CrossTx_demo.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# CrossTx\n",
    "\n",
    "**Panagiotis Chrysinas, Changyou Chen, Rudiyanto Gunawan**\n",
    "**University at Buffalo-SUNY, Buffalo, NY 14260**\n",
    "\n",
    "This script is an implementation of CrossTx, a tool for predicting drug transcriptional signatures in an unseen target cell line using reference drug signature data from CMap and background transcriptome data of the target cell line. CrossTx comprises two components: Predictor and Corrector. The Predictor employs averaging or linear regression model to generate cell line-agnostic drug signatures using the reference data. The Corrector generates target cell-specific drug signature by correcting / projecting the output of the Predictor onto the latent transcriptomic space of the target cell line. For the Corrector, the projection is implemented either using the Principal Component Analysis (PCA) projection or an Autoencoder.\n",
    "\n"]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "# import modules\n",
    "from crosstx_functions_wind import *\n",
    "%matplotlib inline\n",
    "plt.style.use('seaborn-notebook')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load Data/Metadata\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Loading CMap dataset    \n",
    "with open('data_per_cell_table_full.pkl','rb') as f:\n",
    "             full_deepce_data=pkl.load(f)\n",
    "                         \n",
    "# Load the list of drugs whose transcriptome signatures are to be generated\n",
    "# These drugs are the top 100 in terms of the number of samples in the dataset.\n",
    "with open('top100_drugs_list.pkl','rb') as f:\n",
    "             top100_drugs_list=pkl.load(f)             \n",
    "        \n",
    "# Load ground truth drug transcriptomic signatures for the top 100 drugs\n",
    "with open('data_mat_nan.pkl','rb') as f:\n",
    "             data_mat_nan=pkl.load(f)\n",
    "             \n",
    "# Load drugload indices for each of the top 100 drugs in each target cell line\n",
    "with open('drugload_indices.pkl','rb') as f:\n",
    "             drugload_indices=pkl.load(f)\n",
    "             \n",
    "# Load CMap data (grouped by drugs)\n",
    "with open('data_mat_final.pkl','rb') as f:\n",
    "             data_mat_final=pkl.load(f)\n",
    "        \n",
    "# Load metadata on drug concentration values for each cell line for the 100 drugs    \n",
    "with open('conc_final.pkl','rb') as f:\n",
    "             conc_final=pkl.load(f)\n",
    "        \n",
    "# Load metadata on unique (union) drug concentration values for the 100 drugs \n",
    "with open('union_conc_val.pkl','rb') as f:\n",
    "             union_conc_val=pkl.load(f)\n",
    "        \n",
    "# Load reference data (i.e., CMap dataset minus the target cell line)\n",
    "with open('data_mat_final_allexk.pkl','rb') as f:\n",
    "             data_mat_final_allexk=pkl.load(f)\n",
    "        \n",
    "# Load concentration values (conc_final) for all the cell lines minus the target cell line\n",
    "with open('conc_final_allexk.pkl','rb') as f:\n",
    "             conc_final_allexk=pkl.load(f)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Create and train the neural network model (Autoencoder)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From C:\\Users\\rudig\\anaconda3\\lib\\site-packages\\keras\\layers\\normalization.py:524: _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",
      "Model: \"encoder\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "encoder_input (InputLayer)   [(None, 978)]             0         \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 150)               146850    \n",
      "_________________________________________________________________\n",
      "batch_normalization (BatchNo (None, 150)               600       \n",
      "_________________________________________________________________\n",
      "activation (Activation)      (None, 150)               0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 100)               15100     \n",
      "_________________________________________________________________\n",
      "batch_normalization_1 (Batch (None, 100)               400       \n",
      "_________________________________________________________________\n",
      "activation_1 (Activation)    (None, 100)               0         \n",
      "=================================================================\n",
      "Total params: 162,950\n",
      "Trainable params: 162,450\n",
      "Non-trainable params: 500\n",
      "_________________________________________________________________\n",
      "Model: \"decoder\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "decoder_input (InputLayer)   [(None, 100)]             0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 150)               15150     \n",
      "_________________________________________________________________\n",
      "batch_normalization_2 (Batch (None, 150)               600       \n",
      "_________________________________________________________________\n",
      "activation_2 (Activation)    (None, 150)               0         \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 978)               147678    \n",
      "_________________________________________________________________\n",
      "batch_normalization_3 (Batch (None, 978)               3912      \n",
      "_________________________________________________________________\n",
      "activation_3 (Activation)    (None, 978)               0         \n",
      "=================================================================\n",
      "Total params: 167,340\n",
      "Trainable params: 165,084\n",
      "Non-trainable params: 2,256\n",
      "_________________________________________________________________\n",
      "Model: \"autoencoder\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "encoder_input (InputLayer)   [(None, 978)]             0         \n",
      "_________________________________________________________________\n",
      "encoder (Functional)         (None, 100)               162950    \n",
      "_________________________________________________________________\n",
      "decoder (Functional)         (None, 978)               167340    \n",
      "_________________________________________________________________\n",
      "tf_op_layer_mul (TensorFlowO [(None, 978)]             0         \n",
      "=================================================================\n",
      "Total params: 330,290\n",
      "Trainable params: 327,534\n",
      "Non-trainable params: 2,756\n",
      "_________________________________________________________________\n",
      "Train on 900 samples, validate on 100 samples\n",
      "Epoch 1/300\n",
      "900/900 [==============================] - 1s 793us/sample - loss: 36.9699 - mse: 36.9699 - R2: -8.9414 - val_loss: 8.0793 - val_mse: 8.0793 - val_R2: -1.3721\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\rudig\\anaconda3\\lib\\site-packages\\keras\\engine\\training.py:2401: UserWarning: `Model.state_updates` will be removed in a future version. This property should not be used in TensorFlow 2.0, as `updates` are applied automatically.\n",
      "  warnings.warn('`Model.state_updates` will be removed in a future version. '\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 2/300\n",
      "900/900 [==============================] - 0s 181us/sample - loss: 26.5969 - mse: 26.5969 - R2: -6.1326 - val_loss: 11.6921 - val_mse: 11.6921 - val_R2: -2.1873\n",
      "Epoch 3/300\n",
      "900/900 [==============================] - 0s 182us/sample - loss: 20.3348 - mse: 20.3348 - R2: -4.5140 - val_loss: 13.2854 - val_mse: 13.2854 - val_R2: -2.5945\n",
      "Epoch 4/300\n",
      "900/900 [==============================] - 0s 183us/sample - loss: 16.8625 - mse: 16.8625 - R2: -3.6100 - val_loss: 13.3518 - val_mse: 13.3518 - val_R2: -2.6763\n",
      "Epoch 5/300\n",
      "900/900 [==============================] - 0s 157us/sample - loss: 14.7995 - mse: 14.7995 - R2: -3.0291 - val_loss: 13.8465 - val_mse: 13.8465 - val_R2: -2.7903\n",
      "Epoch 6/300\n",
      "900/900 [==============================] - 0s 167us/sample - loss: 13.8866 - mse: 13.8866 - R2: -2.7596 - val_loss: 13.5347 - val_mse: 13.5347 - val_R2: -2.9672\n",
      "Epoch 7/300\n",
      "900/900 [==============================] - 0s 191us/sample - loss: 12.7610 - mse: 12.7610 - R2: -2.4670 - val_loss: 11.9874 - val_mse: 11.9874 - val_R2: -2.2182\n",
      "Epoch 8/300\n",
      "900/900 [==============================] - 0s 187us/sample - loss: 13.1473 - mse: 13.1473 - R2: -2.5741 - val_loss: 12.3384 - val_mse: 12.3384 - val_R2: -2.5310\n",
      "Epoch 9/300\n",
      "900/900 [==============================] - 0s 201us/sample - loss: 12.3035 - mse: 12.3035 - R2: -2.3057 - val_loss: 11.3587 - val_mse: 11.3587 - val_R2: -2.0028\n",
      "Epoch 10/300\n",
      "900/900 [==============================] - 0s 202us/sample - loss: 10.6136 - mse: 10.6136 - R2: -1.8957 - val_loss: 11.5296 - val_mse: 11.5296 - val_R2: -2.1684\n",
      "Epoch 11/300\n",
      "900/900 [==============================] - 0s 204us/sample - loss: 10.5331 - mse: 10.5331 - R2: -1.8743 - val_loss: 10.4414 - val_mse: 10.4414 - val_R2: -2.0750\n",
      "Epoch 12/300\n",
      "900/900 [==============================] - 0s 200us/sample - loss: 10.7056 - mse: 10.7056 - R2: -1.8781 - val_loss: 11.9387 - val_mse: 11.9388 - val_R2: -2.3540\n",
      "Epoch 13/300\n",
      "900/900 [==============================] - 0s 192us/sample - loss: 12.3433 - mse: 12.3433 - R2: -2.3710 - val_loss: 11.4884 - val_mse: 11.4884 - val_R2: -2.3256\n",
      "Epoch 14/300\n",
      "900/900 [==============================] - 0s 167us/sample - loss: 9.9835 - mse: 9.9835 - R2: -1.7361 - val_loss: 9.9446 - val_mse: 9.9446 - val_R2: -1.8219\n",
      "Epoch 15/300\n",
      "900/900 [==============================] - 0s 176us/sample - loss: 9.0825 - mse: 9.0825 - R2: -1.4549 - val_loss: 9.6562 - val_mse: 9.6562 - val_R2: -1.7532\n",
      "Epoch 16/300\n",
      "900/900 [==============================] - 0s 152us/sample - loss: 9.2715 - mse: 9.2715 - R2: -1.5174 - val_loss: 9.3903 - val_mse: 9.3903 - val_R2: -1.6322\n",
      "Epoch 17/300\n",
      "900/900 [==============================] - 0s 153us/sample - loss: 9.4563 - mse: 9.4563 - R2: -1.5604 - val_loss: 9.1021 - val_mse: 9.1021 - val_R2: -1.5189\n",
      "Epoch 18/300\n",
      "900/900 [==============================] - 0s 166us/sample - loss: 9.2225 - mse: 9.2225 - R2: -1.5362 - val_loss: 8.6869 - val_mse: 8.6869 - val_R2: -1.4019\n",
      "Epoch 19/300\n",
      "900/900 [==============================] - 0s 189us/sample - loss: 8.9567 - mse: 8.9567 - R2: -1.4428 - val_loss: 9.2639 - val_mse: 9.2639 - val_R2: -1.6042\n",
      "Epoch 20/300\n",
      "900/900 [==============================] - 0s 188us/sample - loss: 9.8011 - mse: 9.8011 - R2: -1.6850 - val_loss: 9.8852 - val_mse: 9.8852 - val_R2: -1.5540\n",
      "Epoch 21/300\n",
      "900/900 [==============================] - 0s 196us/sample - loss: 8.4195 - mse: 8.4195 - R2: -1.2982 - val_loss: 8.9576 - val_mse: 8.9576 - val_R2: -1.3581\n",
      "Epoch 22/300\n",
      "900/900 [==============================] - 0s 197us/sample - loss: 8.4248 - mse: 8.4248 - R2: -1.2883 - val_loss: 8.6187 - val_mse: 8.6187 - val_R2: -1.4442\n",
      "Epoch 23/300\n",
      "900/900 [==============================] - 0s 204us/sample - loss: 8.5913 - mse: 8.5913 - R2: -1.3579 - val_loss: 8.9262 - val_mse: 8.9262 - val_R2: -1.3465\n",
      "Epoch 24/300\n",
      "900/900 [==============================] - 0s 186us/sample - loss: 9.3725 - mse: 9.3725 - R2: -1.5575 - val_loss: 10.0199 - val_mse: 10.0199 - val_R2: -1.7735\n",
      "Epoch 25/300\n",
      "900/900 [==============================] - 0s 181us/sample - loss: 9.0813 - mse: 9.0813 - R2: -1.4468 - val_loss: 8.6485 - val_mse: 8.6485 - val_R2: -1.4806\n",
      "Epoch 26/300\n",
      "900/900 [==============================] - 0s 171us/sample - loss: 8.6418 - mse: 8.6418 - R2: -1.3607 - val_loss: 8.9527 - val_mse: 8.9527 - val_R2: -1.5448\n",
      "Epoch 27/300\n",
      "900/900 [==============================] - 0s 174us/sample - loss: 7.1414 - mse: 7.1414 - R2: -0.9779 - val_loss: 7.8700 - val_mse: 7.8700 - val_R2: -1.2395\n",
      "Epoch 28/300\n",
      "900/900 [==============================] - 0s 220us/sample - loss: 7.6533 - mse: 7.6533 - R2: -1.1198 - val_loss: 7.6109 - val_mse: 7.6109 - val_R2: -1.2494\n",
      "Epoch 29/300\n",
      "900/900 [==============================] - 0s 225us/sample - loss: 7.8834 - mse: 7.8834 - R2: -1.1678 - val_loss: 7.6765 - val_mse: 7.6765 - val_R2: -1.1986\n",
      "Epoch 30/300\n",
      "900/900 [==============================] - 0s 178us/sample - loss: 7.7780 - mse: 7.7780 - R2: -1.1089 - val_loss: 7.4964 - val_mse: 7.4964 - val_R2: -1.0715\n",
      "Epoch 31/300\n",
      "900/900 [==============================] - 0s 176us/sample - loss: 7.0218 - mse: 7.0218 - R2: -0.9102 - val_loss: 7.2202 - val_mse: 7.2202 - val_R2: -1.0427\n",
      "Epoch 32/300\n",
      "900/900 [==============================] - 0s 161us/sample - loss: 7.4000 - mse: 7.4000 - R2: -1.0107 - val_loss: 7.0741 - val_mse: 7.0741 - val_R2: -0.9051\n",
      "Epoch 33/300\n",
      "900/900 [==============================] - 0s 188us/sample - loss: 7.1210 - mse: 7.1210 - R2: -0.9485 - val_loss: 6.9685 - val_mse: 6.9685 - val_R2: -0.8823\n",
      "Epoch 34/300\n",
      "900/900 [==============================] - 0s 189us/sample - loss: 7.1357 - mse: 7.1357 - R2: -0.9479 - val_loss: 7.0745 - val_mse: 7.0745 - val_R2: -0.9160\n",
      "Epoch 35/300\n",
      "900/900 [==============================] - 0s 192us/sample - loss: 6.9741 - mse: 6.9741 - R2: -0.9194 - val_loss: 6.6267 - val_mse: 6.6267 - val_R2: -0.9071\n",
      "Epoch 36/300\n",
      "900/900 [==============================] - 0s 211us/sample - loss: 6.7898 - mse: 6.7898 - R2: -0.8411 - val_loss: 6.6306 - val_mse: 6.6306 - val_R2: -0.8597\n",
      "Epoch 37/300\n",
      "900/900 [==============================] - 0s 217us/sample - loss: 6.7867 - mse: 6.7867 - R2: -0.8356 - val_loss: 6.6434 - val_mse: 6.6434 - val_R2: -0.8840\n",
      "Epoch 38/300\n",
      "900/900 [==============================] - 0s 200us/sample - loss: 6.3108 - mse: 6.3108 - R2: -0.7190 - val_loss: 6.3336 - val_mse: 6.3336 - val_R2: -0.8017\n",
      "Epoch 39/300\n",
      "900/900 [==============================] - 0s 169us/sample - loss: 6.1190 - mse: 6.1190 - R2: -0.6834 - val_loss: 6.0746 - val_mse: 6.0746 - val_R2: -0.7688\n",
      "Epoch 40/300\n",
      "900/900 [==============================] - 0s 172us/sample - loss: 6.1926 - mse: 6.1926 - R2: -0.6781 - val_loss: 5.8358 - val_mse: 5.8358 - val_R2: -0.6017\n",
      "Epoch 41/300\n",
      "900/900 [==============================] - 0s 178us/sample - loss: 6.0074 - mse: 6.0074 - R2: -0.6244 - val_loss: 5.4935 - val_mse: 5.4935 - val_R2: -0.4782\n",
      "Epoch 42/300\n",
      "900/900 [==============================] - 0s 171us/sample - loss: 5.4739 - mse: 5.4739 - R2: -0.4797 - val_loss: 5.2195 - val_mse: 5.2195 - val_R2: -0.5263\n",
      "Epoch 43/300\n",
      "900/900 [==============================] - 0s 173us/sample - loss: 5.4243 - mse: 5.4243 - R2: -0.4777 - val_loss: 5.1034 - val_mse: 5.1034 - val_R2: -0.5142\n",
      "Epoch 44/300\n",
      "900/900 [==============================] - 0s 178us/sample - loss: 5.3531 - mse: 5.3531 - R2: -0.4622 - val_loss: 4.7056 - val_mse: 4.7056 - val_R2: -0.3010\n",
      "Epoch 45/300\n",
      "900/900 [==============================] - 0s 168us/sample - loss: 4.8523 - mse: 4.8523 - R2: -0.3153 - val_loss: 4.3406 - val_mse: 4.3406 - val_R2: -0.1791\n",
      "Epoch 46/300\n",
      "900/900 [==============================] - 0s 167us/sample - loss: 4.5610 - mse: 4.5610 - R2: -0.2348 - val_loss: 4.0169 - val_mse: 4.0169 - val_R2: -0.2801\n",
      "Epoch 47/300\n",
      "900/900 [==============================] - 0s 177us/sample - loss: 4.2197 - mse: 4.2197 - R2: -0.1421 - val_loss: 3.7005 - val_mse: 3.7005 - val_R2: -0.0880\n",
      "Epoch 48/300\n",
      "900/900 [==============================] - 0s 179us/sample - loss: 4.2698 - mse: 4.2698 - R2: -0.1611 - val_loss: 3.3001 - val_mse: 3.3001 - val_R2: -0.0188\n",
      "Epoch 49/300\n",
      "900/900 [==============================] - 0s 174us/sample - loss: 3.7304 - mse: 3.7304 - R2: -0.0193 - val_loss: 3.0566 - val_mse: 3.0566 - val_R2: 0.1489\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 50/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 3.2971 - mse: 3.2971 - R2: 0.1112 - val_loss: 2.7228 - val_mse: 2.7228 - val_R2: 0.1822\n",
      "Epoch 51/300\n",
      "900/900 [==============================] - 0s 173us/sample - loss: 3.0154 - mse: 3.0154 - R2: 0.1829 - val_loss: 2.4304 - val_mse: 2.4304 - val_R2: 0.3042\n",
      "Epoch 52/300\n",
      "900/900 [==============================] - 0s 181us/sample - loss: 2.6521 - mse: 2.6521 - R2: 0.2788 - val_loss: 2.2000 - val_mse: 2.2000 - val_R2: 0.3465\n",
      "Epoch 53/300\n",
      "900/900 [==============================] - 0s 167us/sample - loss: 2.3895 - mse: 2.3895 - R2: 0.3544 - val_loss: 1.9644 - val_mse: 1.9644 - val_R2: 0.4455\n",
      "Epoch 54/300\n",
      "900/900 [==============================] - 0s 163us/sample - loss: 2.0840 - mse: 2.0840 - R2: 0.4375 - val_loss: 1.7091 - val_mse: 1.7091 - val_R2: 0.5144\n",
      "Epoch 55/300\n",
      "900/900 [==============================] - 0s 177us/sample - loss: 1.8362 - mse: 1.8362 - R2: 0.5053 - val_loss: 1.5352 - val_mse: 1.5352 - val_R2: 0.5627\n",
      "Epoch 56/300\n",
      "900/900 [==============================] - 0s 186us/sample - loss: 1.6392 - mse: 1.6392 - R2: 0.5578 - val_loss: 1.3873 - val_mse: 1.3873 - val_R2: 0.6187\n",
      "Epoch 57/300\n",
      "900/900 [==============================] - 0s 176us/sample - loss: 1.4558 - mse: 1.4558 - R2: 0.6079 - val_loss: 1.2666 - val_mse: 1.2666 - val_R2: 0.6493\n",
      "Epoch 58/300\n",
      "900/900 [==============================] - 0s 207us/sample - loss: 1.2989 - mse: 1.2989 - R2: 0.6501 - val_loss: 1.1789 - val_mse: 1.1789 - val_R2: 0.6551\n",
      "Epoch 59/300\n",
      "900/900 [==============================] - 0s 171us/sample - loss: 1.1948 - mse: 1.1948 - R2: 0.6775 - val_loss: 1.1175 - val_mse: 1.1175 - val_R2: 0.6900\n",
      "Epoch 60/300\n",
      "900/900 [==============================] - 0s 169us/sample - loss: 1.1080 - mse: 1.1080 - R2: 0.7005 - val_loss: 1.0744 - val_mse: 1.0744 - val_R2: 0.7032\n",
      "Epoch 61/300\n",
      "900/900 [==============================] - 0s 169us/sample - loss: 1.0162 - mse: 1.0162 - R2: 0.7226 - val_loss: 1.0486 - val_mse: 1.0486 - val_R2: 0.7020\n",
      "Epoch 62/300\n",
      "900/900 [==============================] - 0s 179us/sample - loss: 0.9649 - mse: 0.9649 - R2: 0.7391 - val_loss: 1.0205 - val_mse: 1.0205 - val_R2: 0.6959\n",
      "Epoch 63/300\n",
      "900/900 [==============================] - 0s 183us/sample - loss: 0.9302 - mse: 0.9302 - R2: 0.7468 - val_loss: 1.0049 - val_mse: 1.0049 - val_R2: 0.7124\n",
      "Epoch 64/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 0.9131 - mse: 0.9131 - R2: 0.7539 - val_loss: 0.9797 - val_mse: 0.9797 - val_R2: 0.7398\n",
      "Epoch 65/300\n",
      "900/900 [==============================] - 0s 173us/sample - loss: 0.8722 - mse: 0.8722 - R2: 0.7637 - val_loss: 0.9649 - val_mse: 0.9649 - val_R2: 0.7178\n",
      "Epoch 66/300\n",
      "900/900 [==============================] - 0s 179us/sample - loss: 0.8710 - mse: 0.8710 - R2: 0.7654 - val_loss: 0.9442 - val_mse: 0.9442 - val_R2: 0.7443\n",
      "Epoch 67/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 0.8571 - mse: 0.8571 - R2: 0.7686 - val_loss: 0.9348 - val_mse: 0.9348 - val_R2: 0.7447\n",
      "Epoch 68/300\n",
      "900/900 [==============================] - 0s 170us/sample - loss: 0.8344 - mse: 0.8344 - R2: 0.7739 - val_loss: 0.9215 - val_mse: 0.9215 - val_R2: 0.7474\n",
      "Epoch 69/300\n",
      "900/900 [==============================] - 0s 171us/sample - loss: 0.8271 - mse: 0.8271 - R2: 0.7765 - val_loss: 0.9083 - val_mse: 0.9083 - val_R2: 0.7478\n",
      "Epoch 70/300\n",
      "900/900 [==============================] - 0s 181us/sample - loss: 0.8061 - mse: 0.8061 - R2: 0.7831 - val_loss: 0.8946 - val_mse: 0.8946 - val_R2: 0.7420\n",
      "Epoch 71/300\n",
      "900/900 [==============================] - 0s 185us/sample - loss: 0.8260 - mse: 0.8260 - R2: 0.7790 - val_loss: 0.8858 - val_mse: 0.8858 - val_R2: 0.7599\n",
      "Epoch 72/300\n",
      "900/900 [==============================] - 0s 207us/sample - loss: 0.8022 - mse: 0.8022 - R2: 0.7841 - val_loss: 0.8749 - val_mse: 0.8749 - val_R2: 0.7521\n",
      "Epoch 73/300\n",
      "900/900 [==============================] - 0s 170us/sample - loss: 0.8013 - mse: 0.8013 - R2: 0.7825 - val_loss: 0.8699 - val_mse: 0.8699 - val_R2: 0.7628\n",
      "Epoch 74/300\n",
      "900/900 [==============================] - 0s 169us/sample - loss: 0.7928 - mse: 0.7928 - R2: 0.7860 - val_loss: 0.8605 - val_mse: 0.8605 - val_R2: 0.7587\n",
      "Epoch 75/300\n",
      "900/900 [==============================] - 0s 168us/sample - loss: 0.7953 - mse: 0.7953 - R2: 0.7860 - val_loss: 0.8531 - val_mse: 0.8531 - val_R2: 0.7574\n",
      "Epoch 76/300\n",
      "900/900 [==============================] - 0s 174us/sample - loss: 0.7897 - mse: 0.7897 - R2: 0.7870 - val_loss: 0.8475 - val_mse: 0.8475 - val_R2: 0.7594\n",
      "Epoch 77/300\n",
      "900/900 [==============================] - 0s 178us/sample - loss: 0.7663 - mse: 0.7663 - R2: 0.7928 - val_loss: 0.8406 - val_mse: 0.8406 - val_R2: 0.7657\n",
      "Epoch 78/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 0.7756 - mse: 0.7756 - R2: 0.7907 - val_loss: 0.8350 - val_mse: 0.8350 - val_R2: 0.7686\n",
      "Epoch 79/300\n",
      "900/900 [==============================] - 0s 175us/sample - loss: 0.7632 - mse: 0.7632 - R2: 0.7946 - val_loss: 0.8297 - val_mse: 0.8297 - val_R2: 0.7522\n",
      "Epoch 80/300\n",
      "900/900 [==============================] - 0s 163us/sample - loss: 0.7584 - mse: 0.7584 - R2: 0.7955 - val_loss: 0.8254 - val_mse: 0.8254 - val_R2: 0.7552\n",
      "Epoch 81/300\n",
      "900/900 [==============================] - 0s 170us/sample - loss: 0.7608 - mse: 0.7608 - R2: 0.7948 - val_loss: 0.8196 - val_mse: 0.8196 - val_R2: 0.7546\n",
      "Epoch 82/300\n",
      "900/900 [==============================] - 0s 167us/sample - loss: 0.7661 - mse: 0.7661 - R2: 0.7927 - val_loss: 0.8161 - val_mse: 0.8161 - val_R2: 0.7707\n",
      "Epoch 83/300\n",
      "900/900 [==============================] - 0s 167us/sample - loss: 0.7412 - mse: 0.7412 - R2: 0.8000 - val_loss: 0.8134 - val_mse: 0.8134 - val_R2: 0.7705\n",
      "Epoch 84/300\n",
      "900/900 [==============================] - 0s 162us/sample - loss: 0.7396 - mse: 0.7396 - R2: 0.8003 - val_loss: 0.8065 - val_mse: 0.8065 - val_R2: 0.7768\n",
      "Epoch 85/300\n",
      "900/900 [==============================] - 0s 163us/sample - loss: 0.7386 - mse: 0.7386 - R2: 0.7999 - val_loss: 0.8057 - val_mse: 0.8057 - val_R2: 0.7688\n",
      "Epoch 86/300\n",
      "900/900 [==============================] - 0s 173us/sample - loss: 0.7318 - mse: 0.7318 - R2: 0.8030 - val_loss: 0.7996 - val_mse: 0.7996 - val_R2: 0.7758\n",
      "Epoch 87/300\n",
      "900/900 [==============================] - 0s 166us/sample - loss: 0.7393 - mse: 0.7393 - R2: 0.8006 - val_loss: 0.7937 - val_mse: 0.7937 - val_R2: 0.7779\n",
      "Epoch 88/300\n",
      "900/900 [==============================] - 0s 154us/sample - loss: 0.7331 - mse: 0.7331 - R2: 0.8013 - val_loss: 0.7900 - val_mse: 0.7900 - val_R2: 0.7791\n",
      "Epoch 89/300\n",
      "900/900 [==============================] - 0s 158us/sample - loss: 0.7213 - mse: 0.7213 - R2: 0.8056 - val_loss: 0.7859 - val_mse: 0.7859 - val_R2: 0.7846\n",
      "Epoch 90/300\n",
      "900/900 [==============================] - 0s 158us/sample - loss: 0.7205 - mse: 0.7205 - R2: 0.8040 - val_loss: 0.7836 - val_mse: 0.7836 - val_R2: 0.7773\n",
      "Epoch 91/300\n",
      "900/900 [==============================] - 0s 170us/sample - loss: 0.7215 - mse: 0.7215 - R2: 0.8052 - val_loss: 0.7805 - val_mse: 0.7805 - val_R2: 0.7812\n",
      "Epoch 92/300\n",
      "900/900 [==============================] - 0s 178us/sample - loss: 0.7224 - mse: 0.7224 - R2: 0.8046 - val_loss: 0.7758 - val_mse: 0.7758 - val_R2: 0.7826\n",
      "Epoch 93/300\n",
      "900/900 [==============================] - 0s 171us/sample - loss: 0.6924 - mse: 0.6924 - R2: 0.8143 - val_loss: 0.7717 - val_mse: 0.7717 - val_R2: 0.7685\n",
      "Epoch 94/300\n",
      "900/900 [==============================] - 0s 159us/sample - loss: 0.7136 - mse: 0.7136 - R2: 0.8076 - val_loss: 0.7696 - val_mse: 0.7696 - val_R2: 0.7814\n",
      "Epoch 95/300\n",
      "900/900 [==============================] - 0s 170us/sample - loss: 0.7063 - mse: 0.7063 - R2: 0.8097 - val_loss: 0.7660 - val_mse: 0.7660 - val_R2: 0.7748\n",
      "Epoch 96/300\n",
      "900/900 [==============================] - 0s 168us/sample - loss: 0.7047 - mse: 0.7047 - R2: 0.8101 - val_loss: 0.7637 - val_mse: 0.7637 - val_R2: 0.7935\n",
      "Epoch 97/300\n",
      "900/900 [==============================] - 0s 182us/sample - loss: 0.6972 - mse: 0.6972 - R2: 0.8120 - val_loss: 0.7621 - val_mse: 0.7621 - val_R2: 0.7899\n",
      "Epoch 98/300\n",
      "900/900 [==============================] - 0s 190us/sample - loss: 0.6897 - mse: 0.6897 - R2: 0.8128 - val_loss: 0.7600 - val_mse: 0.7600 - val_R2: 0.7833\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 99/300\n",
      "900/900 [==============================] - 0s 186us/sample - loss: 0.7006 - mse: 0.7006 - R2: 0.8117 - val_loss: 0.7544 - val_mse: 0.7544 - val_R2: 0.7884\n",
      "Epoch 100/300\n",
      "900/900 [==============================] - 0s 209us/sample - loss: 0.7064 - mse: 0.7064 - R2: 0.8090 - val_loss: 0.7518 - val_mse: 0.7518 - val_R2: 0.7881\n",
      "Epoch 101/300\n",
      "900/900 [==============================] - 0s 184us/sample - loss: 0.6836 - mse: 0.6836 - R2: 0.8147 - val_loss: 0.7517 - val_mse: 0.7517 - val_R2: 0.7858\n",
      "Epoch 102/300\n",
      "900/900 [==============================] - 0s 190us/sample - loss: 0.6975 - mse: 0.6975 - R2: 0.8101 - val_loss: 0.7460 - val_mse: 0.7460 - val_R2: 0.7955\n",
      "Epoch 103/300\n",
      "900/900 [==============================] - 0s 173us/sample - loss: 0.6978 - mse: 0.6978 - R2: 0.8122 - val_loss: 0.7445 - val_mse: 0.7445 - val_R2: 0.7888\n",
      "Epoch 104/300\n",
      "900/900 [==============================] - 0s 174us/sample - loss: 0.6757 - mse: 0.6757 - R2: 0.8183 - val_loss: 0.7435 - val_mse: 0.7435 - val_R2: 0.7993\n",
      "Epoch 105/300\n",
      "900/900 [==============================] - 0s 194us/sample - loss: 0.6816 - mse: 0.6816 - R2: 0.8161 - val_loss: 0.7415 - val_mse: 0.7415 - val_R2: 0.7958\n",
      "Epoch 106/300\n",
      "900/900 [==============================] - 0s 184us/sample - loss: 0.6555 - mse: 0.6555 - R2: 0.8230 - val_loss: 0.7377 - val_mse: 0.7377 - val_R2: 0.7967\n",
      "Epoch 107/300\n",
      "900/900 [==============================] - 0s 160us/sample - loss: 0.6550 - mse: 0.6550 - R2: 0.8229 - val_loss: 0.7322 - val_mse: 0.7322 - val_R2: 0.7965\n",
      "Epoch 108/300\n",
      "900/900 [==============================] - 0s 204us/sample - loss: 0.6602 - mse: 0.6602 - R2: 0.8222 - val_loss: 0.7299 - val_mse: 0.7299 - val_R2: 0.7938\n",
      "Epoch 109/300\n",
      "900/900 [==============================] - 0s 167us/sample - loss: 0.6589 - mse: 0.6589 - R2: 0.8229 - val_loss: 0.7271 - val_mse: 0.7271 - val_R2: 0.7897\n",
      "Epoch 110/300\n",
      "900/900 [==============================] - 0s 162us/sample - loss: 0.6632 - mse: 0.6632 - R2: 0.8216 - val_loss: 0.7257 - val_mse: 0.7257 - val_R2: 0.8047\n",
      "Epoch 111/300\n",
      "900/900 [==============================] - 0s 160us/sample - loss: 0.6645 - mse: 0.6645 - R2: 0.8212 - val_loss: 0.7245 - val_mse: 0.7245 - val_R2: 0.7983\n",
      "Epoch 112/300\n",
      "900/900 [==============================] - 0s 181us/sample - loss: 0.6672 - mse: 0.6672 - R2: 0.8197 - val_loss: 0.7207 - val_mse: 0.7207 - val_R2: 0.8016\n",
      "Epoch 113/300\n",
      "900/900 [==============================] - 0s 176us/sample - loss: 0.6417 - mse: 0.6417 - R2: 0.8255 - val_loss: 0.7209 - val_mse: 0.7209 - val_R2: 0.8028\n",
      "Epoch 114/300\n",
      "900/900 [==============================] - 0s 180us/sample - loss: 0.6659 - mse: 0.6659 - R2: 0.8201 - val_loss: 0.7175 - val_mse: 0.7175 - val_R2: 0.8045\n",
      "Epoch 115/300\n",
      "900/900 [==============================] - 0s 181us/sample - loss: 0.6514 - mse: 0.6514 - R2: 0.8236 - val_loss: 0.7142 - val_mse: 0.7142 - val_R2: 0.7916\n",
      "Epoch 116/300\n",
      "900/900 [==============================] - 0s 193us/sample - loss: 0.6483 - mse: 0.6483 - R2: 0.8255 - val_loss: 0.7100 - val_mse: 0.7100 - val_R2: 0.8066\n",
      "Epoch 117/300\n",
      "900/900 [==============================] - 0s 173us/sample - loss: 0.6347 - mse: 0.6347 - R2: 0.8280 - val_loss: 0.7115 - val_mse: 0.7115 - val_R2: 0.8085\n",
      "Epoch 118/300\n",
      "900/900 [==============================] - 0s 175us/sample - loss: 0.6417 - mse: 0.6417 - R2: 0.8260 - val_loss: 0.7077 - val_mse: 0.7077 - val_R2: 0.8063\n",
      "Epoch 119/300\n",
      "900/900 [==============================] - 0s 167us/sample - loss: 0.6385 - mse: 0.6385 - R2: 0.8279 - val_loss: 0.7045 - val_mse: 0.7045 - val_R2: 0.8055\n",
      "Epoch 120/300\n",
      "900/900 [==============================] - 0s 182us/sample - loss: 0.6396 - mse: 0.6396 - R2: 0.8269 - val_loss: 0.7025 - val_mse: 0.7025 - val_R2: 0.8041\n",
      "Epoch 121/300\n",
      "900/900 [==============================] - 0s 186us/sample - loss: 0.6380 - mse: 0.6380 - R2: 0.8271 - val_loss: 0.7004 - val_mse: 0.7004 - val_R2: 0.8064\n",
      "Epoch 122/300\n",
      "900/900 [==============================] - 0s 177us/sample - loss: 0.6333 - mse: 0.6333 - R2: 0.8281 - val_loss: 0.6987 - val_mse: 0.6987 - val_R2: 0.8041\n",
      "Epoch 123/300\n",
      "900/900 [==============================] - 0s 178us/sample - loss: 0.6240 - mse: 0.6240 - R2: 0.8311 - val_loss: 0.6978 - val_mse: 0.6978 - val_R2: 0.8058\n",
      "Epoch 124/300\n",
      "900/900 [==============================] - 0s 183us/sample - loss: 0.6116 - mse: 0.6116 - R2: 0.8355 - val_loss: 0.6932 - val_mse: 0.6932 - val_R2: 0.8051\n",
      "Epoch 125/300\n",
      "900/900 [==============================] - 0s 190us/sample - loss: 0.6196 - mse: 0.6196 - R2: 0.8333 - val_loss: 0.6897 - val_mse: 0.6897 - val_R2: 0.8027\n",
      "Epoch 126/300\n",
      "900/900 [==============================] - 0s 202us/sample - loss: 0.6117 - mse: 0.6117 - R2: 0.8352 - val_loss: 0.6868 - val_mse: 0.6868 - val_R2: 0.7980\n",
      "Epoch 127/300\n",
      "900/900 [==============================] - 0s 201us/sample - loss: 0.6186 - mse: 0.6186 - R2: 0.8333 - val_loss: 0.6861 - val_mse: 0.6861 - val_R2: 0.8056\n",
      "Epoch 128/300\n",
      "900/900 [==============================] - 0s 230us/sample - loss: 0.6235 - mse: 0.6235 - R2: 0.8325 - val_loss: 0.6841 - val_mse: 0.6841 - val_R2: 0.8076\n",
      "Epoch 129/300\n",
      "900/900 [==============================] - 0s 203us/sample - loss: 0.6036 - mse: 0.6036 - R2: 0.8372 - val_loss: 0.6818 - val_mse: 0.6818 - val_R2: 0.7960\n",
      "Epoch 130/300\n",
      "900/900 [==============================] - 0s 198us/sample - loss: 0.6176 - mse: 0.6176 - R2: 0.8328 - val_loss: 0.6797 - val_mse: 0.6797 - val_R2: 0.7982\n",
      "Epoch 131/300\n",
      "900/900 [==============================] - 0s 203us/sample - loss: 0.6048 - mse: 0.6048 - R2: 0.8370 - val_loss: 0.6774 - val_mse: 0.6774 - val_R2: 0.8074\n",
      "Epoch 132/300\n",
      "900/900 [==============================] - 0s 195us/sample - loss: 0.6044 - mse: 0.6044 - R2: 0.8378 - val_loss: 0.6751 - val_mse: 0.6751 - val_R2: 0.8132\n",
      "Epoch 133/300\n",
      "900/900 [==============================] - 0s 203us/sample - loss: 0.6213 - mse: 0.6213 - R2: 0.8318 - val_loss: 0.6756 - val_mse: 0.6756 - val_R2: 0.8177\n",
      "Epoch 134/300\n",
      "900/900 [==============================] - 0s 171us/sample - loss: 0.5939 - mse: 0.5939 - R2: 0.8394 - val_loss: 0.6712 - val_mse: 0.6712 - val_R2: 0.8088\n",
      "Epoch 135/300\n",
      "900/900 [==============================] - 0s 195us/sample - loss: 0.5978 - mse: 0.5978 - R2: 0.8391 - val_loss: 0.6687 - val_mse: 0.6687 - val_R2: 0.8106\n",
      "Epoch 136/300\n",
      "900/900 [==============================] - 0s 165us/sample - loss: 0.6018 - mse: 0.6018 - R2: 0.8384 - val_loss: 0.6703 - val_mse: 0.6703 - val_R2: 0.8136\n",
      "Epoch 137/300\n",
      "900/900 [==============================] - 0s 157us/sample - loss: 0.6003 - mse: 0.6003 - R2: 0.8372 - val_loss: 0.6648 - val_mse: 0.6648 - val_R2: 0.8243\n",
      "Epoch 138/300\n",
      "900/900 [==============================] - 0s 178us/sample - loss: 0.5912 - mse: 0.5912 - R2: 0.8384 - val_loss: 0.6623 - val_mse: 0.6623 - val_R2: 0.8247\n",
      "Epoch 139/300\n",
      "900/900 [==============================] - 0s 183us/sample - loss: 0.5947 - mse: 0.5947 - R2: 0.8393 - val_loss: 0.6622 - val_mse: 0.6622 - val_R2: 0.8157\n",
      "Epoch 140/300\n",
      "900/900 [==============================] - 0s 183us/sample - loss: 0.5888 - mse: 0.5888 - R2: 0.8407 - val_loss: 0.6589 - val_mse: 0.6589 - val_R2: 0.8219\n",
      "Epoch 141/300\n",
      "900/900 [==============================] - 0s 179us/sample - loss: 0.6014 - mse: 0.6014 - R2: 0.8379 - val_loss: 0.6575 - val_mse: 0.6575 - val_R2: 0.8167\n",
      "Epoch 142/300\n",
      "900/900 [==============================] - 0s 175us/sample - loss: 0.5826 - mse: 0.5826 - R2: 0.8407 - val_loss: 0.6547 - val_mse: 0.6547 - val_R2: 0.8099\n",
      "Epoch 143/300\n",
      "900/900 [==============================] - 0s 182us/sample - loss: 0.5861 - mse: 0.5861 - R2: 0.8390 - val_loss: 0.6562 - val_mse: 0.6562 - val_R2: 0.8181\n",
      "Epoch 144/300\n",
      "900/900 [==============================] - 0s 177us/sample - loss: 0.5954 - mse: 0.5954 - R2: 0.8379 - val_loss: 0.6525 - val_mse: 0.6525 - val_R2: 0.8190\n",
      "Epoch 145/300\n",
      "900/900 [==============================] - 0s 177us/sample - loss: 0.5721 - mse: 0.5721 - R2: 0.8452 - val_loss: 0.6518 - val_mse: 0.6518 - val_R2: 0.8160\n",
      "Epoch 146/300\n",
      "900/900 [==============================] - 0s 174us/sample - loss: 0.5742 - mse: 0.5742 - R2: 0.8449 - val_loss: 0.6484 - val_mse: 0.6484 - val_R2: 0.8238\n",
      "Epoch 147/300\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "900/900 [==============================] - 0s 182us/sample - loss: 0.5887 - mse: 0.5887 - R2: 0.8415 - val_loss: 0.6455 - val_mse: 0.6455 - val_R2: 0.8167\n",
      "Epoch 148/300\n",
      "900/900 [==============================] - 0s 183us/sample - loss: 0.5609 - mse: 0.5609 - R2: 0.8482 - val_loss: 0.6432 - val_mse: 0.6432 - val_R2: 0.8215\n",
      "Epoch 149/300\n",
      "900/900 [==============================] - 0s 185us/sample - loss: 0.5701 - mse: 0.5701 - R2: 0.8466 - val_loss: 0.6401 - val_mse: 0.6401 - val_R2: 0.8294\n",
      "Epoch 150/300\n",
      "900/900 [==============================] - 0s 183us/sample - loss: 0.5745 - mse: 0.5745 - R2: 0.8450 - val_loss: 0.6389 - val_mse: 0.6389 - val_R2: 0.8210\n",
      "Epoch 151/300\n",
      "900/900 [==============================] - 0s 184us/sample - loss: 0.5703 - mse: 0.5703 - R2: 0.8456 - val_loss: 0.6387 - val_mse: 0.6387 - val_R2: 0.8180\n",
      "Epoch 152/300\n",
      "900/900 [==============================] - 0s 178us/sample - loss: 0.5813 - mse: 0.5813 - R2: 0.8426 - val_loss: 0.6359 - val_mse: 0.6359 - val_R2: 0.8335\n",
      "Epoch 153/300\n",
      "900/900 [==============================] - 0s 177us/sample - loss: 0.5474 - mse: 0.5474 - R2: 0.8532 - val_loss: 0.6369 - val_mse: 0.6369 - val_R2: 0.8197\n",
      "Epoch 154/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 0.5653 - mse: 0.5653 - R2: 0.8473 - val_loss: 0.6319 - val_mse: 0.6319 - val_R2: 0.8237\n",
      "Epoch 155/300\n",
      "900/900 [==============================] - 0s 161us/sample - loss: 0.5578 - mse: 0.5578 - R2: 0.8485 - val_loss: 0.6312 - val_mse: 0.6312 - val_R2: 0.8197\n",
      "Epoch 156/300\n",
      "900/900 [==============================] - 0s 166us/sample - loss: 0.5642 - mse: 0.5642 - R2: 0.8476 - val_loss: 0.6307 - val_mse: 0.6307 - val_R2: 0.8197\n",
      "Epoch 157/300\n",
      "900/900 [==============================] - 0s 158us/sample - loss: 0.5454 - mse: 0.5454 - R2: 0.8525 - val_loss: 0.6260 - val_mse: 0.6260 - val_R2: 0.8245\n",
      "Epoch 158/300\n",
      "900/900 [==============================] - 0s 159us/sample - loss: 0.5625 - mse: 0.5625 - R2: 0.8485 - val_loss: 0.6254 - val_mse: 0.6254 - val_R2: 0.8209\n",
      "Epoch 159/300\n",
      "900/900 [==============================] - 0s 151us/sample - loss: 0.5470 - mse: 0.5470 - R2: 0.8529 - val_loss: 0.6243 - val_mse: 0.6243 - val_R2: 0.8270\n",
      "Epoch 160/300\n",
      "900/900 [==============================] - 0s 156us/sample - loss: 0.5469 - mse: 0.5469 - R2: 0.8527 - val_loss: 0.6227 - val_mse: 0.6227 - val_R2: 0.8256\n",
      "Epoch 161/300\n",
      "900/900 [==============================] - 0s 192us/sample - loss: 0.5695 - mse: 0.5695 - R2: 0.8460 - val_loss: 0.6228 - val_mse: 0.6228 - val_R2: 0.8303\n",
      "Epoch 162/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 0.5413 - mse: 0.5413 - R2: 0.8536 - val_loss: 0.6188 - val_mse: 0.6188 - val_R2: 0.8153\n",
      "Epoch 163/300\n",
      "900/900 [==============================] - 0s 201us/sample - loss: 0.5314 - mse: 0.5314 - R2: 0.8558 - val_loss: 0.6164 - val_mse: 0.6164 - val_R2: 0.8299\n",
      "Epoch 164/300\n",
      "900/900 [==============================] - 0s 173us/sample - loss: 0.5600 - mse: 0.5600 - R2: 0.8472 - val_loss: 0.6187 - val_mse: 0.6187 - val_R2: 0.8231\n",
      "Epoch 165/300\n",
      "900/900 [==============================] - 0s 176us/sample - loss: 0.5382 - mse: 0.5382 - R2: 0.8544 - val_loss: 0.6150 - val_mse: 0.6150 - val_R2: 0.8258\n",
      "Epoch 166/300\n",
      "900/900 [==============================] - 0s 204us/sample - loss: 0.5295 - mse: 0.5295 - R2: 0.8570 - val_loss: 0.6130 - val_mse: 0.6130 - val_R2: 0.8198\n",
      "Epoch 167/300\n",
      "900/900 [==============================] - 0s 169us/sample - loss: 0.5360 - mse: 0.5360 - R2: 0.8549 - val_loss: 0.6111 - val_mse: 0.6111 - val_R2: 0.8173\n",
      "Epoch 168/300\n",
      "900/900 [==============================] - 0s 178us/sample - loss: 0.5400 - mse: 0.5400 - R2: 0.8533 - val_loss: 0.6085 - val_mse: 0.6085 - val_R2: 0.8308\n",
      "Epoch 169/300\n",
      "900/900 [==============================] - 0s 178us/sample - loss: 0.5192 - mse: 0.5192 - R2: 0.8601 - val_loss: 0.6081 - val_mse: 0.6081 - val_R2: 0.8328\n",
      "Epoch 170/300\n",
      "900/900 [==============================] - 0s 179us/sample - loss: 0.5249 - mse: 0.5249 - R2: 0.8579 - val_loss: 0.6066 - val_mse: 0.6066 - val_R2: 0.8331\n",
      "Epoch 171/300\n",
      "900/900 [==============================] - 0s 189us/sample - loss: 0.5198 - mse: 0.5198 - R2: 0.8589 - val_loss: 0.6038 - val_mse: 0.6038 - val_R2: 0.8355\n",
      "Epoch 172/300\n",
      "900/900 [==============================] - 0s 189us/sample - loss: 0.5414 - mse: 0.5414 - R2: 0.8533 - val_loss: 0.6039 - val_mse: 0.6039 - val_R2: 0.8256\n",
      "Epoch 173/300\n",
      "900/900 [==============================] - 0s 186us/sample - loss: 0.5289 - mse: 0.5289 - R2: 0.8574 - val_loss: 0.6020 - val_mse: 0.6020 - val_R2: 0.8226\n",
      "Epoch 174/300\n",
      "900/900 [==============================] - 0s 178us/sample - loss: 0.5155 - mse: 0.5155 - R2: 0.8609 - val_loss: 0.6006 - val_mse: 0.6006 - val_R2: 0.8301\n",
      "Epoch 175/300\n",
      "900/900 [==============================] - 0s 171us/sample - loss: 0.5203 - mse: 0.5203 - R2: 0.8594 - val_loss: 0.5992 - val_mse: 0.5992 - val_R2: 0.8316\n",
      "Epoch 176/300\n",
      "900/900 [==============================] - 0s 196us/sample - loss: 0.5168 - mse: 0.5168 - R2: 0.8599 - val_loss: 0.5984 - val_mse: 0.5984 - val_R2: 0.8409\n",
      "Epoch 177/300\n",
      "900/900 [==============================] - 0s 160us/sample - loss: 0.5203 - mse: 0.5203 - R2: 0.8590 - val_loss: 0.5947 - val_mse: 0.5947 - val_R2: 0.8308\n",
      "Epoch 178/300\n",
      "900/900 [==============================] - 0s 151us/sample - loss: 0.5176 - mse: 0.5176 - R2: 0.8596 - val_loss: 0.5978 - val_mse: 0.5978 - val_R2: 0.8395\n",
      "Epoch 179/300\n",
      "900/900 [==============================] - 0s 150us/sample - loss: 0.5295 - mse: 0.5295 - R2: 0.8567 - val_loss: 0.5938 - val_mse: 0.5938 - val_R2: 0.8344\n",
      "Epoch 180/300\n",
      "900/900 [==============================] - 0s 172us/sample - loss: 0.5202 - mse: 0.5202 - R2: 0.8579 - val_loss: 0.5940 - val_mse: 0.5940 - val_R2: 0.8347\n",
      "Epoch 181/300\n",
      "900/900 [==============================] - 0s 166us/sample - loss: 0.5121 - mse: 0.5121 - R2: 0.8621 - val_loss: 0.5900 - val_mse: 0.5900 - val_R2: 0.8215\n",
      "Epoch 182/300\n",
      "900/900 [==============================] - 0s 161us/sample - loss: 0.5150 - mse: 0.5150 - R2: 0.8607 - val_loss: 0.5920 - val_mse: 0.5920 - val_R2: 0.8363\n",
      "Epoch 183/300\n",
      "900/900 [==============================] - 0s 159us/sample - loss: 0.5061 - mse: 0.5061 - R2: 0.8642 - val_loss: 0.5894 - val_mse: 0.5894 - val_R2: 0.8390\n",
      "Epoch 184/300\n",
      "900/900 [==============================] - 0s 169us/sample - loss: 0.5089 - mse: 0.5089 - R2: 0.8619 - val_loss: 0.5905 - val_mse: 0.5905 - val_R2: 0.8367\n",
      "Epoch 185/300\n",
      "900/900 [==============================] - 0s 173us/sample - loss: 0.4999 - mse: 0.4999 - R2: 0.8647 - val_loss: 0.5836 - val_mse: 0.5836 - val_R2: 0.8313\n",
      "Epoch 186/300\n",
      "900/900 [==============================] - 0s 162us/sample - loss: 0.5250 - mse: 0.5250 - R2: 0.8588 - val_loss: 0.5853 - val_mse: 0.5853 - val_R2: 0.8385\n",
      "Epoch 187/300\n",
      "900/900 [==============================] - 0s 166us/sample - loss: 0.5149 - mse: 0.5149 - R2: 0.8608 - val_loss: 0.5828 - val_mse: 0.5828 - val_R2: 0.8416\n",
      "Epoch 188/300\n",
      "900/900 [==============================] - 0s 160us/sample - loss: 0.5048 - mse: 0.5048 - R2: 0.8635 - val_loss: 0.5819 - val_mse: 0.5819 - val_R2: 0.8323\n",
      "Epoch 189/300\n",
      "900/900 [==============================] - 0s 167us/sample - loss: 0.5155 - mse: 0.5155 - R2: 0.8606 - val_loss: 0.5811 - val_mse: 0.5811 - val_R2: 0.8401\n",
      "Epoch 190/300\n",
      "900/900 [==============================] - 0s 168us/sample - loss: 0.5051 - mse: 0.5051 - R2: 0.8637 - val_loss: 0.5811 - val_mse: 0.5811 - val_R2: 0.8402\n",
      "Epoch 191/300\n",
      "900/900 [==============================] - 0s 162us/sample - loss: 0.5134 - mse: 0.5134 - R2: 0.8597 - val_loss: 0.5776 - val_mse: 0.5776 - val_R2: 0.8378\n",
      "Epoch 192/300\n",
      "900/900 [==============================] - 0s 179us/sample - loss: 0.5179 - mse: 0.5179 - R2: 0.8602 - val_loss: 0.5751 - val_mse: 0.5751 - val_R2: 0.8276\n",
      "Epoch 193/300\n",
      "900/900 [==============================] - 0s 174us/sample - loss: 0.5188 - mse: 0.5188 - R2: 0.8600 - val_loss: 0.5768 - val_mse: 0.5768 - val_R2: 0.8403\n",
      "Epoch 194/300\n",
      "900/900 [==============================] - 0s 156us/sample - loss: 0.4830 - mse: 0.4830 - R2: 0.8697 - val_loss: 0.5728 - val_mse: 0.5728 - val_R2: 0.8439\n",
      "Epoch 195/300\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "900/900 [==============================] - 0s 162us/sample - loss: 0.4917 - mse: 0.4917 - R2: 0.8672 - val_loss: 0.5734 - val_mse: 0.5734 - val_R2: 0.8446\n",
      "Epoch 196/300\n",
      "900/900 [==============================] - 0s 165us/sample - loss: 0.4999 - mse: 0.4999 - R2: 0.8650 - val_loss: 0.5735 - val_mse: 0.5735 - val_R2: 0.8436\n",
      "Epoch 197/300\n",
      "900/900 [==============================] - 0s 167us/sample - loss: 0.4981 - mse: 0.4981 - R2: 0.8649 - val_loss: 0.5746 - val_mse: 0.5746 - val_R2: 0.8341\n",
      "Epoch 198/300\n",
      "900/900 [==============================] - 0s 196us/sample - loss: 0.4872 - mse: 0.4872 - R2: 0.8673 - val_loss: 0.5702 - val_mse: 0.5702 - val_R2: 0.8410\n",
      "Epoch 199/300\n",
      "900/900 [==============================] - 0s 165us/sample - loss: 0.4870 - mse: 0.4870 - R2: 0.8684 - val_loss: 0.5677 - val_mse: 0.5677 - val_R2: 0.8411\n",
      "Epoch 200/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 0.4882 - mse: 0.4882 - R2: 0.8685 - val_loss: 0.5661 - val_mse: 0.5661 - val_R2: 0.8460\n",
      "Epoch 201/300\n",
      "900/900 [==============================] - 0s 181us/sample - loss: 0.4860 - mse: 0.4860 - R2: 0.8690 - val_loss: 0.5666 - val_mse: 0.5666 - val_R2: 0.8449\n",
      "Epoch 202/300\n",
      "900/900 [==============================] - 0s 158us/sample - loss: 0.4741 - mse: 0.4741 - R2: 0.8716 - val_loss: 0.5637 - val_mse: 0.5637 - val_R2: 0.8495\n",
      "Epoch 203/300\n",
      "900/900 [==============================] - 0s 163us/sample - loss: 0.4938 - mse: 0.4938 - R2: 0.8663 - val_loss: 0.5628 - val_mse: 0.5628 - val_R2: 0.8441\n",
      "Epoch 204/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 0.4981 - mse: 0.4981 - R2: 0.8639 - val_loss: 0.5640 - val_mse: 0.5640 - val_R2: 0.8395\n",
      "Epoch 205/300\n",
      "900/900 [==============================] - 0s 163us/sample - loss: 0.4872 - mse: 0.4872 - R2: 0.8668 - val_loss: 0.5631 - val_mse: 0.5631 - val_R2: 0.8435\n",
      "Epoch 206/300\n",
      "900/900 [==============================] - 0s 167us/sample - loss: 0.4932 - mse: 0.4932 - R2: 0.8672 - val_loss: 0.5646 - val_mse: 0.5646 - val_R2: 0.8408\n",
      "Epoch 207/300\n",
      "900/900 [==============================] - 0s 173us/sample - loss: 0.4813 - mse: 0.4813 - R2: 0.8670 - val_loss: 0.5588 - val_mse: 0.5588 - val_R2: 0.8422\n",
      "Epoch 208/300\n",
      "900/900 [==============================] - 0s 186us/sample - loss: 0.4977 - mse: 0.4977 - R2: 0.8659 - val_loss: 0.5560 - val_mse: 0.5560 - val_R2: 0.8471\n",
      "Epoch 209/300\n",
      "900/900 [==============================] - 0s 166us/sample - loss: 0.4681 - mse: 0.4681 - R2: 0.8735 - val_loss: 0.5546 - val_mse: 0.5546 - val_R2: 0.8476\n",
      "Epoch 210/300\n",
      "900/900 [==============================] - 0s 170us/sample - loss: 0.4656 - mse: 0.4656 - R2: 0.8749 - val_loss: 0.5535 - val_mse: 0.5535 - val_R2: 0.8523\n",
      "Epoch 211/300\n",
      "900/900 [==============================] - 0s 175us/sample - loss: 0.4899 - mse: 0.4899 - R2: 0.8673 - val_loss: 0.5522 - val_mse: 0.5522 - val_R2: 0.8485\n",
      "Epoch 212/300\n",
      "900/900 [==============================] - 0s 161us/sample - loss: 0.4945 - mse: 0.4945 - R2: 0.8646 - val_loss: 0.5528 - val_mse: 0.5528 - val_R2: 0.8448\n",
      "Epoch 213/300\n",
      "900/900 [==============================] - 0s 163us/sample - loss: 0.4826 - mse: 0.4826 - R2: 0.8700 - val_loss: 0.5494 - val_mse: 0.5494 - val_R2: 0.8446\n",
      "Epoch 214/300\n",
      "900/900 [==============================] - 0s 152us/sample - loss: 0.4622 - mse: 0.4622 - R2: 0.8750 - val_loss: 0.5488 - val_mse: 0.5488 - val_R2: 0.8440\n",
      "Epoch 215/300\n",
      "900/900 [==============================] - 0s 168us/sample - loss: 0.4733 - mse: 0.4733 - R2: 0.8705 - val_loss: 0.5474 - val_mse: 0.5474 - val_R2: 0.8478\n",
      "Epoch 216/300\n",
      "900/900 [==============================] - 0s 168us/sample - loss: 0.4809 - mse: 0.4809 - R2: 0.8707 - val_loss: 0.5463 - val_mse: 0.5463 - val_R2: 0.8483\n",
      "Epoch 217/300\n",
      "900/900 [==============================] - 0s 180us/sample - loss: 0.4554 - mse: 0.4554 - R2: 0.8771 - val_loss: 0.5455 - val_mse: 0.5455 - val_R2: 0.8436\n",
      "Epoch 218/300\n",
      "900/900 [==============================] - 0s 200us/sample - loss: 0.4698 - mse: 0.4698 - R2: 0.8727 - val_loss: 0.5470 - val_mse: 0.5470 - val_R2: 0.8463\n",
      "Epoch 219/300\n",
      "900/900 [==============================] - 0s 156us/sample - loss: 0.4635 - mse: 0.4635 - R2: 0.8752 - val_loss: 0.5450 - val_mse: 0.5450 - val_R2: 0.8440\n",
      "Epoch 220/300\n",
      "900/900 [==============================] - 0s 167us/sample - loss: 0.4540 - mse: 0.4540 - R2: 0.8774 - val_loss: 0.5438 - val_mse: 0.5438 - val_R2: 0.8462\n",
      "Epoch 221/300\n",
      "900/900 [==============================] - 0s 181us/sample - loss: 0.4709 - mse: 0.4709 - R2: 0.8721 - val_loss: 0.5451 - val_mse: 0.5451 - val_R2: 0.8435\n",
      "Epoch 222/300\n",
      "900/900 [==============================] - 0s 166us/sample - loss: 0.4540 - mse: 0.4540 - R2: 0.8774 - val_loss: 0.5425 - val_mse: 0.5425 - val_R2: 0.8458\n",
      "Epoch 223/300\n",
      "900/900 [==============================] - 0s 156us/sample - loss: 0.4701 - mse: 0.4701 - R2: 0.8715 - val_loss: 0.5412 - val_mse: 0.5412 - val_R2: 0.8486\n",
      "Epoch 224/300\n",
      "900/900 [==============================] - 0s 179us/sample - loss: 0.4596 - mse: 0.4596 - R2: 0.8751 - val_loss: 0.5414 - val_mse: 0.5414 - val_R2: 0.8529\n",
      "Epoch 225/300\n",
      "900/900 [==============================] - 0s 170us/sample - loss: 0.4853 - mse: 0.4853 - R2: 0.8690 - val_loss: 0.5399 - val_mse: 0.5399 - val_R2: 0.8534\n",
      "Epoch 226/300\n",
      "900/900 [==============================] - 0s 165us/sample - loss: 0.4654 - mse: 0.4654 - R2: 0.8740 - val_loss: 0.5385 - val_mse: 0.5385 - val_R2: 0.8466\n",
      "Epoch 227/300\n",
      "900/900 [==============================] - 0s 167us/sample - loss: 0.4548 - mse: 0.4548 - R2: 0.8766 - val_loss: 0.5377 - val_mse: 0.5377 - val_R2: 0.8466\n",
      "Epoch 228/300\n",
      "900/900 [==============================] - 0s 160us/sample - loss: 0.4492 - mse: 0.4492 - R2: 0.8784 - val_loss: 0.5364 - val_mse: 0.5364 - val_R2: 0.8509\n",
      "Epoch 229/300\n",
      "900/900 [==============================] - 0s 158us/sample - loss: 0.4458 - mse: 0.4458 - R2: 0.8794 - val_loss: 0.5322 - val_mse: 0.5322 - val_R2: 0.8455\n",
      "Epoch 230/300\n",
      "900/900 [==============================] - 0s 162us/sample - loss: 0.4644 - mse: 0.4644 - R2: 0.8739 - val_loss: 0.5338 - val_mse: 0.5338 - val_R2: 0.8552\n",
      "Epoch 231/300\n",
      "900/900 [==============================] - 0s 168us/sample - loss: 0.4543 - mse: 0.4543 - R2: 0.8770 - val_loss: 0.5369 - val_mse: 0.5369 - val_R2: 0.8513\n",
      "Epoch 232/300\n",
      "900/900 [==============================] - 0s 168us/sample - loss: 0.4535 - mse: 0.4535 - R2: 0.8766 - val_loss: 0.5320 - val_mse: 0.5320 - val_R2: 0.8458\n",
      "Epoch 233/300\n",
      "900/900 [==============================] - 0s 156us/sample - loss: 0.4537 - mse: 0.4537 - R2: 0.8772 - val_loss: 0.5324 - val_mse: 0.5324 - val_R2: 0.8550\n",
      "Epoch 234/300\n",
      "900/900 [==============================] - 0s 157us/sample - loss: 0.4654 - mse: 0.4654 - R2: 0.8730 - val_loss: 0.5294 - val_mse: 0.5294 - val_R2: 0.8474\n",
      "Epoch 235/300\n",
      "900/900 [==============================] - 0s 171us/sample - loss: 0.4512 - mse: 0.4512 - R2: 0.8780 - val_loss: 0.5329 - val_mse: 0.5329 - val_R2: 0.8476\n",
      "Epoch 236/300\n",
      "900/900 [==============================] - 0s 166us/sample - loss: 0.4508 - mse: 0.4508 - R2: 0.8784 - val_loss: 0.5296 - val_mse: 0.5296 - val_R2: 0.8482\n",
      "Epoch 237/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 0.4521 - mse: 0.4521 - R2: 0.8772 - val_loss: 0.5300 - val_mse: 0.5300 - val_R2: 0.8596\n",
      "Epoch 238/300\n",
      "900/900 [==============================] - 0s 163us/sample - loss: 0.4353 - mse: 0.4353 - R2: 0.8814 - val_loss: 0.5268 - val_mse: 0.5268 - val_R2: 0.8467\n",
      "Epoch 239/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 0.4533 - mse: 0.4533 - R2: 0.8764 - val_loss: 0.5250 - val_mse: 0.5250 - val_R2: 0.8552\n",
      "Epoch 240/300\n",
      "900/900 [==============================] - 0s 162us/sample - loss: 0.4318 - mse: 0.4318 - R2: 0.8832 - val_loss: 0.5286 - val_mse: 0.5286 - val_R2: 0.8489\n",
      "Epoch 241/300\n",
      "900/900 [==============================] - 0s 166us/sample - loss: 0.4340 - mse: 0.4340 - R2: 0.8826 - val_loss: 0.5244 - val_mse: 0.5244 - val_R2: 0.8527\n",
      "Epoch 242/300\n",
      "900/900 [==============================] - 0s 180us/sample - loss: 0.4266 - mse: 0.4266 - R2: 0.8843 - val_loss: 0.5235 - val_mse: 0.5235 - val_R2: 0.8515\n",
      "Epoch 243/300\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "900/900 [==============================] - 0s 171us/sample - loss: 0.4549 - mse: 0.4549 - R2: 0.8754 - val_loss: 0.5251 - val_mse: 0.5251 - val_R2: 0.8557\n",
      "Epoch 244/300\n",
      "900/900 [==============================] - 0s 171us/sample - loss: 0.4423 - mse: 0.4423 - R2: 0.8799 - val_loss: 0.5218 - val_mse: 0.5218 - val_R2: 0.8538\n",
      "Epoch 245/300\n",
      "900/900 [==============================] - 0s 167us/sample - loss: 0.4404 - mse: 0.4404 - R2: 0.8814 - val_loss: 0.5222 - val_mse: 0.5222 - val_R2: 0.8628\n",
      "Epoch 246/300\n",
      "900/900 [==============================] - 0s 229us/sample - loss: 0.4425 - mse: 0.4425 - R2: 0.8804 - val_loss: 0.5216 - val_mse: 0.5216 - val_R2: 0.8511\n",
      "Epoch 247/300\n",
      "900/900 [==============================] - 0s 162us/sample - loss: 0.4234 - mse: 0.4234 - R2: 0.8856 - val_loss: 0.5187 - val_mse: 0.5187 - val_R2: 0.8608\n",
      "Epoch 248/300\n",
      "900/900 [==============================] - 0s 169us/sample - loss: 0.4252 - mse: 0.4252 - R2: 0.8846 - val_loss: 0.5170 - val_mse: 0.5170 - val_R2: 0.8602\n",
      "Epoch 249/300\n",
      "900/900 [==============================] - 0s 177us/sample - loss: 0.4281 - mse: 0.4281 - R2: 0.8836 - val_loss: 0.5232 - val_mse: 0.5232 - val_R2: 0.8527\n",
      "Epoch 250/300\n",
      "900/900 [==============================] - 0s 162us/sample - loss: 0.4210 - mse: 0.4210 - R2: 0.8851 - val_loss: 0.5171 - val_mse: 0.5171 - val_R2: 0.8536\n",
      "Epoch 251/300\n",
      "900/900 [==============================] - 0s 167us/sample - loss: 0.4238 - mse: 0.4238 - R2: 0.8844 - val_loss: 0.5193 - val_mse: 0.5193 - val_R2: 0.8530\n",
      "Epoch 252/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 0.4227 - mse: 0.4227 - R2: 0.8840 - val_loss: 0.5150 - val_mse: 0.5150 - val_R2: 0.8598\n",
      "Epoch 253/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 0.4272 - mse: 0.4272 - R2: 0.8844 - val_loss: 0.5160 - val_mse: 0.5160 - val_R2: 0.8548\n",
      "Epoch 254/300\n",
      "900/900 [==============================] - 0s 165us/sample - loss: 0.4259 - mse: 0.4259 - R2: 0.8840 - val_loss: 0.5141 - val_mse: 0.5141 - val_R2: 0.8478\n",
      "Epoch 255/300\n",
      "900/900 [==============================] - 0s 160us/sample - loss: 0.4255 - mse: 0.4255 - R2: 0.8835 - val_loss: 0.5121 - val_mse: 0.5121 - val_R2: 0.8569\n",
      "Epoch 256/300\n",
      "900/900 [==============================] - 0s 161us/sample - loss: 0.4298 - mse: 0.4298 - R2: 0.8838 - val_loss: 0.5157 - val_mse: 0.5157 - val_R2: 0.8559\n",
      "Epoch 257/300\n",
      "900/900 [==============================] - 0s 178us/sample - loss: 0.4287 - mse: 0.4287 - R2: 0.8826 - val_loss: 0.5123 - val_mse: 0.5123 - val_R2: 0.8590\n",
      "Epoch 258/300\n",
      "900/900 [==============================] - 0s 155us/sample - loss: 0.4351 - mse: 0.4351 - R2: 0.8827 - val_loss: 0.5143 - val_mse: 0.5143 - val_R2: 0.8614\n",
      "Epoch 259/300\n",
      "900/900 [==============================] - 0s 165us/sample - loss: 0.4175 - mse: 0.4175 - R2: 0.8867 - val_loss: 0.5099 - val_mse: 0.5099 - val_R2: 0.8523\n",
      "Epoch 260/300\n",
      "900/900 [==============================] - ETA: 0s - loss: 0.4099 - mse: 0.4099 - R2: 0.88 - 0s 163us/sample - loss: 0.4094 - mse: 0.4094 - R2: 0.8896 - val_loss: 0.5135 - val_mse: 0.5135 - val_R2: 0.8500\n",
      "Epoch 261/300\n",
      "900/900 [==============================] - 0s 159us/sample - loss: 0.4126 - mse: 0.4126 - R2: 0.8882 - val_loss: 0.5129 - val_mse: 0.5129 - val_R2: 0.8594\n",
      "Epoch 262/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 0.4241 - mse: 0.4241 - R2: 0.8854 - val_loss: 0.5070 - val_mse: 0.5070 - val_R2: 0.8522\n",
      "Epoch 263/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 0.4147 - mse: 0.4147 - R2: 0.8883 - val_loss: 0.5097 - val_mse: 0.5097 - val_R2: 0.8605\n",
      "Epoch 264/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 0.4206 - mse: 0.4206 - R2: 0.8860 - val_loss: 0.5065 - val_mse: 0.5065 - val_R2: 0.8567\n",
      "Epoch 265/300\n",
      "900/900 [==============================] - 0s 171us/sample - loss: 0.4056 - mse: 0.4056 - R2: 0.8907 - val_loss: 0.5052 - val_mse: 0.5052 - val_R2: 0.8593\n",
      "Epoch 266/300\n",
      "900/900 [==============================] - 0s 198us/sample - loss: 0.4264 - mse: 0.4264 - R2: 0.8848 - val_loss: 0.5082 - val_mse: 0.5082 - val_R2: 0.8567\n",
      "Epoch 267/300\n",
      "900/900 [==============================] - 0s 157us/sample - loss: 0.4019 - mse: 0.4019 - R2: 0.8906 - val_loss: 0.5022 - val_mse: 0.5022 - val_R2: 0.8629\n",
      "Epoch 268/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 0.4103 - mse: 0.4103 - R2: 0.8897 - val_loss: 0.5058 - val_mse: 0.5058 - val_R2: 0.8575\n",
      "Epoch 269/300\n",
      "900/900 [==============================] - 0s 168us/sample - loss: 0.4074 - mse: 0.4074 - R2: 0.8898 - val_loss: 0.5054 - val_mse: 0.5054 - val_R2: 0.8550\n",
      "Epoch 270/300\n",
      "900/900 [==============================] - 0s 154us/sample - loss: 0.4139 - mse: 0.4139 - R2: 0.8874 - val_loss: 0.5074 - val_mse: 0.5074 - val_R2: 0.8607\n",
      "Epoch 271/300\n",
      "900/900 [==============================] - 0s 177us/sample - loss: 0.4132 - mse: 0.4132 - R2: 0.8879 - val_loss: 0.5003 - val_mse: 0.5003 - val_R2: 0.8596\n",
      "Epoch 272/300\n",
      "900/900 [==============================] - 0s 164us/sample - loss: 0.4102 - mse: 0.4102 - R2: 0.8890 - val_loss: 0.5030 - val_mse: 0.5030 - val_R2: 0.8610\n",
      "Epoch 273/300\n",
      "900/900 [==============================] - 0s 163us/sample - loss: 0.4126 - mse: 0.4126 - R2: 0.8871 - val_loss: 0.5024 - val_mse: 0.5024 - val_R2: 0.8616\n",
      "Epoch 274/300\n",
      "900/900 [==============================] - 0s 169us/sample - loss: 0.4039 - mse: 0.4039 - R2: 0.8901 - val_loss: 0.5017 - val_mse: 0.5017 - val_R2: 0.8513\n",
      "Epoch 275/300\n",
      "900/900 [==============================] - 0s 169us/sample - loss: 0.4189 - mse: 0.4189 - R2: 0.8856 - val_loss: 0.5008 - val_mse: 0.5008 - val_R2: 0.8595\n",
      "Epoch 276/300\n",
      "900/900 [==============================] - 0s 162us/sample - loss: 0.4149 - mse: 0.4149 - R2: 0.8875 - val_loss: 0.5014 - val_mse: 0.5014 - val_R2: 0.8591\n",
      "Epoch 277/300\n",
      "900/900 [==============================] - 0s 155us/sample - loss: 0.4044 - mse: 0.4044 - R2: 0.8904 - val_loss: 0.4972 - val_mse: 0.4972 - val_R2: 0.8609\n",
      "Epoch 278/300\n",
      "900/900 [==============================] - 0s 161us/sample - loss: 0.4084 - mse: 0.4084 - R2: 0.8894 - val_loss: 0.5002 - val_mse: 0.5002 - val_R2: 0.8606\n",
      "Epoch 279/300\n",
      "900/900 [==============================] - 0s 157us/sample - loss: 0.3943 - mse: 0.3943 - R2: 0.8933 - val_loss: 0.4942 - val_mse: 0.4942 - val_R2: 0.8574\n",
      "Epoch 280/300\n",
      "900/900 [==============================] - 0s 183us/sample - loss: 0.4005 - mse: 0.4005 - R2: 0.8913 - val_loss: 0.4968 - val_mse: 0.4968 - val_R2: 0.8661\n",
      "Epoch 281/300\n",
      "900/900 [==============================] - 0s 168us/sample - loss: 0.3965 - mse: 0.3965 - R2: 0.8926 - val_loss: 0.4938 - val_mse: 0.4938 - val_R2: 0.8614\n",
      "Epoch 282/300\n",
      "900/900 [==============================] - 0s 171us/sample - loss: 0.4150 - mse: 0.4150 - R2: 0.8879 - val_loss: 0.4969 - val_mse: 0.4969 - val_R2: 0.8575\n",
      "Epoch 283/300\n",
      "900/900 [==============================] - 0s 161us/sample - loss: 0.3964 - mse: 0.3964 - R2: 0.8929 - val_loss: 0.4917 - val_mse: 0.4917 - val_R2: 0.8637\n",
      "Epoch 284/300\n",
      "900/900 [==============================] - 0s 166us/sample - loss: 0.3911 - mse: 0.3911 - R2: 0.8942 - val_loss: 0.4926 - val_mse: 0.4926 - val_R2: 0.8639\n",
      "Epoch 285/300\n",
      "900/900 [==============================] - 0s 157us/sample - loss: 0.3781 - mse: 0.3781 - R2: 0.8978 - val_loss: 0.4911 - val_mse: 0.4911 - val_R2: 0.8616\n",
      "Epoch 286/300\n",
      "900/900 [==============================] - 0s 156us/sample - loss: 0.4007 - mse: 0.4007 - R2: 0.8905 - val_loss: 0.4907 - val_mse: 0.4907 - val_R2: 0.8602\n",
      "Epoch 287/300\n",
      "900/900 [==============================] - 0s 165us/sample - loss: 0.3860 - mse: 0.3860 - R2: 0.8951 - val_loss: 0.4934 - val_mse: 0.4934 - val_R2: 0.8611\n",
      "Epoch 288/300\n",
      "900/900 [==============================] - 0s 161us/sample - loss: 0.4127 - mse: 0.4127 - R2: 0.8882 - val_loss: 0.4920 - val_mse: 0.4920 - val_R2: 0.8624\n",
      "Epoch 289/300\n",
      "900/900 [==============================] - 0s 168us/sample - loss: 0.3981 - mse: 0.3981 - R2: 0.8912 - val_loss: 0.4935 - val_mse: 0.4935 - val_R2: 0.8565\n",
      "Epoch 290/300\n",
      "900/900 [==============================] - 0s 156us/sample - loss: 0.4035 - mse: 0.4035 - R2: 0.8910 - val_loss: 0.4894 - val_mse: 0.4894 - val_R2: 0.8610\n",
      "Epoch 291/300\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "900/900 [==============================] - 0s 169us/sample - loss: 0.3838 - mse: 0.3838 - R2: 0.8960 - val_loss: 0.4957 - val_mse: 0.4957 - val_R2: 0.8673\n",
      "Epoch 292/300\n",
      "900/900 [==============================] - 0s 192us/sample - loss: 0.4023 - mse: 0.4023 - R2: 0.8917 - val_loss: 0.4878 - val_mse: 0.4878 - val_R2: 0.8626\n",
      "Epoch 293/300\n",
      "900/900 [==============================] - 0s 171us/sample - loss: 0.3873 - mse: 0.3873 - R2: 0.8951 - val_loss: 0.4888 - val_mse: 0.4888 - val_R2: 0.8672\n",
      "Epoch 294/300\n",
      "900/900 [==============================] - 0s 181us/sample - loss: 0.3937 - mse: 0.3937 - R2: 0.8939 - val_loss: 0.4868 - val_mse: 0.4868 - val_R2: 0.8628\n",
      "Epoch 295/300\n",
      "900/900 [==============================] - 0s 186us/sample - loss: 0.3955 - mse: 0.3955 - R2: 0.8904 - val_loss: 0.4876 - val_mse: 0.4876 - val_R2: 0.8620\n",
      "Epoch 296/300\n",
      "900/900 [==============================] - 0s 170us/sample - loss: 0.3902 - mse: 0.3902 - R2: 0.8950 - val_loss: 0.4864 - val_mse: 0.4864 - val_R2: 0.8643\n",
      "Epoch 297/300\n",
      "900/900 [==============================] - 0s 209us/sample - loss: 0.3923 - mse: 0.3923 - R2: 0.8934 - val_loss: 0.4833 - val_mse: 0.4833 - val_R2: 0.8596\n",
      "Epoch 298/300\n",
      "900/900 [==============================] - 0s 162us/sample - loss: 0.3917 - mse: 0.3917 - R2: 0.8936 - val_loss: 0.4869 - val_mse: 0.4869 - val_R2: 0.8609\n",
      "Epoch 299/300\n",
      "900/900 [==============================] - 0s 172us/sample - loss: 0.3739 - mse: 0.3739 - R2: 0.8991 - val_loss: 0.4822 - val_mse: 0.4822 - val_R2: 0.8627\n",
      "Epoch 300/300\n",
      "900/900 [==============================] - 0s 187us/sample - loss: 0.3762 - mse: 0.3762 - R2: 0.8975 - val_loss: 0.4831 - val_mse: 0.4831 - val_R2: 0.8698\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x396 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAhMAAAF6CAYAAACjsNorAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuNCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8QVMy6AAAACXBIWXMAAAsTAAALEwEAmpwYAAA6aklEQVR4nO3deZxcVZ3//9enqrp637J2OjthCYmshiUIBkRAUGTc2HQQha+i4A6jo6gRHeeHA6LjBo64iyiM6IACEjBh38ImZIGQlWydpffu6trO749b3al0qrqr01V9O13v5+NRj0qde2/V6cOl693nnHuuOecQERER2V8BvysgIiIiBzaFCRERERkWhQkREREZFoUJERERGRaFCRERERkWhQkREREZFoUJEcmJmS02M2dmr2XZvia1fXG/8vea2UNm1mJmPWb2qpl9y8wmpO3jsjxOTm3/5QD7XFTQH1xEBhXyuwIickCJALPNbIFz7tneQjM7DpiZ2k5a+Y3AZ4FfADcBbcA84ApgPvCetN1vBO7s93mvpJ6/Cdzcb9sngIuBJfv/44hIPihMiMhQdALPARcCz6aVXwg8BLy5t8DMzgU+D1zmnPt52r7LzOynwJn93nu9c+7JTB/qnHsdeD29zMxuBZY453bs588iInmiYQ4RGarbgfPNzABSz+enytN9DniuX5AAwDmXcM7du78VMLMj8Xo4fr+/7yEi+aMwISJD9SdgMnBy6vUpwETgrt4dzKwEOAm4bwjvGzCzUNojOMC+F+INqfx5KBUXkcJQmBCRIXHOteCFhAtTRRcC96XKe40HSoGNQ3jr7wOxtMeyAfa9APibc65tCO8vIgWiMCEi++N24P1mVgq8n32HOHoN5U6C/wUcl/a4LNNOZnYCcBAa4hAZNTQBU0T2x/8BPwP+A6gE7u63fRfQA8wYwntuTL9CZAAXAu3AX4fw3iJSQOqZEJEhc851AvfgTbK8O/U6fXsMeAw4K5+fa2YB4APAn51z3fl8bxHZfwoTIrK/foLXI9F//Yde3wMWmNmH+28ws4CZvWM/PvOtwFQ0xCEyqmiYQ0T2i3NuKbB0gO13m9l3gVvN7C3AX4AOYC7eolXrGdrVHuANcewCHhh6jUWkUBQmRKRgnHNfMLPHgauA24ByvBDxf8ANQ3kvMwsB7wPudM7F81xVERkGc24ok63z/OFmBwPXACcCbwIecc6dmsNxtXhdqP+CN1RzD/Bp59yuQtVVREREMvO7Z2I+cA7wJBAewnF/AA4DLgeSwPV4i9eckuf6iYiIyCD87pkIOOeSqX/fCUwYrGfCzBYCjwOLnHMPp8qOB54CznDO6aY/IiIiI8jXqzl6g8QQnQ1s7w0Sqfd5GliX2iYiIiIj6EC8NHQusCpD+crUNhERERlBfs+Z2B/1QEuG8ma8JXYHZGaLga8DlJeXM2/evHzWTUREZFRbvnz5TufcxHy+54EYJiDzev+WpXzvA51bDCwGWLBggXv22VxW7xURERkbzGxDvt/zQBzmaAbqMpTXkbnHQkRERAroQAwTq8g8NyLbXAoREREpoAMxTNwLNJjZyb0FZrYAb77Evb7VSkREpEj5OmfCzCrwFq0C7+Y9NWb2/tTrvznnusxsDbDMOXcZgHPuCTO7H/i1mV3NnkWrHtUaEyIiIiPP7wmYk4A7+pX1vp6Nt4Z/CAj22+dC4Cbg56Qtp12wWoqIiEhWvoYJ59x6vKswBtpnVoayFuAjqYeIiBQJ5xyJRIJkcn/WPBy7AoEAwWAQswG/Ugv3+b58qoiIyBAlEglaW1uJRqN+V2XUiUajtLa2kkgkfPl8v4c5REREBuWco729ndraWt/++h7NwuEw5eXltLa2+tJG6pkQEZFRL5FIEA6HFSQGYGaEw2FfeicUJkREZNRLJpOEQupMH0wwGPRlPonChIiIyBihCZgiIiJyQFKYEBERkWFRmBARERlBixcvxsz6Hg0NDbzrXe/ipZde6ttn9erVXHnllRx++OFUVFRw0EEH8ZnPfIaWlhb/Kj4AzWYREREZYbW1tdx3330ArF+/nq997WucccYZrFy5knHjxvHAAw/w2GOP8YlPfIIjjzyStWvXcu211/LEE0/w5JNPEgiMrr4AhQkREZERFgqFOPHEEwE48cQTmTVrFgsXLuS+++7j4osv5qKLLuLKK6/sm1B56qmnMm3aNM466yweeeQRFi1a5Gf196EwISIi4rOjjjoKgE2bNgEwfvz4ffY55phjAGhqahq5iuVodPWTiIiIFKGNGzcCMHv27Kz7PP744wDMmzdvROo0FOqZEBGRA9b7f/I4W1sjvnz2lNoy7vzESft9fDweB2DDhg1cddVVHH300Zx33nkZ9+3q6uJLX/oSixYtYv78+fv9mYWiMCEiIjLCdu3aRUlJSd/r8ePH88wzz1BaWrrPvs45LrvsMpqamvjrX/86ktXMmcKEiIgcsIbTM+Cn2tpalixZQiKR4MUXX+Tqq6/m4osv5rHHHtvnSo0vfvGL3HXXXTzwwAMcdNBBPtV4YAoTIiIiIywUCrFgwQIATjjhBMrLy7nkkku44447uOCCC/r2u+mmm7jhhhu4/fbbOeWUU/yq7qA0AVNERMRnH/rQh5g/fz7XX399X9ltt93GF77wBW688UbOP/98H2s3OPVMiIiI+MzM+PKXv8wHP/hBHnzwQUKhEJdeeilnnnkmCxcu5Mknn+zbd9q0aUybNs3H2u5LPRMiIiKjwAUXXMAhhxzCd77zHf7xj38Qi8W4//77Wbhw4V6Pn/3sZ35XdR/mnPO7Dr5ZsGCBe/bZZ/2uhoiIDCIajQIQDod9rsnolks7mdly59yCfH6ueiZERERkWDRnQkREZAxbvmE3AMdMrycQsIJ8hsKEiIjIKNATS/BGczcAsydWggMHBPsFgI5IDDOjsjREIulwzhEKegMNzjl2dfQwZVwJZsY9L23hU79/HudgYnUp37/w6ILUXWFCRESkwJxztHTFiMQT1JWXEE04QgEvELRFYuzuiNLREyeZmsf4RnM3nT1xDDh4UlVfoGjpirGpuQsD6ivDtHbFSDhHWUmQhpoydrdFeP/Nj1NTWc6M8RU8tmYnleEQZ81vYOnqJmaNryzIz6cwISIio14gECAajeZtAmYy6WjuimIGVaUlhEMBnHMkndcTsKO9h7ZIjOn1FbRHYuzujFIaChIIeNvHV4YpCQaIJpJEYgnKSoI4Bzvae6irKKG6rIR4IklTew+RWAIzoz0SA7x9etVXhGnpiuKAcCjApOpymtojtHRF+/bZsLuLeMLRE08A3ucbxu7OKEEzqkpDdEYTrN/VSWdXF/VVpbzREmHtzk7qK0r40QeP5aQ5E0gmnYY5RESkeAWDQSI9PSQsRCgUpCwUwMyIJZIkk46SYCDnL8qOSJw3WrqIxpMAGN3UVYTpjiWIxBJUhIN0Rb0v7rU7OogmvP26Y4m+99jZHgXzehx6GYbD64GoqyihPRIjntyzvSIcZFxlKZ09cUqCAZq7ojR3RQkGjFnjK6ks9b6SS0MBtrR0M6G6lN2d0VQPhdeLYUBDbRmhgLG7K8a4ijDhUIDuaJw3mrvY3N7FXVeeQkkoQDzhqAgHMfPapVBBAhQmRETEB8454qkQcPeLW3hi7S6qy0I88upOJlaX8vNLj+vr2m+LxCgvCXLtX1/nzVNKmTd9PFVlYUIB6/uihz1zC5JJ74szFDCCASOZ9OYUlJUESCahOfVXf31FmJKQ0dIZY3d7JwDhYJDO7hglwQDlJSHaIj0EzJg5roJAwHAOIvEEzZ1eL0MoZJSGAnRFEySSSWrLS9jV2cPudi8kTKgMU1ceJp5whEsCBAyqSryv3qqSEnZ1RqkvL6HEkn2XdZYYzKz3bvgVrgqx27z3LSsJpn7SJCRhXFkAknF6ehwukWBiaZKDjp5NMOjtVzqC3/AKEyIisl+6onGWb2hm1dZ2zpg3mVkTso/Ht0dibG+LUBoKEgwYV/x2OTvae/jCmYfxxf99iUTaX/BshTue3UTCOX71+Hpe3d5BWUmASCzJjo561rUmeb2pnS0t3cxvrGF8VSk723toavduRV5VFqKtO86uzihdPXFKSwL0xPaEjvrKMN9+zxFMri0DoKashKWrdzB9XDkHTazk1e3tTKopo7wkyP0vb+PYmfWUhfd8XZaWBKktzz7c0hlNsH5nB8fPHt8XAELBffcLh4JMqS0fsI1LggEm15QNuI+ZEQ6HKS8v7+uFGGlatEqLVonIGJJMOlq6Y4yr3PvLzjlHc9ee8p54gq/++WVmjq/kE4vmsGF3F7cse52Xt7TyvQuO5pHXdvJ/L27hvy88hsk1ZezujNKQ+vJ9fUcHP3poDfe9sq1vOKChpoxvnDef25/eSDSRJGBGS1eMi46fQX1FCZ/9wwv0pIYVggHbKzyYwfcvPIbxlWEmVZfy7h8+RiLpiCaShEMB3jyjnq2t3Rw+pYYbzz+KinDufwcnko5gwNjWGmHltjYMOHp6HXUVxbv4VSEWrVKYUJgQkVHMOUdTe8+gf50657jh76u57amNtHTH+PHFx3L2EVMA7wv1mjte5E/Pb+YLZxzKVW87mK/+5WV+++RGAOY2VPPq9nZ6v99rykK0ReIAHDKpCjNY09TBrR8+jhVb2/juA6+SSDpmjq/gHfMbiCcdtz66bp869YYGM6goCfLuoxtp646zpqmDi46fTiBgfOPuFXz29EP41OmH9B33vSWv8r0lr3HY5Gp+8ZHjaKwb+K93GRqFiTxTmBCRkRaJJfifh9dyxLRaFh06cZ9u6a2t3Wzc1UV7JE5rd4w7lm/iybW7ufzk2XzlnYdn7ca+56UtXHXb80yoCtMeiVNVGuLvn3srdRVhPv/HF/jLC1swA+e8XoRtbRHmNlRTEQ7y3MYW5jfW8MlTD2ZzSxff/tsqJlSFOfngCfz5hS0AhAJGwLw5ClPryvnquw7nrPkNffX53pJXefjVHXzxHXM5ekYdiaRjV0eUj/7yGba0dPPLjx7PcbPG7VPvrmh8n56GeCLJkpXbOengCdSUleSj2SWNwkSeKUyIyHA454jEkpSHMwyIZ/GTpa9z/X2rAJjfWMMVi+ZwWEM1S1c38avHN7C5pXufY6rLQrRH4lx28mw+8pZZ/OaJDSzf0ExzV5RwKMjUujKe39hCR0+c+z77Vh5cuZ1v/XUlx88ax4zxFdy5/A3ePLOe/3r/kXztL6+wdkcHjXXl3PCBo2isK+f1HR3MbajuCwaPvLaD2RMqaagp4+ZlrzOvsYb2SJzP3P4CM8ZV8PuPncjUHHsLYokkXdEEteUKBaOFwkSeKUyISC46e+Ks2tbG9PoKvvfgazy2ZifnHtnIkpXbeaO5m99dfgJHTa/jhU0t3ProOo6fPY5DJ1Xxk2Wvc86bpnD+cdMBbxLiKd/5B8mk45RDJ/K3f24l/VdwZTjISQdP4NDJVdSUeWsVzJ1STWNtOR+45XE27d4TNIIBo7a8hGg8SUePNyTx5XPm8rG3ziGRdHzs18/y4KomwBvG+MPHFw77C/2lN1qYMa6iqOcbjAUKE3mmMCEytq3a1saP//E6Z8ybzIxxFWxvi3D64ZMJBoznNzbz1b+8TDzhOG7WONojMWrLS5haX86uzijHzqjnzHmTeX1HJ5f/6hnW7+rqe9/e4YLeEYcJVaXMbajmkdd2ZqzHWw4e33dJ4qpt7Vx95qFc9bZDWLezkzuXb6K1O0ZjXTkfPH4mtRWZv/DbIjF+9dh6HlrdxDlvmsIlJ82kNHWJwPa2CG80d3HsjPq+3gXnHA+s2M7SV3fwmdMPGXTOhRQPhYk8U5gQOXBta43w+Os7OeeIKZSVBHltezs/WfY6yaTjouNnMK+xhnN/8OheIQDgkoUzmTGugv+8dxVJ561z0Lt4UX/T6svZ2hohkXS888gp7Oro4fjZ4/nwwpkse3UHhzVU89Ta3Vx3zwoAjptVz8ffOocHV23n9aZOPvKWWfz3Q2tYubWt7z2n1pVz/+feStVILgIgkkZhIs8UJkT81dkT58dL13DuUY3MbajpK4/EEtzz0lZOPWwiBvzy8fVE40lOOngCiw6dyH/dv4pblq0lnnRcdvJs3jyznqtue470pQrKS4J0xxJcfMIMKkqC9MSTPL1uN6u3twN7bnp09PQ61jR1UF8RZndnlK2tEapKQ/zisXU89vpO5jfWculJszj3qMaMP4NzjmWv7mBqXTmHTK7eZ3sy6djZ0UN9ZRiXWqq5/42bREaSwkSeKUyIFF40nqQtEmNCVek+23740Gvc8PdXmVhdyu0fO5GOSJy6ihK+ctfLPLpmJ421ZYRDgb7ehYDBxxfN4SdLX2dafTmJpGNHew/l4SDxhOO/LzqGmrIQf3h2Ew+s2M7hU2r43eUnUJK6o+LW1m4u+umTjK8q5UcXH9u3bkI2zjnfFgESKRSFiTxTmBDJv1e2tNIVTTBzXAXjq0q55OdP8diaXRw5rZaz5jdwxNRaWrtjHDuznnf/4FFau/e+f0GvuQ3VrN7ejnPwiVPncOTUWq76/fMkko6SoHHPp05h3c5OrvjtcgC+9S9v4kMnzuw7Ppl6z/73IyjkzY5EDgSFCBMatBORvFn26g4+/POnAW9dghMOGsdja3Yxta6cFVvaeOmN1r59QwEjnnRcedocDOOJtbuY31jDrs4o0+sr+PwZh7J8QzPtkRhnzm8A4N92d/Gf967is28/lMMaqjl0chWXLPQCxAdPmLFXXbIFBgUJkfxTz4R6JqTIOedIpG6EtGl3F3csf4P6ihIaasqYWl/O3IYaOnviPLN+N8+s383T65vpiMT4yFtm8/zGFl58o4VTDpnAW+ZM4No/v8zOjh4uPWkW97y0lW1tEabWlfO3T58CwJKV29nU7A1Z3ProOpyDpdecmnEIJJvNLd05r3EgIvvSMEeeKUzIWBSJJXjktZ2ccNA4QgFj9bZ25jfWEg4FaO2KcfWdLzKxupTPvv0Qrrt7BY+t2UlXNMFPL1nAjX9fvVfvAUA4tPfVDqGAEQhYX1lJ0Igl9vwe+fwZh/Lp0w+htTvGb5/cwFnzJ3PwpH0nJrZ2xeiKxQe90ZGI5JfCRJ4pTMhY09wZ5bJfPcNzG1uoTK3K2BlNMK4yzKmHTmTF1jZWbfOuZggHA31LI+9o7yGeTJJ0cPabGnjXkY1sa4uwbmcHL2xqoa48zHGzxnHc7HqOmV5Pc1eUnz68lnmNNbz7qEaeXrebp9btIp5wXH3WYX0THkVk9FGYyDOFCTmQxBJJEklHWUmQaDxJPJnsu6dBLJHkoVVNXHf3Cja3dPOWg8ezYVcXZnD8rPH8Y3UTuzujAFx8wgxauqL87Z/b+MSpc7jmzMO46/nNfOGOF5lQFeaBzy2ivlIrHIqMVZqAKVIE3mjuorkzxhHTagFo7Y7xh2c2csuytXRG45w0ZwLPrN9NR0+c2eMrcXjzCKLxJKGA8enTD+Gzpx+y10TDeCLJpuZuuqJx5k3x1nPY1Rntm6vwvjdPY1xlmOnjyhUkRGTIFCZERkAkluCHD61he1uE6rISptSWsbsrStI572ZM4yqYUlfOzvYe3vPjx2jtjnHTBUfz3IZmbn9mEz3xJFWlISZWl/LQqiYm15RyeEMNrza1UxIMcOjkKo6YWsdlJ8/KOD8hFAwwe0LlXmX9Jz2eNndSQdtARMYuDXNomEMKaOOuLiZUh/n/7l3Fr5/YMOC+oYBRVRaipSvWN58BYOb4Ci48bgYXHT+dmrISNu7uYlp9OSHNSxCR/aBhDpFRzDlHe0+cmjLvRk33v7KNK367nKpS7/bRh06u4uYPvZn2SJytrRHGVYaJJZI8u76Z7e0RXt7cyktvtPLJU+ewcM54rr7jRd59VCNXn3VY3w2dAGb162EQEfGbrz0TZjYP+AGwEGgBfgZ8wzmXGOS4BcC3gTcDBjwHfMU599RQPl89EzKQ3zy5gRv/vppff/R4jpxW11e+aXcXD7+2gw27uphYVcrCOeOZ21DNx3+znEde28l3LziKKbXlXHLrUyQdlIeDdEcT3HXlSXvdfyKTrmi8b1KliEghjKmeCTOrB5YAK4DzgDnAjUAAuHaA46anjnsOuCRVfA3wdzM70jk3cF+ySA7aIzFu/PtqWrpiXHnbc1yxaA6PvLqTLa3d+6zDAN5QxIbU/SOuuu35vvIfXXwspx8+ia7U5ZmDUZAQkQORn7+5rgDKgfc659qAB8ysBlhsZt9JlWXyTqA6dVwLgJk9DuwEzgF+UvCay5jS0hXlybW7mN9Yi3OwblcnS1c30dIVY87ESl7f0clX7noZ8O74ePyscfzLMVM5dHIVm1u6uf3pTTyxdhfzptTw9XPn8aU//ZPp4yr4f6fM5pRDJgJQVhIcqAoiIgc0P8PE2cD9/ULD7cD1wCLg7izHlQBxoCOtrCNVpkX3ZUCdPXF+9cR6QgHjwuNnsLm5myt+u7yvVyHduMowf/rkW/jvB1+jMhzkAwumM7WufK9LLhcA7z6qkVe2tDFrQiVVpSH+cfWpI/cDiYiMAn6GibnAQ+kFzrmNZtaV2pYtTPwvcB1wo5n9R6rsa0AzcEeB6ioHqNXb2vnmPStoao8QMGNXZ5Qd7T0A/Oe9q+idMnT+gmns7owRDhmHTq6muTPK6YdPpra8hK++a96An2FmvGlqbaF/FBGRUcvPMFGPN+myv+bUtoycc1vM7DTgHuDTqeKtwFnOuR2DfaiZLQa+DjBlypSh1VhGtbU7Ovjn5laSznHIpGoOnlTFlbc9x5qmDmrLS0gkvRtaXXXawZSHgyxZuZ2GmjLOPaqRc47QuSAisr/8nu2V6VISy1LubTSbAtwJLAcuTxVfCfzVzE5yzm0c8AOdWwwsBu9qjqFXWUaLnR09/M/DaznxoPE8sHI7tz2193/62vISWrtjfHjhTL5x3psA7/JNM2+Y4srTDh7xOouIjEV+holmoC5DeS2Zeyx6XYNX7/c752IAZvYQ8BpwNXt6K2QM644muOxXz/LiphZueXgtAHMbqrnguOmEAsazG5q59+VtHDypii+dfXjfcb1BQkRE8sfPMLEKb25En9Rln5WpbdnMBV7pDRIAzrmomb2Cd3mpjEHOOe59eRtPr9tNTzzJs+t381pTB+88YgqVpUFKQ0G+fM7hlKfulPmvC2fxHz1xgmZ9ZSIiUhh+hol7gWvMrNo5154quwDoBpYNcNwG4BwzCzvnogBmVgq8ieyTNuUA8pcXNrNqWzsff+tBfPYPL/Dy5jZmja/g2Q3NffuEAsY7j5zCTecfTTiUeVnpqlK/R/FERIqDn79tb8YbkviTmV0PHIQ3l+G76ZeLmtkaYJlz7rJU0c/w5krcZWY/xptjcSUwBfjpyFVfCuHpdbv5/B9fJJF0/PKx9XTHEpSGAjy7oZnjZtXzpbPnUlVawszxFVq7QURklPAtTDjnms3sdOCHeD0KLcBNpCZHpgkBwbTjlpvZO/CuyPhNqvifwBnOuRcLXG0poMfX7ORzf3wBgLcfPpklK7dz+txJ/OiDx7Kzo4fG2r3XeBARkdHB135g59wK4G2D7DMrQ9mDwIMFqpaMsETS8YU/vsCfX9gCwLXvPJzLTp7N6zs6mD2himDAmFZf4XMtRUQkGw0qi6/iiSRfuetl/vzCFo6aXsc3z5vfd1OtgydV+1s5ERHJicKE+CKZdDy9fjf/+beVvPhGK0dMreW3lx1Pder23SIicuBQmJAR19QW4YKfPsm6nZ0AvOeYqSw+d76ChIjIAUphQkbcj/6xhnU7OznniAYuWTiLEw8a73eVRERkGBQmZERtbunm909vYsa4Cr5/4TGUBDOvESEiIgcO/SaXEXXTA68STST59OmHKEiIiIwR+m0uI2bp6ibuXP4Gcxuq+ZejG/2ujoiI5InChIyI1u4YX/rffxIKGDeefxQh9UqIiIwZ+o0uI+Kb96xgW1uET73tEOY31vpdHRERySOFCSm4h1Zt587lbzC/sYZPnqYbu4qIjDUKE1JwNy9dixnceP5RmnQpIjIG6Te7FFRzZ5RnN+zm6Ol1zG2o8bs6IiJSAAoTUlBLX20i6by7gIqIyNikMCEFtWRlE6AwISIylilMSMFE40keXr2DafXlHDq5yu/qiIhIgShMSMH8c3Mr7T1xTjtsEmbmd3VERKRAFCakYJ7f2AzAgln1PtdEREQKSWFCCub5jS0AHDNdYUJEZCxTmJCCeX5jMxOqwkwfV+53VUREpIAUJqQgtrVG2NIa4ejp9ZovISIyxilMSEH0zpc4dmadvxUREZGCU5iQgnguFSY0X0JEZOxTmJCCeH5jCwGDo6brDqEiImOdwoTkXTSe5KXNrcxtqKEiHPK7OiIiUmAKE5J3K7e2EY0nNV9CRKRIKExI3mm+hIhIcVGYkLzrW6xqRp2v9RARkZGhMCF599zGZuoqSpg9odLvqoiIyAhQmJC82tHewxvN3RwzvU6LVYmIFAmFCcmrFVvbADhiqi4JFREpFgoTklerUmHi8Ck1PtdERERGisKE5NXKVJiYqzAhIlI0FCYkr1Zta6e8JMjMcRV+V0VEREaIwoTkTU88wZqmDg5rqCYQ0ORLEZFioTAhebOmqYN40mm+hIhIkVGYkLxZtbUdgHlTqn2uiYiIjCSFCcmbVds0+VJEpBgpTEjerN/VBcBBWvlSRKSoKExI3mxp6aasJMC4yrDfVRERkRGkMCF5s6Wlm8a6ci2jLSJSZBQmJC+6onGau2JMrSv3uyoiIjLCFCYkL7a0RABorFWYEBEpNgoTkhdbWroBmFqvMCEiUmwUJiQvNqfCRKOGOUREio7ChOTFlr4wUeZzTUREZKQpTEhe9PZMaAKmiEjxUZiQvOjtmWioVc+EiEix8TVMmNk8M3vQzLrMbIuZXWdmwRyPfa+ZPWNm3Wa2y8zuMzMtveiTLS0RJlaXUhrK6T+fiIiMIb6FCTOrB5YADjgPuA74AvCNHI69HLgNuBc4G7gceA0IFaq+kl0y6dja2q0hDhGRIuXnl+8VQDnwXudcG/CAmdUAi83sO6myfZjZBOAm4FPOuf9J23RXwWssGe3s6CGWcJp8KSJSpPwc5jgbuL9faLgdL2AsGuC481PPvypUxWRotrf1ADCpWmFCRKQY+Rkm5gKr0guccxuBrtS2bE4AVgOXmdkbZhYzs6fM7KTCVVUGsqPDW/1yUk2pzzURERE/+Bkm6oGWDOXNqW3ZNACHAdcCXwTOBTqB+8xs8mAfamaLzcyZmduyZcuQKy37alLPhIhIUfP70lCXocyylPcKAFXAZc653znn7gP+BUgAVw36gc4tds6Zc84aGxv3o8rSX1N7b5hQz4SISDHyM0w0A3UZymvJ3GPRa3fqeWlvQWrexXJgXn6qJkPR1O4Nc0xUmBARKUp+holV9JsbYWbTgUr6zaXoZyVez4X1Kzcgmc8KSm72DHMoTIiIFCM/w8S9wFlmVp1WdgHQDSwb4Lh78ILDab0FZlYLvBl4sQD1lEHs6OghFDDqK8J+V0VERHzgZ5i4GegB/mRmbzezjwGLge+mXy5qZmvM7Nbe1865Z4G/ALea2YfN7J3A/wEx4Ecj+QOIp6mth4nVpQQC/TuLRESkGPgWJpxzzcDpQBC4G2/ly5uAr/fbNZTaJ92HgD8D3wXuxAsSb0u9p4wg5xw72ns0xCEiUsR8XX7aObcCeNsg+8zKUNYBfCL1EB+1dseIJpKafCkiUsT8vjRUDnC9l4VO1BoTIiJFS2FChkVXcoiIiMKEDIuW0hYREYUJGRYtpS0iIgoTMixaSltERBQmZFj2TMBUmBARKVYKEzIsTW3enIkJVQoTIiLFSmFChmVHew/jKsOEQzqVRESKlb4BZFi0+qWIiChMyH7rjiZo74lrvoSISJFTmJD91tSeWmNCl4WKiBQ1hQnZb7qSQ0REQGFChkFLaYuICChMyDD0DXNoKW0RkaKmMCH7bUe7ltIWERGFCRkGLaUtIiIwhDBhZlPN7AYz+5uZ/cbMzsmy33lmtjZ/VZTRShMwRUQEIJTLTmY2BVgOTAKagBrgYjO7C7jUOdeRtnsVMDPfFZXRp6ktQmU4SGVpTqeRiIiMUbn2THwDKAdOcc41ABOArwPvBpaa2cQC1U9GsR3tPUyq0XwJEZFil2uYOA34oXPuMQDnXJdz7lvAO4A5wCNmNqNAdZRRKJZIsqszqiEOERHJOUw0Amv6FzrnHgJOBeqAR81sbt5qJqNac2cUgIm6W6iISNHLNUxsBWZn2uCcexE4BXDAw8CC/FRNRrPmrhgA9ZUlPtdERET8lmuYeAp4T7aNzrnXgJOBZuDTeaiXjHLNXV7PRF152OeaiIiI33INE/8LNJjZomw7OOc24QWKlwDLQ91kFGvpDRMV6pkQESl2OV3T55z7E/CnHPbbARxjZuOHWzEZ3Vp6hzkq1DMhIlLs8roCpplNMbMbgXX5fF8ZfTRnQkREeuW82lBqLYlLgIOB3cCdzrnnU9umAF8FPgKUAkvzXlMZVfYMc6hnQkSk2OW6AuahwCN4i1X1zof4NzO7CIgDv8Rb+fIu4Hrn3LP5r6qMJnsmYKpnQkSk2OXaM3EdUA18Ei9UzAZuAr4H1AMPAP/mnHu1AHWUUUhzJkREpFeuYWIRcItz7pbU6xVmlgT+CvzeOffBgtRORq2WrhhmUKOeCRGRopfrBMyJwIv9yl5IPf8xb7WRA0ZzV5Ta8hKCAV0FLCJS7HINEwEg2q+s93V7/qojB4rmrpiGOEREBBjC1RzAIWZ2Utrr2tTzfDOL9N/ZOff4sGomo5ZzjpauKNPqy/2uioiIjAJDCRNfSz36+z7efTl6Wep1cBj1klGsM5ognnTUa/VLEREh9zDxkYLWQg4ovXcM1TCHiIhA7stp/6rQFZEDR+9loVqwSkREIM/LaUtx6F2wSsMcIiICChOyH1q6e3smFCZERERhQvaD7sshIiLpFCZkyJo7tZS2iIjsoTAhQ9bS3dszoWEOERFRmJD90NYdB6CmTGFCREQUJmQ/tEe8YY6a8qGseSYiImOVwoQMWVsqTFSVKkyIiIjChOyH9kicinCQUFCnj4iIKEzIfmiPxDVfQkRE+ihMyJC1RWJUl2mIQ0REPL6GCTObZ2YPmlmXmW0xs+vMLOe7jZpZwMyWm5kzs3cVsq7icc55PRPl6pkQERGPb39emlk9sARYAZwHzAFuxAs41+b4NpcDUwtSQcmoO5YgkXTqmRARkT5+9kxcAZQD73XOPeCcuxn4BvB5M6sZ7OBUGPkP4CuFraak611jolpzJkREJMXPMHE2cL9zri2t7Ha8gLEoh+O/CTwGPFiAukkWfWtMqGdCRERS/AwTc4FV6QXOuY1AV2pbVmZ2JPAR4OqC1U4yaouoZ0JERPbmZ5ioB1oylDentg3kB8CPnHNrhvqhZrY4NWHTbdmyZaiHF73eBas0Z0JERHr5fWmoy1BmWcq9jWYXAocB39qvD3RusXPOnHPW2Ni4P29R1NpTPRO6mkNERHr5GSaagboM5bVk7rHAzEqA/wKuBwJmVgf0TtasNLPqvNdS9qI5EyIi0p+fYWIV/eZGmNl0oJJ+cynSVALTgO/ihZFm4MXUttuB5wtSU+mz52oOhQkREfH4+Y1wL3CNmVU759pTZRcA3cCyLMd0AKf1K2sAfg98GXioEBWVPfb0TGiYQ0REPH6GiZuBTwN/MrPrgYOAxcB30y8XNbM1wDLn3GXOuTiwNP1NzGxW6p//dM49NQL1LmrtuppDRET68S1MOOeazex04IfA3XjzJG7CCxTpQkDOS2xLYelqDhER6c/XbwTn3ArgbYPsM2uQ7evxrgCREaCrOUREpD+/Lw2VA0x7JEbAoDKsziIREfEoTMiQtHXHqSoNYabOIBER8ShMyJC0R2Ia4hARkb0oTMiQtEfiupJDRET2ojAhOUskHe09cV3JISIie1GYkJx19KSu5FDPhIiIpFGYkJy1dXtrTNRqzoSIiKRRmJCctabCRE25hjlERGQPhQnJWat6JkREJAOFCcmZwoSIiGSiMCE5U5gQEZFMFCYkZwoTIiKSicKE5ExXc4iISCYKE5Iz9UyIiEgmChOSsz2XhipMiIjIHgoTkjP1TIiISCYKE5Kztu4Y4VCAspKg31UREZFRRGFCctbaHVOvhIiI7ENhQnLWFokrTIiIyD4UJiQnzjn1TIiISEYKE5KTzmiCRNIpTIiIyD4UJiQnfZeFlumOoSIisjeFCclJa5cuCxURkcwUJiQnWmNCRESyUZiQnGj1SxERyUZhQnLSFlHPhIiIZKYwITnRHUNFRCQbhQnJiYY5REQkG4UJyUl7JA5AtS4NFRGRfhQmJCcdPakwUaqeCRER2ZvChOSkMxUmKkt1x1AREdmbwoTkpLdnokrDHCIi0o/ChOSkoydOSdAoDalnQkRE9qYwITnp7IlTWapeCRER2ZfChOSkIxKnMqwwISIi+1KYkJx09MR1WaiIiGSkMCGDcs7RGU1omENERDJSmJBB9cSTJJJOYUJERDJSmJBB9a1+qTAhIiIZKEzIoLRglYiIDERhQgbV0Rcm1DMhIiL7UpiQQfWtfqkwISIiGShMyKA6FSZERGQAChMyKA1ziIjIQBQmZFCdPQlAPRMiIpKZwoQMqqMnBihMiIhIZr6GCTObZ2YPmlmXmW0xs+vMbMDrD83sODP7hZmtSR232sy+bmZlI1XvYtOR6pnQMIeIiGTi27eDmdUDS4AVwHnAHOBGvIBz7QCHXpDa93rgNeBI4Jup5/cVsMpFSxMwRURkIH5+O1wBlAPvdc61AQ+YWQ2w2My+kyrL5Hrn3I6010vNLALcYmYznXMbClzvotORWgGzSjf6EhGRDPwc5jgbuL9faLgdL2AsynZQvyDR6/nU86T8VU96dUS1AqaIiGTnZ5iYC6xKL3DObQS6UtuG4iQgCazOT9UknYY5RERkIH6GiXqgJUN5c2pbTsysAfgK8JsBhkZkGDoicQIG5SXqmRARkX35fWmoy1BmWcr33dEsDPwR6AA+l+Mxi83MmZnbsmVLzhUtZh09cSpLQ5iZ31UREZFRyM8w0QzUZSivJXOPxV7M+2b7NTAfOMc515zLhzrnFjvnzDlnjY2Nude2iHVG4xriEBGRrPwME6voNzfCzKYDlfSbS5HFTXiXlJ7nnMtlf9lPnT0JrTEhIiJZ+Rkm7gXOMrPqtLILgG5g2UAHmtm/A58CPuSce7RwVRTw5kyoZ0JERLLxM0zcDPQAfzKzt5vZx4DFwHfTJ1KmVrq8Ne31xcC38YY4NpvZiWmPiSP7I4x90XiSaCKpMCEiIln59g3hnGs2s9OBHwJ3482TuAkvUKQLAemXEZyZer409Uj3EeCXea1okWvt9u7LUVte4nNNRERktPL1z03n3ArgbYPsM6vf60vZN0RIgbR2RwGorVCYEBGRzPy+NFRGuZYur2eiTj0TIiKShcKEDKgvTKhnQkREslCYkAFpzoSIiAxGYUIG1NIXJsI+10REREYrhQkZUGuXNwFTwxwiIpKNwoQMqLdnQmFCRESyUZiQAe25mkPDHCIikpnChAxIEzBFRGQwChMyoJbuGOFQgLISnSoiIpKZviFkQK1dUerKS/Du+C4iIrIvhQkZUEt3TJMvRURkQAoTklUy6Wjrjmm+hIiIDEhhQrJq74mTdFqwSkREBqYwIVm16r4cIiKSA4UJyaoldftx3TFUREQGojAhWemOoSIikguFCclKC1aJiEguFCYkq747hlZoAqaIiGSnMCFZ9d0xVD0TIiIyAIUJyaq9Jw5AVVnI55qIiMhopjAhWfXEkgCUlwR9romIiIxmChOSVXc0AShMiIjIwBQmJKvuWCpMhBUmREQkO4UJyao3TJSFFCZERCQ7hQnJKtIbJsI6TUREJDt9S0hWkViCgEE4qNNERESy07eEZNUdS1BeEsTM/K6KiIiMYgoTklV3NEGZruQQEZFBKExIVpFYUmFCREQGpTAhWUViCV0WKiIig1KYkKx650yIiIgMRGFCMnLO0R1LUFaiU0RERAambwrJKJpI4hyaMyEiIoNSmJCMIlHd5EtERHKjMCEZ6b4cIiKSK4UJyUj35RARkVwpTEhGEfVMiIhIjhQmJKO+ngnNmRARkUEoTEhGkWiqZ0JhQkREBqEwIRnt6ZnQKSIiIgPTN4VkFImlLg3VnAkRERmEwoRkpDkTIiKSK4UJyahvnQmFCRERGYTChGTUOwFTPRMiIjIYhQnJKKKeCRERyZHChGS0ZzltnSIiIjIwfVNIRpqAKSIiufI1TJjZPDN70My6zGyLmV1nZoN+e5lZrZn9wsyazazVzH5nZuNHos7FIqIwISIiOQr59cFmVg8sAVYA5wFzgBvxAs61gxz+B+Aw4HIgCVwP/Bk4pUDVLTp960woTIiIyCB8CxPAFUA58F7nXBvwgJnVAIvN7Dupsn2Y2ULgLGCRc+7hVNlm4Ckze7tzbskI1X9M69Zy2iIikiM/hznOBu7vFxpuxwsYiwY5bntvkABwzj0NrEttkzzo1l1DRUQkR36GibnAqvQC59xGoCu1LefjUlYOcpwMQW+YKA1pjq6IiAzMz2GOeqAlQ3lzatv+HHfQYB9qZouBr6dexszspcGOKWaB6/v+2Qhs8a8mBwy1U+7UVrlRO+VG7ZS7w/P9hn6GCQCXocyylOfjOJxzi4HFAGbmnHMLBjtG+tqq0e96jHZqp9yprXKjdsqN2il3Zjbod+VQ+dmH3QzUZSivJXPPw2DH1Q1ynIiIiBSAn2FiFf3mOJjZdKCSzHMish6Xkm0uhYiIiBSQn2HiXuAsM6tOK7sA6AaWDXJcg5md3FtgZgvw5kvcO8Q6fGOI+xcztVVu1E65U1vlRu2UG7VT7vLeVuZc3odOcvtgb9GqFcDLeItOHQR8F/iec+7atP3WAMucc5elld0HHApczZ5Fq5qcc1q0SkREZIT51jPhnGsGTgeCwN14Sekm9lxp0SuU2ifdhXi9Fz8Hfg0sB95TyPqKiIhIZr71TIiIiMjYoBWJREREZFgUJkRERGRYFCZERERkWBQmREREZFiKMkyY2Twze9DMusxsi5ldZ2ZFfXtMM7vUzFyGxxVp+5iZfdnMNplZt5k9bGZH+1jtgjOzg83sFjN70cwSZrY0wz45tctYPu9ybKf1Gc6vbRn2G8vt9AEz+z8z22xmHWa23Mwu6rePzqfc2qnozycAM3u/mT1uZrvMLGJmq83sWjMLp+1T8HPK73tzjLjU+hZL8Na4OA+YA9yIF6yuHeDQYvE2vIXDeq1N+/eXgK8C1+CtNvp5YImZvck5t8//xGPEfOAc4EkgnGWfQdulCM67XNoJ4DbgB2mvo+kbi6CdPg+sAz4H7MRrs9vMbIJzrrdddD7l1k6g8wlgPPAP4L/wbilxPN79pxqAq1L7FP6ccs4V1QP4d7z7e9Sklf0b3q3Pa/yql98P4FK8G6VVZdleBrQCX0srqwR2AN/yu/4FbJdA2r/vBJbuT7uM9fNusHZKla8HbhjkfcZ6O03IUHYbsE7nU+7tpPNp0Pb7D7xgYSN1ThXjMMfZwP3Ouba0stuBcmCRP1U6IJwE1AB/7C1wznXiLTh2tl+VKjTnXHKQXXJtlzF93uXQTrka6+20M0Px88Ck1L91PpFTO+VqTLfTAHaxp4dwRM6pYgwT+9wQzDm3ES99ZbqBWLF53cziqXG3j6eVzwUSwGv99l9Jcbdbru2i887zUTOLmlmrmd1pZjP7bS/GdjoJr2sZdD4NJL2deul8SjGzoJlVmHffqk8DP3Fe98KInFNFN2cCqCfzrcqbU9uK1Va8MbWn8ZYvvwi42cwqnHM34bVNh3Mu0e+4ZqDCzMLOuSjFJ9d20XkHf8GbU/EGcDje0vmPmNkRzrnW1D5F1U5mdjre+PRHU0U6nzLI0E6g86m/TqA09e9f482PgBE6p4oxTIA3N6A/y1JeFJxz9wP3pxXda2alwLVm9v3e3TIcagNsKxa5tktRn3fOuc+kvXzEzB4HXgA+AnwvfdcMh4+5djKzWXjzAP7inPtl2iadT2mytZPOp32cBFTgTcD8GvBD4JOpbQU/p4oxTDQDdRnKa8mcyorZncD5wCy8dqs2s2C/hFsHdDnnYiNfvVEh13bRedePc+5lM1sNHJtWXBTtZGbjgHuBjcCH0jbpfEozQDvto5jPJwDn3HOpfz5qZjuBX5nZjYzQOVWMcyZW0W/8x8ym481uXZXxCHF4bRMEDu63bZ9xtiKTa7vovMsu/a+eMd9OZlYB3IM3Qe6dqclwvXQ+pQzSTgMpqvMpi95gMZsROqeKMUzcC5xlZtVpZRfgra2wzJ8qjVrvw7vGewPwONAGfKB3Y+p/9nPx2rRY5douOu/6MbM3AYcBy9OKx3Q7mVkIuAM4BDjbOdfUbxedT+TUTpmOKbrzaQBvST2vY6TOKb+vh/Xh+tt6vMmGDwBvBz4GdDCG10rIsV3+F/gi3uVB7wJ+g5fwP5W2z7/jzey9Ejgd+Cte2Jjsd/0L2C4VwPtTjyeAV9JeV+TaLmP9vBusnYB3Ar8HPgicBnwC2Iy3KFr6de1jvZ1+mvr/6tPAif0epTqfcmsnnU97tdV9wNWp391nAt9I/Yy3p+1T8HPK94bwqfHnAQ/hJa6twDeBoN/18rlNvg2sTp1w3Xjp/l/77WPAV/BmT3cDjwDH+F33ArfLrNQvtUyPWUNpl7F83g3WTsCRwIN4C+XEgG3AL4HGImun9Tqfht9OOp/2+vm+Cbyc+uJvwRvi+BRQkrZPwc8pS72BiIiIyH4pxjkTIiIikkcKEyIiIjIsChMiIiIyLAoTIiIiMiwKEyIiIjIsChMiIiIyLAoTInJAMrP1ZrbE73qIiMKEiIiIDJPChIiIiAyLwoSIiIgMi8KEiGRlZpPN7GYz22xmUTNbY2b/bmaB1PZZZubM7Foz+3hqe8TMnjezMzO833Qz+62Z7Ujt96KZXZphP0u933Iz6zKzZjN71MzOy7DvcWb2mJl1m9kmM/t8QRpDRLLSvTlEJCMzmwA8A5Th3cVxC96tjf8VuMU5d4WZzcK7zfGLwGTgx0AE+DgwA3ibc+7RtPd7HhgP/ADvLo/np97zGufcDWmffXPqPZbi3eEwChwHtDvnPpnaZ32qvAbvLrdr8W6ZvAh4h3Pu/vy3iohkojAhIhmZ2S3A+4AjnHNb08q/DXwJmIv3Zb4OiAPznXOvpvaZCLwGrHTOLUyV3QB8gbQvejMrAZYBxwDTnHO7zOytqbJfAh91ab+kzMx6X6fCxEzgbOfcfamyUmAj8LBz7gOFaBcR2ZeGOURkH2ZmwAeAvwExM5vQ+wDux7ul8Wlph/ytN0gAOOd2AL8DTjSz8anidwEvp/cYOOdiwE14vR+np4p7Q8BXXL+/dvq/Btb3BonU9h7gSeCg/fixRWQ/KUyISCYTgXq8IY0d/R5LU/tMStt/dYb36C2blfa8MsN+K1LPs1PPBwO7nXNbcqjn+gxlzcC4HI4VkTwJ+V0BERmVev/Q+APwsyz7rE37d6bxUsvxs3r3c2mvcx1/TQzyniIyAhQmRCSTHUAbEHbOZV1lMjUBE7z5E/0dmnrekHpen2W/uWnbwZtrcZaZTXXObc69yiLiFw1ziMg+nHMJ4A7g3WZ2XP/tZladmuzY6xwzOzRt+0TgYuAp59zOVPHdwBFmdkbafiHgs3hXgPSGljtSz99Kzd1I/1z1OIiMQuqZEJFs/h04FXjEzG4FXgKqgfnA+4Ej0vZ9BVhmZj8CevAu66wC/i1tn+uBC4E/m1nvpaEfYM+lobsBnHMPm9nPgMuBWWZ2N95VI28GuoArC/LTish+U5gQkYycczvM7ATgWuA84P8BLXjDEN8EtgENqd3vxBsauQaYjjfR8lzn3MNp77fTzN4C/CdeUKjGm6T5UefcL/p9/MeAF1LP38YLEa8A38n3zykiw6d1JkRkv6UtWvVV59y3fK6OiPhEcyZERERkWBQmREREZFgUJkRERGRYNGdCREREhkU9EyIiIjIsChMiIiIyLAoTIiIiMiwKEyIiIjIsChMiIiIyLP8/2AoWGy11kK4AAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 576x396 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "np.random.seed(123)\n",
    "tf.random.set_seed(456)\n",
    "list_cell_lines_names=['MCF7','A375','HT29','PC3', 'HA1E','YAPC','HELA']\n",
    "numb_drugs=100\n",
    "numb_genes=978\n",
    "num_cell_lines=7\n",
    "\n",
    "# Set landmark gene names\n",
    "gene_names=full_deepce_data[0].columns[5:]\n",
    "\n",
    "# Convert data matrix to dataframes\n",
    "data_mat_nan1=data_mat_nan.copy()\n",
    "data_mat_nan_pd=[]\n",
    "for i1 in range(0, num_cell_lines):\n",
    "    data_mat_nan_pd.append(pd.DataFrame(data_mat_nan1[i1], columns=gene_names.transpose()))\n",
    "\n",
    "# Set autoencoder hyperparameters    \n",
    "features_nr=978\n",
    "latent_dim=100\n",
    "hidden_layer_dim=150\n",
    "test_set_percent = 0.1\n",
    "num_epochs=300  \n",
    "size_batch=40\n",
    "optim = adam_v2.Adam(learning_rate=0.001)\n",
    "\n",
    "# Fonts sizes for the R2, train and validation loss plots during the training of the autoencoder\n",
    "SMALL_SIZE = 15\n",
    "MEDIUM_SIZE = 17\n",
    "BIGGER_SIZE = 20\n",
    "\n",
    "# Build autoencoder for MCF7 in this example      \n",
    "for i in [0]: \n",
    "    \n",
    "    # Load background data    \n",
    "    full_data = full_deepce_data[i].drop(full_deepce_data[i].columns[np.arange(0,5)], axis=1) \n",
    "    \n",
    "    # Split background dataset to test and train parts\n",
    "    lincs_train_df, lincs_test_df= train_test_split(full_data, test_size=test_set_percent, random_state=50)\n",
    "    \n",
    "    # Set encoder parameters\n",
    "    e_i = Input(shape=(features_nr, ), name='encoder_input')\n",
    "    encoded = layers.Dense(hidden_layer_dim, kernel_initializer=glorot_uniform(seed=123))(e_i)\n",
    "    encoded_batchnorm=BatchNormalization()(encoded)\n",
    "    encoded_activ = Activation('tanh')(encoded_batchnorm)\n",
    "    encoded1 = layers.Dense(latent_dim, kernel_initializer=glorot_uniform(seed=123))(encoded_activ)\n",
    "    encoded_batchnorm1=BatchNormalization()(encoded1)\n",
    "    l_s = Activation('tanh')(encoded_batchnorm1)\n",
    "    \n",
    "    # Instantiate encoder\n",
    "    encoder = Model(e_i, l_s, name='encoder')\n",
    "    encoder.summary()\n",
    "\n",
    "    # Set decoder parameters\n",
    "    d_i   = Input(shape=(latent_dim, ), name='decoder_input')\n",
    "    decoded0 = layers.Dense(hidden_layer_dim, kernel_initializer=glorot_uniform(seed=123))(d_i)\n",
    "    decoded_batchnorm=BatchNormalization()(decoded0)\n",
    "    decoded_activ = Activation('tanh')(decoded_batchnorm)\n",
    "    decoded1 = layers.Dense(features_nr, kernel_initializer=glorot_uniform(seed=123))(decoded_activ)\n",
    "    decoded_batchnorm1=BatchNormalization()(decoded1)\n",
    "    decoded = Activation('tanh')(decoded_batchnorm1)\n",
    "    \n",
    "    # Instantiate decoder\n",
    "    decoder = Model(d_i, decoded, name='decoder')\n",
    "    decoder.summary()\n",
    "\n",
    "    # Instantiate and compile autoencoder\n",
    "    autoencoder_outputs = 10*decoder(encoder(e_i))\n",
    "    autoencoder = Model(e_i, autoencoder_outputs, name='autoencoder')\n",
    "    autoencoder.summary()\n",
    "    autoencoder.compile(optimizer=optim, loss='mse',metrics=['mse',R2])\n",
    "\n",
    "    # Train autoencoder\n",
    "    H=autoencoder.fit(lincs_train_df,lincs_train_df,\n",
    "                epochs=num_epochs,\n",
    "                batch_size=size_batch,\n",
    "                shuffle=True,\n",
    "                validation_data=(lincs_test_df,lincs_test_df))\n",
    "   \n",
    "    # Assess autoencoder training\n",
    "    plt.figure()\n",
    "    plt.rc('font', size=SMALL_SIZE)          # controls default text sizes\n",
    "    plt.rc('axes', titlesize=SMALL_SIZE)     # fontsize of the axes title\n",
    "    plt.rc('axes', labelsize=MEDIUM_SIZE)    # fontsize of the x and y labels\n",
    "    plt.rc('xtick', labelsize=SMALL_SIZE)    # fontsize of the tick labels\n",
    "    plt.rc('ytick', labelsize=SMALL_SIZE)    # fontsize of the tick labels\n",
    "    plt.rc('legend', fontsize=SMALL_SIZE)    # legend fontsize\n",
    "    plt.rc('figure', titlesize=BIGGER_SIZE)  # fontsize of the figure title\n",
    "    plt.plot(H.history['loss'])\n",
    "    plt.plot(H.history['val_loss'])\n",
    "    plt.title(list_cell_lines_names[i])\n",
    "    plt.ylabel('loss')\n",
    "    plt.xlabel('epoch')\n",
    "    plt.legend(['train loss', 'validation loss'], loc='upper right');\n",
    "    plt.axis([0, 300, 0, 3])\n",
    "\n",
    "    plt.figure()\n",
    "    plt.plot(H.history['R2'])\n",
    "\n",
    "    plt.title(list_cell_lines_names[i])\n",
    "    plt.ylabel('R2')\n",
    "    plt.xlabel('epoch')\n",
    "    plt.legend(['R2'], loc='upper right');\n",
    "    plt.axis([0, 300, 0, 1])\n",
    "    plt.rc('font', size=SMALL_SIZE)          # controls default text sizes\n",
    "    plt.rc('axes', titlesize=SMALL_SIZE)     # fontsize of the axes title\n",
    "    plt.rc('axes', labelsize=MEDIUM_SIZE)    # fontsize of the x and y labels\n",
    "    plt.rc('xtick', labelsize=SMALL_SIZE)    # fontsize of the tick labels\n",
    "    plt.rc('ytick', labelsize=SMALL_SIZE)    # fontsize of the tick labels\n",
    "    plt.rc('legend', fontsize=SMALL_SIZE)    # legend fontsize\n",
    "    plt.rc('figure', titlesize=BIGGER_SIZE)  # fontsize of the figure title"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## PREDICTOR using Mean and Regression method"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\rudig\\Desktop\\Panos\\cross_tx_demo\\cross_tx_demo\\crosstx_functions_wind.py:52: FutureWarning: Using the level keyword in DataFrame and Series aggregations is deprecated and will be removed in a future version. Use groupby instead. df.median(level=1) should use df.groupby(level=1).median().\n",
      "  pred_final_mat=pd.concat(data_mat_nan_pd1).mean(skipna=True,level=0)\n"
     ]
    }
   ],
   "source": [
    "mean=[]\n",
    "regression=[]\n",
    "\n",
    "# Target cell line is MCF7 in this example\n",
    "for i1 in [0]:\n",
    "    \n",
    "    # Run Predictor using Mean method\n",
    "    mean.append(mean_method(data_mat_nan_pd,union_conc_val,drugload_indices,i1,numb_drugs))\n",
    "    \n",
    "    # Run Predictor using Regression method\n",
    "    regression.append(regression_method(data_mat_final_allexk,conc_final_allexk,data_mat_final,conc_final,i1,numb_genes,numb_drugs))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## CORRECTOR using PCA projection\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# preallocation\n",
    "indiv_drug = [[0]*numb_drugs for i in range(num_cell_lines)]\n",
    "all_drugs_minus_indiv_drug= [ [0]*numb_drugs for i in range(num_cell_lines)]\n",
    "pca_proj_mean=[[None]*numb_drugs for i in range(0,num_cell_lines)]\n",
    "pca_proj_regression=[[None]*numb_drugs for i in range(0,num_cell_lines)]\n",
    "\n",
    "# Set the number of nearest neighbors\n",
    "num_nn=5\n",
    "\n",
    "# Set explained variation threshold\n",
    "expl_var=.85     \n",
    "\n",
    "# Target cell line is MCF7 in this example\n",
    "for i in [0]:\n",
    "    \n",
    "    # Find non-empty drug indices for the target cell line\n",
    "    non_empty_cells_indices=[i2 for i2,x in enumerate(mean[i]) if x is not None]\n",
    "    \n",
    "    for i1 in non_empty_cells_indices:\n",
    "        \n",
    "        # Prepare background data (all the full data for the MCF7 cell line minus the data for the i1-th drug)\n",
    "        indiv_drug[i][i1]=ismember(full_deepce_data[i]['pert_id'],top100_drugs_list[i1])    \n",
    "        all_drugs_minus_indiv_drug[i][i1]= ~ indiv_drug[i][i1][0]        \n",
    "        data_mat_genes=full_deepce_data[i].values[:,np.arange(5,983)]    \n",
    "        bg_data=data_mat_genes[all_drugs_minus_indiv_drug[i][i1],:]\n",
    "\n",
    "        # Find nearest neighbors of the output of the Mean method\n",
    "        nn_pca_find_mean=find_nearest_neighbors(mean[i],bg_data,num_nn,i1)\n",
    "        \n",
    "        # Find nearest neighbors of the output of the Regression method\n",
    "        nn_pca_find_regession=find_nearest_neighbors(regression[i],bg_data,num_nn,i1)\n",
    "\n",
    "        # PCA projection of the output of the Mean Method (i.e., Mean+PCA method)         \n",
    "        pca_proj_mean[i][i1]=pca_projection(mean[i],nn_pca_find_mean,expl_var,i1)\n",
    "        \n",
    "        # PCA projection of the output of the Regression Method (i.e., Regression+PCA method)         \n",
    "        pca_proj_regression[i][i1]=pca_projection(regression[i],nn_pca_find_regession,expl_var,i1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## CORRECTOR using Autoencoder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\rudig\\anaconda3\\lib\\site-packages\\keras\\engine\\training.py:2401: UserWarning: `Model.state_updates` will be removed in a future version. This property should not be used in TensorFlow 2.0, as `updates` are applied automatically.\n",
      "  warnings.warn('`Model.state_updates` will be removed in a future version. '\n"
     ]
    }
   ],
   "source": [
    "# preallocation\n",
    "mean_ae=[[None]*numb_drugs for i in range(0,num_cell_lines)]\n",
    "regression_ae=[[None]*numb_drugs for i in range(0,num_cell_lines)]\n",
    "mean_pca_proj_ae=[[None]*numb_drugs for i in range(0,num_cell_lines)]\n",
    "regression_pca_proj_ae=[[None]*numb_drugs for i in range(0,num_cell_lines)]\n",
    "    \n",
    "# Target cell line is MCF7 in this example\n",
    "for i in [0]:\n",
    "        \n",
    "    # Find non-empty drug indices for the target cell line\n",
    "    non_empty_cells_indices1=[i2 for i2,x in enumerate(pca_proj_mean[i]) if  (x is not None) and (len(x)>0)]\n",
    "    \n",
    "    for i1 in non_empty_cells_indices1:\n",
    "        \n",
    "        # AE projection of the output of the Mean method (i.e., Mean+AE method)\n",
    "        mean_ae[i][i1]=autoencoder.predict(np.array(mean[i][i1].transpose()))\n",
    "        \n",
    "        # AE projection of the output of the Mean+PCA method (i.e., Mean+PCA+AE method)\n",
    "        mean_pca_proj_ae[i][i1]=autoencoder.predict(np.array(pca_proj_mean[i][i1]))\n",
    "        \n",
    "        # AE projection of the output of the Regression method (i.e., Regression+AE method)\n",
    "        regression_ae[i][i1]=autoencoder.predict(np.array(regression[i][i1].transpose()))\n",
    "        \n",
    "        # AE projection of the output of the Regression+PCA method (i.e., Regression+PCA+AE method)\n",
    "        regression_pca_proj_ae[i][i1]=autoencoder.predict(np.array(pca_proj_regression[i][i1]))\n",
    "            "
   ]
  }
 ],
 "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": "3.8.8"
  },
  "varInspector": {
   "cols": {
    "lenName": 16,
    "lenType": 16,
    "lenVar": 40
   },
   "kernels_config": {
    "python": {
     "delete_cmd_postfix": "",
     "delete_cmd_prefix": "del ",
     "library": "var_list.py",
     "varRefreshCmd": "print(var_dic_list())"
    },
    "r": {
     "delete_cmd_postfix": ") ",
     "delete_cmd_prefix": "rm(",
     "library": "var_list.r",
     "varRefreshCmd": "cat(var_dic_list()) "
    }
   },
   "types_to_exclude": [
    "module",
    "function",
    "builtin_function_or_method",
    "instance",
    "_Feature"
   ],
   "window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}