmRdJONZ4B9zGoWm+DvQC/gETM7w8z+CfQDHslgjPWY2Ylm5o1sK1Nsa0YDbShpKkZDhoA7/POfcUciItIipDvPxp2ER1s3IfRy3AzsCxwOHBDVeZ7w1Eo2TQMuT3Ls/6KYHk2jvYXAtQnKF6cZlxSCrbaCnXcOt1LOPTfuaEREil5ayYa7vw6cXuv1CuBIM+sLbAHMAF5191WJW8gMd59GSDjqMbMXox/Hp9HkN+4+prlxSQEZOhRGjoTPPoPNNos7GhGRopaRybfc/TV3/7u7v5ztRKMhZrY94TbO58C/44pDCsCxx4b93/8ebxwiIi1Asc30OTza/9ndUxqzEWlnZseZ2cVmdo6ZDTSzVtkIUPJEeTn066dkQ0QkBxpMNsxshJm1b2rjZrazmR3U1Penea4OwHHASuC2NN/eHbgLGEsYu/EU8JGZ7ZPRICW/DB0K06bB++/HHYmISFFrrGdjLPCJmY00s41TadCCA83sfmAqsGNzg0zRscB6wGPuPiuN990B7EdIODoC2wO3ENZ5edTMksZvZsPNbKqZTZ0zZ06TA5eYHHNMeBRWvRsiIllltWYdr3/QrDfwB8I05CuBF4DnCEnEl8ACoD2wAWGa8n6s+eKeB1wG3JKLcRxm9jxhfZbB7v5QBtr7HXAB8IC7H9FY/YqKCp86dWpzTyu5NmAAfP01vPtuSDxERCQlZvaau1ekUrfBp1Hc/UPgEDPbEzgDOIrwaGmiDKXmL/UHhFVh73D3RSlH3Qxmti0h0fgfmZvj42ZCsrF3htqTfDR0KJx+Orz9NuywQ9zRiIgUpZQefXX3F4AXzOw0wpfvXkAZoUdjCTAbeAt42t3fyVKsDWnqwNCG1NwX6Zih9iQfHXUUnHlmmHNDyYaISFakO8/GIsIjpXnzWGk0gPV4wm2eP2ew6X7R/tMMtin5pls32G+/MG5j7FjdShERyYJiePT1GKAL8GiygaFm1sbMtjazXnXK+5hZvZ4LMysHxkUvJ2Y2XMk7Q4fCp5+CxtyIiGRFMSQbNbdQGpoxtAfwHvBknfL0woXjAAAgAElEQVQhwFdm9m8zu9HMrjKze6K6WxDGf/wu0wFLnjniCGjTRivBiohkSUEnG2bWhzB+pKkDQycDDxMWk/sJcD6wD+GJmxOAQ9z9+8xEK3lrvfXC0vPXXQclJWHCr0mT4o5KRKRopLsQW15x9/dY8xRMQ/VmJKrn7s8QVrKVlmzSJHjnHVgZjS2uqoLhUYdZZWV8cYmIFImC7tkQyYhRo+D7Oh1Y1dWhXEREmk3JhsjMmemVi4hIWpRsiJSVpVcuIiJpyViyYWZdEj1GKpL3xo6F0tK1y0pLQ7mIiDRbWsmGme1nZlebWZdaZT8ws2eAucB8M/tDpoMUyarKShg/fk1PRqdO4bUGh4qIZES6PRtnAUe6+4JaZb8jrJfyCWHxtXPM7NgMxSeSG5WV4SmUww6DH/xAiYaISAalm2zsSJiDAgAz6wAcDfzH3XsDWwGzgNMyFqFILg0cGGYT1eBQEZGMSTfZ+AHwRa3XuxOWmJ8Aq9dOeZiQdIgUnn33DfvJk+ONQ0SkiKSbbCwDOtR6XbPc/LO1yr4F1m9mXCLx2HZb6NoVnnoq7khERIpGusnGZ8C+tV4fBXzk7p/XKtuUMFhUpPCUlMCAAaFnwz3uaEREikK6ycadwPZm9rKZTQG2B/5ap84OwAeZCE4kFvvuC7NmhbEbIiLSbOkmGzcBdwMVQH/C+Iyrag6a2XaEBOTpDMUnknsDB4a9bqWIiGREWsmGuy93958AXYB13f0wd19Wq8pXwM7ADRmMUSS3ttoKunfXIFERkQxp0qqv7v5tkvK5aLyGFDqzcCvlqafCuA1rdGFhERFpQLoziHYxs23MrF2d8pPM7F9m9lcz2y2zIYrEYOBA+OoreP/9uCMRESl46Y7Z+A3wcu33mdlZwG3AocBQ4Gkz2yZjEYrEoWbchm6liIg0W7rJRn/gSXdfUqtsBPA5sDdQM035+RmITSQ+m28e1krRIFERkWZLd8xGD+DJmhdRD8amwEh3fy4qO4aQeIgULrPQu/Hww7BqVZh/Q0REmiTdv6AdgKW1XvcnzCD631plnxCSEpHCNnAgzJsH06fHHYmISEFLN9n4HNi61usDCdOTv1mrrAtQ+zaLSGHSfBsiIhmRbrIxGTjYzM40s2HAYOAxd19Vq04vwsqvIoWtrAx69dIgURGRZko32fgtsBi4DhhPuKUypuagma0D7AW8kKH4ROI1cCA88wysXBl3JCIiBSvdGUQ/A7YFzgHOBrZz99rroGwB3EK05LxIwdt3X1i4EN54I+5IREQKVtoziLr7V8C4JMdeB15vblAieWPAgLCfPBkqKmINRUSkUDX5eT4za2Nm25vZ/5nZDmbWJpOBieSFjTaCrbfWuA0RkWZIO9kws3XM7GbgG2AaYYXXN4BvzOxmM1svsyGKxGzffeHZZ2H58rgjEREpSOmujbIO8DwwHFgBTAH+Ee2XR+XPRfVEisPAgfDddzB1atyRiIgUpHR7Ni4iDBC9Cejp7gPc/cfuPgDoCfwJ2CaqJ1Icao/bEBGRtKWbbBwJvOTuZ7j7N7UPuPtCdz8LeBE4KlMBisSua1fYYQdN7iUi0kTpJhs9CWM0GvIMYb0UkeIxcCA8/zwsWxZ3JCIiBSfdZOM74AeN1OkGVDctHJE8te++sHQpvPxy3JGIiBScdJONV4FjzGzLRAfNrBdhmflXmxuYSF7Ze++w8qtupYiIpC3dZOMaoBPwqpldYWb7mlkfMxtoZpcTkoxOwO8yHahIrNZbD3beWYNERUSaIN3pyp8Efg60By4G/gNMJywxfwnQETjT3f+btBGRQrXvvvDSS1Ctu4QiIulIe1Ivd78F6A1cCtwPPBXtLwF6u/tNGY1QJF8MHAjffw8vaJ1BEZF0pL02CoC7zwTGJjpmZu2Btu7+bXMCa4yZzSA8HZPI1+7ePY22NgF+BQwCNgC+BB4ALnf3Bc0MVYrFXntBq1bhVsoPfxh3NCIiBaNJyUYjbgKOz1LbdS0Erk1QvjjVBqJBrS8QnrL5F/A+sBthZdtBZtbf3edlIFYpdJ07w267adyGiEiaspUQWJbaresbdx/TzDZuJCQaZ7v7DTWFZvYH4DxCD85pzTyHFIuBA+Gqq2DRopB8iIhIo5q86msxiHo1DgBmEKZar+0ywrwix5tZxxyHJvlq4EBYuRKeey7uSERECkahJxvtzOw4M7vYzM6JHsFtlcb7B0b7J9x9Ve0D7r6IsOhcKdAvQ/FKoZs1K+wPPhjKy2HSpFjDEREpBLkYV5FN3YG76pR9ZmYnufszKbx/q2j/YZLjHxF6PnoDTzYtRCkakybBmWeueV1VBcOHh58rK+OJSUSkABRyz8YdwH6EhKMjsD1wC1AOPGpmO6bQxrrRfmGS4zXl6zU9TCkao0bVn2OjujqUi4hIUgXbs+Hul9cpmg6cZmaLgQuAMcAR2YzBzIYDwwHKysqyeSrJBzNnplcuIiJACj0bZrYynQ34aQ7ibsjN0X7vFOrW9Fysm+R4Tfk3iQ66+3h3r3D3im7duqURohSkZAmlEk0RkQalchvFmrDFaU60T+UJkg+ife8kx2sWnEs2pkNakrFjobR07bLS0lAuIiJJNXobxd0LbVxHzZMjn6ZQt2Z2pgPMrKT2Eylm1hnoD1QDL2U2RClINYNAR40Kg0MB/vhHDQ4VEWlEoSUSAEQrzdbruTCzcmBc9HJirfI2ZrZ1NK/Gau7+CfAEYVDpGXWau5zQO3KXu3+XseClsFVWwowZ8Ez0sFP3lGfFFxFpsQp1gOgQ4AIzexaoAhYBvYAfEVakfYS1l7nvAbwX1S2v09bPCdOVX29m+0X1difMwfEhoEcNpL7ddoP27eHpp2Hw4LijERHJa4WabEwmzJGxM+FWR0fCIM7nCPNu3OXunkpD7v6JmVWwZiG2gwkLsV2HFmKTZNq3hz32CMmGiIg0qCCTjWjCrlQm7aqpP4MGBq66+yzgpOZHJi3KgAEwZgwsWABdusQdjYhI3irIMRsieWHgQHCHZ5+NOxIRkbymZEOkqWqP2xARkaSUbIg0Vbt2sOeeSjZERBqhZEOkOQYMgDffhPnz445ERCRvKdkQaY4BA8K4jSlT4o5ERCRvKdkQaY7ddoMOHWDy5Mbrioi0UEo2RJpD4zZERBqlZEOkuQYMgLfe0rgNEZEklGyINFfNuA3NtyEikpCSDZHm2nXXMG5Dt1JERBJSsiHSXO3aQf/+GiQqIpKEkg2RTKgZtzFvXtyRiIjkHSUbIpkwYEDYa9yGiEg9SjZEMkHjNkREklKyIZIJbduGcRtKNkRE6lGyIZIpGrchIpKQkg2RTBk4MOyfeSbeOERE8oySDZFMqaiA0lLdShERqUPJhkimaNyGiEhCSjZEMmnAAHj7bZg7N+5IRETyhpINkUzSfBsiIvUo2RDJpF13DeM2NHW5iMhqSjZEMqlNG9hrL43bEBGpRcmGSKYNGADTp8OcOXFHIiKSF5RsiGSaxm2IiKxFyYZIplVUQMeOupUiIhJRsiGSaTXjNjRIVEQEULIhkh0DBsA778Ds2XFHIiISOyUbItmgcRsiIqsp2RDJhr59NW5DRCSiZEMkG9q0gc03h/HjoaQEysth0qS4oxIRiUXruAMQKUqTJsH778Py5eF1VRUMHx5+rqyMLy4RkRioZ0MkG0aNWpNo1KiuDuUiIi2Mkg2RbJg5M71yEZEipmRDJBvKytIrFxEpYko2RLJh7Niw+mttpaWhXESkhSnIZMPMNjCzYWZ2v5l9bGZLzGyhmT1nZqeYWcqfy8xmmJkn2b7K5ueQIlZZGZ5E6dEjvO7SJbzW4FARaYEK9WmUY4CbgC+BycBMYEPgSOA24CAzO8bdPcX2FgLXJihfnIFYpaWqrAzbttuGpEOJhoi0UIWabHwIDAb+7e6ragrN7GLgFeAoQuJxb4rtfePuYzIdpAgAgwbBn/4Unkape2tFRKQFKMjbKO7+lLs/VDvRiMq/Am6OXg7IeWAiiRx4ICxbBs88E3ckIiKxKNSejYbUTG6wIo33tDOz44Ay4DvgLeBZd1+Z6eCkBdp7b+jQAR57DA46KO5oRERyrqiSDTNrDfw0evlYGm/tDtxVp+wzMzvJ3fW/o9I87duHhdkeS+c/SRGR4lGQt1EacCWwHfCIuz+e4nvuAPYjJBwdge2BW4By4FEz2zELcUpLM2gQfPghfPpp3JGIiORc0SQbZnY2cAHwPnB8qu9z98ujMSBfu3u1u09399OAPwAdgDENnHO4mU01s6lz5sxp5ieQojZoUNg/nmoOLCJSPIoi2TCzM4HrgHeBge4+PwPN1gw03TtZBXcf7+4V7l7RrVu3DJxSitaWW8Jmm+lWioi0SAWfbJjZucANwHRCopGpibhquio6Zqg9acnMQu/Gk0/C99/HHY2ISE4VdLJhZiOBPwLTCInG7Aw23y/a6ya7ZMagQfDdd/D883FHIiKSUwWbbJjZJYQBoa8B+7n73AbqtjGzrc2sV53yPmZWr+fCzMqBcdHLiRkLWlq2gQOhTRvdShGRFqcgH301sxOAXwErgSnA2WZWt9oMd58Q/dwDeA+oIjxlUmMIcIGZPRsdWwT0An4EtAceAX6XlQ8hLU/nzrDXXiHZuOqquKMREcmZgkw2gM2ifSvg3CR1ngEmNNLOZGArYGegP2F8xjfAc4R5N+5KY30VkcYNGgQjR8IXX8DGG8cdjYhITpi+SzOjoqLCp06dGncYku/eegt23BFuvx1OOinuaEREmszMXnP3ilTqFuyYDZGCtP32sNFGGrchIi2Kkg2RXKp5BPY//4GVWnpHRFoGJRsiuXbggbBgAbz6atyRiIjkhJINkVz74Q+hpES3UkSkxVCyIZJrG2wAu+2mZENEWgwlGyJxGDQIXnkF5s2LOxIRkaxTsiESh0GDwD0MFBURKXJKNkTiUFEB66+vWyki0iIo2RCJQ6tWcMABIdlYtSruaEREskrJhkhcBg2Cr78Os4qKiBQxJRsicTnggLDXrRQRKXJKNkTistFGsNNOSjZEpOgp2RCJ06BB8Pzz8O23cUciIpI1SjZE4jRoEKxYAU89ldl2J02C8vIwU2l5eXgtIhITJRsicdpjD+jcObO3UiZNguHDoaoqzOVRVRVeK+EQkZgo2RCJU9u2sN9+Idlwz0ybo0ZBdfXaZdXVoVxEJAZKNkTi1qVL6H1o1SoztzxmzkxcXlUFf/4zfPTRmsRGt1tEJAdaxx2ASIs2aRLcfXf4ufYtD4DKyqa1WVYW2qmrpASGDQs/d+8e6r3xBixfHsoycW4RkQTUsyESp1GjYMmStcuae8vjZz+rX1ZaCn/5C7z7Ltx8c7h189praxKNTJ1bRCQB80zdJ27hKioqfOrUqXGHIYWmpCTxWA2zpk1j7g4HHhgep11/ffj889CDMXZs/d6KTJ9bRFoUM3vN3StSqaueDZE4lZWlV96Yxx8PK8mOHQuzZoWkYcaMxLdFMn1uEZEklGyIxGns2HCLo7bWrUN5ulasgBEjoFcv+PnPm3ZugB//OP1zi4g0QMmGSJwqK2H8eOjZM9y+6Nw5JA3l5em3NWECvPMOXHVVeKQ23XNvuilsvDHccUdYIE5EJEM0ZiNDNGZDMmLxYthuO2jfHqZNC/tU37fllrD55vDccyF5aIrp02HXXeH//i/M/VGi/x8RkcQ0ZkOkUHXqBLfeCh98AJdfnvr7fvc7+Oor+P3vm55oQEh0rrsujPu4+uqmtyMiUouSDZF8s//+cPLJcM014fHUxnzxRah77LHQr1/zz3/qqXDMMTB6NLzwQvPbE5EWT8mGSD76/e/hBz8IScf33zdc95JLwjiP3/42M+c2C70rZWVhsOj8+ZlpV0RaLCUbIvlovfXC5FtvvRUGfCbz1lthQOdZZ4XxGpmy7rphZtMvvgizjmpsl4g0g5INkXw1eDAMHQpXXBGeMknkwgtDYpKNWT932y30ltx/P9x4Y+bbF5EWQ8mGSD67/vrQy3DyybBy5drHHn8cnngCLr00LOaWDeefDwcdFPbTpmXnHCJS9JRsiOSzbt3ghhvglVfg2mvXlK9cmd4EXk1VUgJ33gkbbBCSjrIyrRArImlTsiGS74YMCbdURo8Oy8NDmMBr+nS48srUJvBqjm7d4KSTwqO1s2atvTptrhKOSZNCgqNER6QgKdkQyXdmcNNN0K4dHHpo6F0YNiy8XrYsNzEk+nLPxAqxqSQRkyaFxKaqqvFER0mJSF5SsiFSCDbeOMx98cEHoXcBQqKRq96FmTMTl1dVwT//GWYwrZHqF35DSYQ7LFoUznvhhSGxqa26OpTPmAFz58LSpeklJSKSU5quPEM0XblkXc+eib/0e/YMX7rZVF4evrzrKikJK8u2bx/GdGy0UXgUd8mSNXVKS8MaLD/5CcybF5KlmTPDrZkFCxK3WVIS5g7JhB494H//q18+aVLomZk5M/QWjR2beHVcEUkonenKlWxkiJINybqSksTzXZiFL/xsquk1qN3DUFoa5gIpK4N77oH77gvzciTSujW0abN2EtKQiy4KT9h06RJ+nju3fp2uXcOU6osXh+3ii5O3t+WWsPfesM8+YZsyJfHnGT9eCYdIitJJNnD3gtyATYDbgS+AZcAM4FqgS5rtrB+9b0bUzhdRu5uk007fvn1dJKt69nQP6cbaW8+euTn/xInhXGZhP3Hi2sdXrgzHEsUI7uef737tte733uv+6qvum2yS2ueZONG9tHTtOqWl9c+f7Pez3nruhx4a9jVlrVql/rts7HM3pW6m64nEAJjqqX7XploxnzagF/A14MADwJXAU9Hr94ENUmxnA+CD6H1PRu08EL3+Gtg81ZiUbEjWpfqlG6d0EqJ0Pk8qX7qNtbdypfu0ae7XXZc8IQL3Pfd0P/ZY9wsucD/uOPd27VKPMZXPk+l66fyO0qmXjTaVPBWVlpBsPB4lBGfVKf9DVH5ziu3cEtX/fZ3ys6Pyx1KNScmG5ES+/7HO1hdkOudPpb1kSVHHju4DB7pvsYV7+/bJExIz97Iy9x12cP+//3M/5JD6n7t2m0OGuB9+uPtBB9VPXmr/nk44wf30091HjHBfd93E9bp3d3/vPfcvvnBfvNh91arsJDBKnvL73NlqMw1FnWxEvRoOfAaU1DnWGVgMfAd0bKSdTkB1VL9znWMl0W0VT7V3Q8mGSCTfEyL31L74Vq1q+LbQT3/qfthh7gMGuO+8c/J64N67t/v227tXVDRcr2dP965dkycuibZWrdxLShIfa9fO/YAD3A8+2H3wYPcOHRLX69AhJEyDBrnvv39IuJIlRW3buu+yS9h23tm9TZvE9UpL3YcNcz/3XPdRo9a+jVV7+8EP3J9+2v3558PttWnT3K+6qn6y16GD+5/+5P711+6zZ7vPnet+yy31P1OHDu433+w+Z06o+9VXITG74YbkdRcscP/2W/fvvnOfMCH/E7dstZmmdJKNghsgambDgFuB8e7+swTHHwcOAH7o7k820M4Pgf8AT7j7gQmO3wIMB4a5+58bi0sDREUKTCpPoyR7CifRE0Cp1k21XrKnj7p1C9PYL1wYtm+/DbEn069feLJnxYqGp5zfZRdo1SoM5m3dOgyiTeZHPwoDkwEefjh5vY02gu++CwN4sz2IORc6dgwDtVu1Cr/3RJ+pVavwBJTZmm3WrMRPV7VuDb17r6n3wQewfHn9em3awA47rF329tuJV4Ru2xZ22in8bAZvvJG4XgaeYktngGjrZp0pHltF+w+THP+IkGz0JozDaE47RO2ISLGprGz8yZOxYxM/tZLoyz3VuqnW+81vEtf74x/DAn21TZyYPIF58cU1rxtKdF57be2yhurWTjBSSZ7cw+uaOWJq23BD+Nvfwhfi8uVhf/TR4T2JjBu35v/Rzz47cR0I0/ybheTADE4/PXndP/whLAGwciX88pfJ65122pp6N9yQuM7KlTBw4Nr9NxMnJq67YgX06bOm3vTpiestXw7du6957V7/etX4/ntYf/01v79EiQYknzsnW1LtAsmXDRhPuL0xLMnxsdHxixpp5+Ko3q+THD81On5LA20MB6YCU8vKytLsgBKRglAIYwIKoZs+nXOnOtA4nQHJmW4zznNnq800UeRjNvIm2ai9acyGiMSqEAYgttTkqVDaTFOxJxvXREnABUmOj4uOn95IO2dE9W5IcnxEdPyqVOJSsiEikkHFlDwVUptpSCfZ0ABRDRAVERFJWzoDRAtxIbbJ0f4AM1srfjPrDPQnPNL6UiPtvAQsAfpH76vdTgkhYal9PhEREWmCgks23P0T4AmgnHArpLbLgY7AXe7+XU2hmW1tZlvXaWcxcFdUf0ydds6M2n/c3T/NYPgiIiItTiE++grwc+AF4Hoz2w94D9gdGEh4lHVUnfrvRXurU34xMAA438x2Al4B+gCHAbOpn8yIiIhImgquZwNW925UABMIScYFhJlFrwP6ufu8FNuZB+wBXA9sEbWzO3AH0Dc6j4iIiDRDofZs4O6zgJNSrFu3R6P2sfnAOdEmIiIiGVaQPRsiIiJSOJRsiIiISFYp2RAREZGsUrIhIiIiWaVkQ0RERLKq4KYrz1dmNgdIsM5yk3UF5mawPcksXZ/8puuT33R98luq16enu3dLpUElG3nKzKamOue85J6uT37T9clvuj75LRvXR7dRREREJKuUbIiIiEhWKdnIX+PjDkAapOuT33R98puuT37L+PXRmA0RERHJKvVsiIiISFYp2RAREZGsUrKRR8xsEzO73cy+MLNlZjbDzK41sy5xx9ZSmNnRZnaDmU0xs2/NzM1sYiPv2dPMHjGz+Wa2xMzeMrNzzaxVruJuCcxsAzMbZmb3m9nH0e96oZk9Z2anmFnCv2e6PrljZleZ2ZNmNiv6Xc83szfM7DIz2yDJe3R9YmRmx0V/59zMhiWpc4iZPR39e1tsZi+b2QlpnUdjNvKDmfUCXgB+APwLeB/YDRgIfAD0d/d58UXYMpjZNGBHYDHwP2BrYJK7H5ek/mHAvcBS4O/AfOBQYCvgHnc/JhdxtwRmdhpwE/AlMBmYCWwIHAmsS7gOx3itP2q6PrllZt8DrwPvArOBjkA/oAL4Aujn7rNq1df1iZGZbQq8DbQCOgGnuvttdeqcCdwAzCNco++Bo4FNgN+7+4iUTubu2vJgAx4HHDirTvkfovKb446xJWyE5G5LwIAB0e9+YpK66xD+oC4DKmqVtyckjg4MjfszFcsG7Ev4IiqpU96dkHg4cJSuT6zXqH2S8rHR7/tGXZ/82KK/cf8FPgGuiX7fw+rUKSckgvOA8lrlXYCPo/fskcr5dBslD0S9GgcAM4A/1Tl8GfAdcLyZdcxxaC2Ou0929488+hfViKOBbsDd7j61VhtLgdHRy9OzEGaL5O5PuftD7r6qTvlXwM3RywG1Dun65Fj0u03kH9F+y1pluj7xOpuQwJ9E+I5J5GSgHTDO3WfUFLr7AuA30cvTUjmZko38MDDaP5HgD+ki4HmglNAdKflj32j/WIJjzwLVwJ5m1i53IbVYy6P9ilpluj7549Bo/1atMl2fmJhZH+BK4Dp3f7aBqg1do0fr1GmQko38sFW0/zDJ8Y+ife8cxCKpS3rd3H0F8BnQGtg8l0G1NGbWGvhp9LL2H0Vdn5iY2QgzG2NmfzSzKcAVhETjylrVdH1iEP17uYtw6/HiRqo3dI2+JPSIbGJmpY2dt3WacUp2rBvtFyY5XlO+Xg5ikdTpuuWHK4HtgEfc/fFa5bo+8RlBGLxb4zHgRHefU6tM1ycelwI7A3u5+5JG6qZyjTpG9aobakg9GyJSsMzsbOACwtNbx8ccjkTcvbu7G2Hw7pGE3ok3zGyXeCNr2cxsd0Jvxu/d/cVcnlvJRn6oyRrXTXK8pvybHMQiqdN1i1H0SN51hMcsB7r7/DpVdH1i5u5fu/v9hAHwGwB/qXVY1yeHotsnfyHcErkkxbeleo2S9XyspmQjP3wQ7ZONyagZwZ1sTIfEI+l1i/5hb0YYsPhpLoNqCczsXMKz/9MJicZXCarp+uQJd68iJIXbmlnXqFjXJ7c6EX7XfYCltSbycsJTjwC3RmXXRq8bukYbEW6h/M/dG7yFAko28sXkaH9A3VkQzawz0J9wP+ylXAcmDXoq2g9KcGxvwhNEL7j7styFVPzMbCTwR2AaIdGYnaSqrk9+2Tjar4z2uj65tQz4c5LtjajOc9HrmlssDV2jg+rUaVjcE4toWz1Jiib1yrON1Cb1moMmJcrlNbkk+r1OBdZvpK6uT26vTW9g3QTlJayZ1Ot5XZ/824AxJJ7UazMyNKmXpivPEwmmK38P2J0wB8eHwJ6u6cqzzswOBw6PXnYHDiR0406JyuZ6rel5o/r3EP5B3k2Ybnkw0XTLwLGuf2QZEa3FMIHwf8Y3kPg+8Qx3n1DrPbo+ORLd2vot4f+OPyN8QW0I7EMYIPoVsJ+7v1vrPbo+ecDMxhBupSSarvws4Ho0XXnxbMCmwB2EtR++B6qAa4EuccfWUjbWZPjJthkJ3tMfeARYACwhrDVwHtAq7s9TTFsK18aBp3V9Yrs+2wHjCLe35hLGWywEXo2uXcKeKF2f+DeS9GzUOn4o8AywiDC3xqvACemcQz0bIiIiklUaICoiIiJZpWRDREREskrJhoiIiGSVkg0RERHJKiUbIiIiklVKNkRERCSrlGyIiIhIVinZEBGJmNmYaCGqAXHHIlJMlGyISMbUXkmygW1A3HGKSG61jjsAESlKlzdwbEaughCR/KBkQ0Qyzt3HxB2DiOQP3UYRkdjUHiNhZieY2RtmtsTMZpvZ7WbWPcn7tjSzv5jZ52b2vZl9Eb3eMkn9VmZ2mpk9b2YLo3N8bGa3NfCeo83sFTOrNrP5Zna3mfXI5OcXaSnUsyEi+eA84ADCEtaPAXsBJwEDzPh1tgMAAAMlSURBVGx3d59TU9HMdgX+C3QGHgTeBbYGjgMOM7Mfuvurteq3BR4G9gdmAX8FvgXKgSMIS6J/VCeenxOWOn+QsNrl7sAQYEcz28ndl2Xyw4sUOyUbIpJxZjYmyaGl7n5lgvKDgN3d/Y1abfwROBe4EjglKjPgL8A6wHHuPqlW/SHA3cBdZraNu6+KDo0hJBoPAcfUThTMrF3UVl2DgF3d/e1adf8K/Bg4DPhH0g8vIvVoiXkRyRgza+wPykJ3X69W/THAZcDt7n5KnbbWBar4/3bunjWKKArA8HsIRK0SCxEFO0FtxNgIGokBCxvBQrAyaf0BCiIBBbEUO7G0C9qoIFgIwa8ggkiw0aQKiPEDBYMWKsqxmFkYht3ElR2y4PvActgzd+/c6Q6z91xYBwxn5o+IOEDxJuJpZu5vc//HFG9FxjLzUUQMAJ+BQWB7Zi6tsv7Wei5l5lTt2jgwA1zOzNOrPKekCvdsSOq5zIwOn+EOP3nYZo5lYA5YD+wq03vLONNhnlZ+pIw7gSHg5WqFRs3zNrk3ZdzYxTySsNiQ1B8+dMi/L+NQLb7rML6VH67Ft12u50ub3K8yDnQ5l/Tfs9iQ1A82d8i3ulGWa7FtlwqwpTauVTTYRSKtIYsNSf1grJ4o92zsAb4Dr8p0awPpoQ7zjJfxRRlfUxQcuyNia09WKqlrFhuS+sHJiBip5S5Q/G0yXekgmQXmgdGIOF4dXH4/CCxQbCIlM38DV4ENwLWy+6T6m8GI2NTjZ5FUY+urpJ5bofUV4HZmztVy94DZiLhJse9itPwsAmdbgzIzI2ISuA/ciIg7FG8vdgDHgK/ARKXtFYqj0/cBR4GFiLhbjttGcbbHGeD6Pz2opL9isSGpCedXuLZI0WVSdQW4RXGuxgngG0UBcC4zP1YHZuaz8mCvKeAwRRHxCZgGLmbmfG38z4g4ApwCJoBJIICl8p5Pun88Sd3wnA1Ja6ZyrsV4Zj5Y29VIaop7NiRJUqMsNiRJUqMsNiRJUqPcsyFJkhrlmw1JktQoiw1JktQoiw1JktQoiw1JktQoiw1JktQoiw1JktSoP4Bl+X1jnWc9AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 576x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "plt.rcParams['font.size'] = 20\n",
    "\n",
    "plt.figure(figsize=(8, 6))\n",
    "plt.plot(model.history.history['loss'], 'o-r', label='Train Loss')\n",
    "plt.xlabel('Epoch')\n",
    "plt.ylabel(\"Loss (a.u.)\")\n",
    "plt.legend(frameon=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}