LSTMmolecule.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "fitting-spelling",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "from matplotlib import pyplot as plt\n",
    "\n",
    "from collections import OrderedDict\n",
    "from itertools import chain\n",
    "from sklearn.utils import shuffle\n",
    "\n",
    "from rdkit import Chem\n",
    "\n",
    "import tensorflow as tf\n",
    "from tensorflow.keras.models import Sequential, Model\n",
    "from tensorflow.keras.layers import Dense\n",
    "from tensorflow.keras.layers import Dropout\n",
    "from tensorflow.keras.layers import LSTM\n",
    "from keras.utils import np_utils"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "cooked-third",
   "metadata": {},
   "source": [
    "## Data Loading and Visualization"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "mobile-victoria",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "There are 100 molecules in total. \n",
      "\n",
      "This is the chemical structure of the 60th molecule: \n",
      "\n",
      "CCOC(C)=O\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<rdkit.Chem.rdchem.Mol at 0x2c3c89147b0>"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Load the data\n",
    "train_data = np.genfromtxt('small_smiles.csv', comments='@', dtype='U')\n",
    "n_chars = len(train_data)\n",
    "print(\"There are \" + str(len(train_data)) +\" molecules in total. \\n\" )\n",
    "\n",
    "# Pick a random molecule from the dataset to visualize\n",
    "random_molecule = np.random.randint(0, len(train_data)-1)\n",
    "\n",
    "# Note that this will only work if you installed the rdkit package!\n",
    "print(\"This is the chemical structure of the \" + str(random_molecule+1) +\"th molecule: \\n\" )\n",
    "print(str(train_data[random_molecule]))\n",
    "Chem.MolFromSmiles(train_data[random_molecule])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fresh-afghanistan",
   "metadata": {},
   "source": [
    "## Mapping the Characters and Integers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "assumed-difference",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Get the unique characters from the dataset\n",
    "unique_chars = sorted(list(OrderedDict.fromkeys(chain.from_iterable(train_data))))\n",
    "\n",
    "# create mapping of characters to integers and the reverse\n",
    "char_to_int = dict((c, i) for i, c in enumerate(unique_chars))\n",
    "int_to_char = dict((i, c) for i, c in enumerate(unique_chars))\n",
    "\n",
    "# add stop letter to dictionary\n",
    "char_to_int.update({\"|\" : len(char_to_int)})\n",
    "int_to_char.update({len(int_to_char) : \"|\"})\n",
    "\n",
    "# how many unique characters do we have?\n",
    "mapping_size = len(char_to_int)\n",
    "reverse_mapping_size = len(int_to_char)\n",
    "#print (\"Size of the character to integer dictionary is: \", mapping_size)\n",
    "#print (\"Size of the integer to character dictionary is: \", reverse_mapping_size)\n",
    "\n",
    "# get longest sequence\n",
    "embed = max([len(seq) for seq in train_data])\n",
    "#print (\"The longest sequence has \" + str(embed) + \" characters.\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "unlike-sending",
   "metadata": {},
   "source": [
    "## One Hot Encoding"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "determined-tracker",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The 1th molecule CCC=O is encoded as [5 5 5 4 7 8 8 8 8 8 8 8]\n",
      "The 2th molecule CN=CO is encoded as [5 6 4 5 7 8 8 8 8 8 8 8]\n",
      "The 3th molecule O=CCO is encoded as [7 4 5 5 7 8 8 8 8 8 8 8]\n",
      "The 4th molecule CC1CC1 is encoded as [5 5 3 5 5 3 8 8 8 8 8 8]\n",
      "The 5th molecule CC(C)C=O is encoded as [5 5 1 5 2 5 4 7 8 8 8 8]\n",
      "The 6th molecule CC(O)C=O is encoded as [5 5 1 7 2 5 4 7 8 8 8 8]\n",
      "The 7th molecule CN(C)C=O is encoded as [5 6 1 5 2 5 4 7 8 8 8 8]\n",
      "The 8th molecule CC(=O)CO is encoded as [5 5 1 4 7 2 5 7 8 8 8 8]\n",
      "The 9th molecule CCC(C)=O is encoded as [5 5 5 1 5 2 4 7 8 8 8 8]\n",
      "The 10th molecule CCC(=N)O is encoded as [5 5 5 1 4 6 2 7 8 8 8 8]\n",
      "The 11th molecule CN=C(C)O is encoded as [5 6 4 5 1 5 2 7 8 8 8 8]\n",
      "The 12th molecule CNC(=N)O is encoded as [5 6 5 1 4 6 2 7 8 8 8 8]\n",
      "The 13th molecule COC(C)=O is encoded as [5 7 5 1 5 2 4 7 8 8 8 8]\n",
      "The 14th molecule COC(=N)O is encoded as [5 7 5 1 4 6 2 7 8 8 8 8]\n",
      "The 15th molecule N=C(O)CO is encoded as [6 4 5 1 7 2 5 7 8 8 8 8]\n",
      "The 16th molecule NCC(=O)O is encoded as [6 5 5 1 4 7 2 7 8 8 8 8]\n",
      "The 17th molecule CC(C)CO is encoded as [5 5 1 5 2 5 7 8 8 8 8 8]\n",
      "The 18th molecule COC(C)C is encoded as [5 7 5 1 5 2 5 8 8 8 8 8]\n",
      "The 19th molecule CC1(C)CC1 is encoded as [5 5 3 1 5 2 5 5 3 8 8 8]\n",
      "The 20th molecule CC1(C)CO1 is encoded as [5 5 3 1 5 2 5 7 3 8 8 8]\n",
      "The 21th molecule CCN=COC=O is encoded as [5 5 6 4 5 7 5 4 7 8 8 8]\n",
      "The 22th molecule COC=NCC=O is encoded as [5 7 5 4 6 5 5 4 7 8 8 8]\n",
      "The 23th molecule C#CC#CCCC is encoded as [5 0 5 5 0 5 5 5 5 8 8 8]\n",
      "The 24th molecule CCCC#CC#N is encoded as [5 5 5 5 0 5 5 0 6 8 8 8]\n",
      "The 25th molecule C#CC#CCOC is encoded as [5 0 5 5 0 5 5 7 5 8 8 8]\n",
      "The 26th molecule COCC#CC#N is encoded as [5 7 5 5 0 5 5 0 6 8 8 8]\n",
      "The 27th molecule C#CC#CCCO is encoded as [5 0 5 5 0 5 5 5 7 8 8 8]\n",
      "The 28th molecule N#CC#CCCO is encoded as [6 0 5 5 0 5 5 5 7 8 8 8]\n",
      "The 29th molecule CCCC#CC=O is encoded as [5 5 5 5 0 5 5 4 7 8 8 8]\n",
      "The 30th molecule COCC#CC=O is encoded as [5 7 5 5 0 5 5 4 7 8 8 8]\n",
      "The 31th molecule CC(=O)CC#N is encoded as [5 5 1 4 7 2 5 5 0 6 8 8]\n",
      "The 32th molecule C#CCC(=N)O is encoded as [5 0 5 5 5 1 4 6 2 7 8 8]\n",
      "The 33th molecule N#CCC(=N)O is encoded as [6 0 5 5 5 1 4 6 2 7 8 8]\n",
      "The 34th molecule CC(=N)N=CO is encoded as [5 5 1 4 6 2 6 4 5 7 8 8]\n",
      "The 35th molecule CC(=N)OC=O is encoded as [5 5 1 4 6 2 7 5 4 7 8 8]\n",
      "The 36th molecule CC(=O)CC=O is encoded as [5 5 1 4 7 2 5 5 4 7 8 8]\n",
      "The 37th molecule CC(O)=NC=N is encoded as [5 5 1 7 2 4 6 5 4 6 8 8]\n",
      "The 38th molecule CC(=O)N=CO is encoded as [5 5 1 4 7 2 6 4 5 7 8 8]\n",
      "The 39th molecule CC(=O)OC=N is encoded as [5 5 1 4 7 2 7 5 4 6 8 8]\n",
      "The 40th molecule N=C(N)N=CO is encoded as [6 4 5 1 6 2 6 4 5 7 8 8]\n",
      "The 41th molecule N=C(O)CC=O is encoded as [6 4 5 1 7 2 5 5 4 7 8 8]\n",
      "The 42th molecule N=CNC(=N)O is encoded as [6 4 5 6 5 1 4 6 2 7 8 8]\n",
      "The 43th molecule N=C(O)N=CO is encoded as [6 4 5 1 7 2 6 4 5 7 8 8]\n",
      "The 44th molecule N=COC(=N)O is encoded as [6 4 5 7 5 1 4 6 2 7 8 8]\n",
      "The 45th molecule C#CCC(C)C is encoded as [5 0 5 5 5 1 5 2 5 8 8 8]\n",
      "The 46th molecule CC(C)CC#N is encoded as [5 5 1 5 2 5 5 0 6 8 8 8]\n",
      "The 47th molecule C#CCC(C)O is encoded as [5 0 5 5 5 1 5 2 7 8 8 8]\n",
      "The 48th molecule CC(O)CC#N is encoded as [5 5 1 7 2 5 5 0 6 8 8 8]\n",
      "The 49th molecule CN(C)CC#N is encoded as [5 6 1 5 2 5 5 0 6 8 8 8]\n",
      "The 50th molecule CC(C)CC=O is encoded as [5 5 1 5 2 5 5 4 7 8 8 8]\n",
      "The 51th molecule CC(C)N=CO is encoded as [5 5 1 5 2 6 4 5 7 8 8 8]\n",
      "The 52th molecule CC(C)OC=O is encoded as [5 5 1 5 2 7 5 4 7 8 8 8]\n",
      "The 53th molecule CC(O)CC=O is encoded as [5 5 1 7 2 5 5 4 7 8 8 8]\n",
      "The 54th molecule CN(C)CC=O is encoded as [5 6 1 5 2 5 5 4 7 8 8 8]\n",
      "The 55th molecule CC(=O)CCO is encoded as [5 5 1 4 7 2 5 5 7 8 8 8]\n",
      "The 56th molecule CCCC(C)=O is encoded as [5 5 5 5 1 5 2 4 7 8 8 8]\n",
      "The 57th molecule CCCC(=N)O is encoded as [5 5 5 5 1 4 6 2 7 8 8 8]\n",
      "The 58th molecule CCN=C(C)O is encoded as [5 5 6 4 5 1 5 2 7 8 8 8]\n",
      "The 59th molecule CCNC(=N)O is encoded as [5 5 6 5 1 4 6 2 7 8 8 8]\n",
      "The 60th molecule CCOC(C)=O is encoded as [5 5 7 5 1 5 2 4 7 8 8 8]\n",
      "The 61th molecule CCOC(=N)O is encoded as [5 5 7 5 1 4 6 2 7 8 8 8]\n",
      "The 62th molecule CNCC(=O)O is encoded as [5 6 5 5 1 4 7 2 7 8 8 8]\n",
      "The 63th molecule COCC(C)=O is encoded as [5 7 5 5 1 5 2 4 7 8 8 8]\n",
      "The 64th molecule COCC(=N)O is encoded as [5 7 5 5 1 4 6 2 7 8 8 8]\n",
      "The 65th molecule N=C(O)CCO is encoded as [6 4 5 1 7 2 5 5 7 8 8 8]\n",
      "The 66th molecule NCCC(=O)O is encoded as [6 5 5 5 1 4 7 2 7 8 8 8]\n",
      "The 67th molecule CC(C)CCO is encoded as [5 5 1 5 2 5 5 7 8 8 8 8]\n",
      "The 68th molecule CC(O)CCO is encoded as [5 5 1 7 2 5 5 7 8 8 8 8]\n",
      "The 69th molecule CCCC(C)C is encoded as [5 5 5 5 1 5 2 5 8 8 8 8]\n",
      "The 70th molecule CCCC(C)O is encoded as [5 5 5 5 1 5 2 7 8 8 8 8]\n",
      "The 71th molecule CCOC(C)C is encoded as [5 5 7 5 1 5 2 5 8 8 8 8]\n",
      "The 72th molecule COCC(C)C is encoded as [5 7 5 5 1 5 2 5 8 8 8 8]\n",
      "The 73th molecule COCC(C)O is encoded as [5 7 5 5 1 5 2 7 8 8 8 8]\n",
      "The 74th molecule CC(=O)C(C)=O is encoded as [5 5 1 4 7 2 5 1 5 2 4 7]\n",
      "The 75th molecule CC(=O)C(=N)N is encoded as [5 5 1 4 7 2 5 1 4 6 2 6]\n",
      "The 76th molecule CC(=O)C(N)=O is encoded as [5 5 1 4 7 2 5 1 6 2 4 7]\n",
      "The 77th molecule N=C(N)C(=O)O is encoded as [6 4 5 1 6 2 5 1 4 7 2 7]\n",
      "The 78th molecule NC(=O)C(N)=O is encoded as [6 5 1 4 7 2 5 1 6 2 4 7]\n",
      "The 79th molecule CC(=O)C(C)C is encoded as [5 5 1 4 7 2 5 1 5 2 5 8]\n",
      "The 80th molecule CC(C)C(=N)O is encoded as [5 5 1 5 2 5 1 4 6 2 7 8]\n",
      "The 81th molecule CC(N)C(=O)O is encoded as [5 5 1 6 2 5 1 4 7 2 7 8]\n",
      "The 82th molecule CC(=O)C(C)O is encoded as [5 5 1 4 7 2 5 1 5 2 7 8]\n",
      "The 83th molecule CC(O)C(=N)O is encoded as [5 5 1 7 2 5 1 4 6 2 7 8]\n",
      "The 84th molecule CC(=O)N(C)C is encoded as [5 5 1 4 7 2 6 1 5 2 5 8]\n",
      "The 85th molecule CN(C)C(=N)O is encoded as [5 6 1 5 2 5 1 4 6 2 7 8]\n",
      "The 86th molecule CC(C)C(C)C is encoded as [5 5 1 5 2 5 1 5 2 5 8 8]\n",
      "The 87th molecule CC(C)C(C)O is encoded as [5 5 1 5 2 5 1 5 2 7 8 8]\n",
      "The 88th molecule CC(O)C(C)O is encoded as [5 5 1 7 2 5 1 5 2 7 8 8]\n",
      "The 89th molecule CC1(C)CCC1 is encoded as [5 5 3 1 5 2 5 5 5 3 8 8]\n",
      "The 90th molecule CC1(C)CCO1 is encoded as [5 5 3 1 5 2 5 5 7 3 8 8]\n",
      "The 91th molecule CC1(C)COC1 is encoded as [5 5 3 1 5 2 5 7 5 3 8 8]\n",
      "The 92th molecule CC1(O)CCC1 is encoded as [5 5 3 1 7 2 5 5 5 3 8 8]\n",
      "The 93th molecule CC1(O)COC1 is encoded as [5 5 3 1 7 2 5 7 5 3 8 8]\n",
      "The 94th molecule CC1(C)CC1O is encoded as [5 5 3 1 5 2 5 5 3 7 8 8]\n",
      "The 95th molecule CC1(O)CC1O is encoded as [5 5 3 1 7 2 5 5 3 7 8 8]\n",
      "The 96th molecule CC1CC1(C)C is encoded as [5 5 3 5 5 3 1 5 2 5 8 8]\n",
      "The 97th molecule CC1CC1(C)O is encoded as [5 5 3 5 5 3 1 5 2 7 8 8]\n",
      "The 98th molecule CC1OC1(C)C is encoded as [5 5 3 7 5 3 1 5 2 5 8 8]\n",
      "The 99th molecule C#CC(=O)C=O is encoded as [5 0 5 5 1 4 7 2 5 4 7 8]\n",
      "The 100th molecule N#CC(=O)C=O is encoded as [6 0 5 5 1 4 7 2 5 4 7 8]\n"
     ]
    }
   ],
   "source": [
    "# Define the length of the sequence of learning \n",
    "seq_length = 1\n",
    "\n",
    "# Initialize both input/output data matrix as the stop letter \"|\"\n",
    "dataX = np.full((len(train_data), embed-seq_length+1), char_to_int[\"|\"])\n",
    "dataY = np.full((len(train_data), embed-seq_length+1), char_to_int[\"|\"])\n",
    "\n",
    "# Consider the next character in a molecule as the output of the current character\n",
    "for i, smile in enumerate(train_data):\n",
    "    for j, char in enumerate(smile):\n",
    "        dataX[i, j] = char_to_int[char]\n",
    "        if j > 0:\n",
    "            dataY[i, j-1] = char_to_int[char]\n",
    "    print(\"The \"+str(i+1)+\"th molecule \"+ str(smile) +\" is encoded as \"+str(dataX[i,:]))\n",
    "\n",
    "# Keras utility for one-hot encoding of the characters\n",
    "X = np_utils.to_categorical(dataX)\n",
    "Y = np_utils.to_categorical(dataY)\n",
    "\n",
    "# X and Y array should be (number of training samples, length of longest sequences (with end characters), length of dictionary)\n",
    "X, Y = shuffle(X, Y)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "equivalent-extreme",
   "metadata": {},
   "source": [
    "## LSTM Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "normal-complement",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "lstm (LSTM)                  (None, None, 256)         272384    \n",
      "_________________________________________________________________\n",
      "dropout (Dropout)            (None, None, 256)         0         \n",
      "_________________________________________________________________\n",
      "lstm_1 (LSTM)                (None, None, 256)         525312    \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, None, 256)         0         \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, None, 9)           2313      \n",
      "=================================================================\n",
      "Total params: 800,009\n",
      "Trainable params: 800,009\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n"
     ]
    }
   ],
   "source": [
    "# Create the model (simple 2 layer LSTM)\n",
    "model = Sequential()\n",
    "model.add(LSTM(256, input_shape=(None, X.shape[2]), return_sequences = True))\n",
    "model.add(Dropout(0.25))\n",
    "# model.add(LSTM(512, return_sequences = True))\n",
    "# model.add(Dropout(0.25))\n",
    "model.add(LSTM(256, return_sequences = True))\n",
    "model.add(Dropout(0.25))\n",
    "model.add(Dense(Y.shape[-1], activation='softmax'))\n",
    "print (model.summary())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "burning-estonia",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 100 samples\n",
      "Epoch 1/500\n",
      "100/100 [==============================] - 6s 60ms/sample - loss: 2.1993\n",
      "Epoch 2/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 2.1543\n",
      "Epoch 3/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 2.1043\n",
      "Epoch 4/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 2.0373\n",
      "Epoch 5/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.9404\n",
      "Epoch 6/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.8114\n",
      "Epoch 7/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.7080\n",
      "Epoch 8/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.7504\n",
      "Epoch 9/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.7878\n",
      "Epoch 10/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.7359\n",
      "Epoch 11/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.6753\n",
      "Epoch 12/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.6458\n",
      "Epoch 13/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.6627\n",
      "Epoch 14/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.6764\n",
      "Epoch 15/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.6772\n",
      "Epoch 16/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.6525\n",
      "Epoch 17/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.6173\n",
      "Epoch 18/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.5939\n",
      "Epoch 19/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.5888\n",
      "Epoch 20/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.5973\n",
      "Epoch 21/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.5893\n",
      "Epoch 22/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.5782\n",
      "Epoch 23/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.5576\n",
      "Epoch 24/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.5411\n",
      "Epoch 25/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.5397\n",
      "Epoch 26/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.5408\n",
      "Epoch 27/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.5314\n",
      "Epoch 28/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.5093\n",
      "Epoch 29/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4957\n",
      "Epoch 30/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4979\n",
      "Epoch 31/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4989\n",
      "Epoch 32/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4816\n",
      "Epoch 33/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4697\n",
      "Epoch 34/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4653\n",
      "Epoch 35/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4648\n",
      "Epoch 36/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4494\n",
      "Epoch 37/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4414\n",
      "Epoch 38/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4476\n",
      "Epoch 39/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4393\n",
      "Epoch 40/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4237\n",
      "Epoch 41/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4210\n",
      "Epoch 42/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4185\n",
      "Epoch 43/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4094\n",
      "Epoch 44/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4019\n",
      "Epoch 45/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.4061\n",
      "Epoch 46/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3974\n",
      "Epoch 47/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3930\n",
      "Epoch 48/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3843\n",
      "Epoch 49/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3739\n",
      "Epoch 50/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3801\n",
      "Epoch 51/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3718\n",
      "Epoch 52/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3543\n",
      "Epoch 53/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3565\n",
      "Epoch 54/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3540\n",
      "Epoch 55/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3475\n",
      "Epoch 56/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3453\n",
      "Epoch 57/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3412\n",
      "Epoch 58/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3336\n",
      "Epoch 59/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3312\n",
      "Epoch 60/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3275\n",
      "Epoch 61/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3215\n",
      "Epoch 62/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3238\n",
      "Epoch 63/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3201\n",
      "Epoch 64/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3119\n",
      "Epoch 65/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3101\n",
      "Epoch 66/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.3024\n",
      "Epoch 67/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2990\n",
      "Epoch 68/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2945\n",
      "Epoch 69/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2956\n",
      "Epoch 70/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2866\n",
      "Epoch 71/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2774\n",
      "Epoch 72/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2750\n",
      "Epoch 73/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2698\n",
      "Epoch 74/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2696\n",
      "Epoch 75/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2642\n",
      "Epoch 76/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2541\n",
      "Epoch 77/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2468\n",
      "Epoch 78/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2520\n",
      "Epoch 79/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2406\n",
      "Epoch 80/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2438\n",
      "Epoch 81/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2561\n",
      "Epoch 82/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2413\n",
      "Epoch 83/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2234\n",
      "Epoch 84/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2412\n",
      "Epoch 85/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2081\n",
      "Epoch 86/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2245\n",
      "Epoch 87/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.1997\n",
      "Epoch 88/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.2045\n",
      "Epoch 89/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.1890\n",
      "Epoch 90/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.1929\n",
      "Epoch 91/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.1776\n",
      "Epoch 92/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.1671\n",
      "Epoch 93/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.1637\n",
      "Epoch 94/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.1673\n",
      "Epoch 95/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.1413\n",
      "Epoch 96/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.1421\n",
      "Epoch 97/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.1353\n",
      "Epoch 98/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.1173\n",
      "Epoch 99/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.1228\n",
      "Epoch 100/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.1047\n",
      "Epoch 101/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.0986\n",
      "Epoch 102/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.0947\n",
      "Epoch 103/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.0801\n",
      "Epoch 104/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.0730\n",
      "Epoch 105/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.0588\n",
      "Epoch 106/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.0615\n",
      "Epoch 107/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.0549\n",
      "Epoch 108/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.0347\n",
      "Epoch 109/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.0349\n",
      "Epoch 110/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.0344\n",
      "Epoch 111/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.0136\n",
      "Epoch 112/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 1.0073\n",
      "Epoch 113/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.9929\n",
      "Epoch 114/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.9969\n",
      "Epoch 115/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.9879\n",
      "Epoch 116/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.9749\n",
      "Epoch 117/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.9787\n",
      "Epoch 118/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.9667\n",
      "Epoch 119/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.9570\n",
      "Epoch 120/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.9520\n",
      "Epoch 121/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.9316\n",
      "Epoch 122/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.9318\n",
      "Epoch 123/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.9237\n",
      "Epoch 124/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.9146\n",
      "Epoch 125/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.9107\n",
      "Epoch 126/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.8928\n",
      "Epoch 127/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.8810\n",
      "Epoch 128/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.8843\n",
      "Epoch 129/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.8835\n",
      "Epoch 130/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.8670\n",
      "Epoch 131/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.8731\n",
      "Epoch 132/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.8590\n",
      "Epoch 133/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.8582\n",
      "Epoch 134/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.8379\n",
      "Epoch 135/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.8339\n",
      "Epoch 136/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.8342\n",
      "Epoch 137/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.8285\n",
      "Epoch 138/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.8209\n",
      "Epoch 139/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.8131\n",
      "Epoch 140/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.8049\n",
      "Epoch 141/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.8150\n",
      "Epoch 142/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7980\n",
      "Epoch 143/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7949\n",
      "Epoch 144/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7819\n",
      "Epoch 145/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7858\n",
      "Epoch 146/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7658\n",
      "Epoch 147/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7821\n",
      "Epoch 148/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7631\n",
      "Epoch 149/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7664\n",
      "Epoch 150/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7470\n",
      "Epoch 151/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7591\n",
      "Epoch 152/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7496\n",
      "Epoch 153/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7402\n",
      "Epoch 154/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7410\n",
      "Epoch 155/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7282\n",
      "Epoch 156/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7314\n",
      "Epoch 157/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7169\n",
      "Epoch 158/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7206\n",
      "Epoch 159/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7337\n",
      "Epoch 160/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7095\n",
      "Epoch 161/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7081\n",
      "Epoch 162/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7054\n",
      "Epoch 163/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7030\n",
      "Epoch 164/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7013\n",
      "Epoch 165/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6982\n",
      "Epoch 166/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.7104\n",
      "Epoch 167/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6993\n",
      "Epoch 168/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6760\n",
      "Epoch 169/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6959\n",
      "Epoch 170/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6824\n",
      "Epoch 171/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6816\n",
      "Epoch 172/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6746\n",
      "Epoch 173/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6709\n",
      "Epoch 174/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6526\n",
      "Epoch 175/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6620\n",
      "Epoch 176/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6747\n",
      "Epoch 177/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6589\n",
      "Epoch 178/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6634\n",
      "Epoch 179/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6466\n",
      "Epoch 180/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6603\n",
      "Epoch 181/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6570\n",
      "Epoch 182/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6439\n",
      "Epoch 183/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6596\n",
      "Epoch 184/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6486\n",
      "Epoch 185/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6505\n",
      "Epoch 186/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6371\n",
      "Epoch 187/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6292\n",
      "Epoch 188/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6360\n",
      "Epoch 189/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6326\n",
      "Epoch 190/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6196\n",
      "Epoch 191/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6307\n",
      "Epoch 192/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6313\n",
      "Epoch 193/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6174\n",
      "Epoch 194/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6126\n",
      "Epoch 195/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6171\n",
      "Epoch 196/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6235\n",
      "Epoch 197/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6149\n",
      "Epoch 198/500\n",
      "100/100 [==============================] - 0s 2ms/sample - loss: 0.6198\n",
      "Epoch 199/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6160\n",
      "Epoch 200/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6203\n",
      "Epoch 201/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6134\n",
      "Epoch 202/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6102\n",
      "Epoch 203/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6083\n",
      "Epoch 204/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6072\n",
      "Epoch 205/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5976\n",
      "Epoch 206/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.6051\n",
      "Epoch 207/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5907\n",
      "Epoch 208/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5903\n",
      "Epoch 209/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5898\n",
      "Epoch 210/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5866\n",
      "Epoch 211/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5919\n",
      "Epoch 212/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5827\n",
      "Epoch 213/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5887\n",
      "Epoch 214/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5879\n",
      "Epoch 215/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5917\n",
      "Epoch 216/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5825\n",
      "Epoch 217/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5704\n",
      "Epoch 218/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5933\n",
      "Epoch 219/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5766\n",
      "Epoch 220/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5651\n",
      "Epoch 221/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5837\n",
      "Epoch 222/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5575\n",
      "Epoch 223/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5835\n",
      "Epoch 224/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5765\n",
      "Epoch 225/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5835\n",
      "Epoch 226/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5548\n",
      "Epoch 227/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5780\n",
      "Epoch 228/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5823\n",
      "Epoch 229/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5854\n",
      "Epoch 230/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5531\n",
      "Epoch 231/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5617\n",
      "Epoch 232/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5618\n",
      "Epoch 233/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5626\n",
      "Epoch 234/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5646\n",
      "Epoch 235/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5640\n",
      "Epoch 236/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5566\n",
      "Epoch 237/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5572\n",
      "Epoch 238/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5415\n",
      "Epoch 239/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5557\n",
      "Epoch 240/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5460\n",
      "Epoch 241/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5550\n",
      "Epoch 242/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5452\n",
      "Epoch 243/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5493\n",
      "Epoch 244/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5387\n",
      "Epoch 245/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5363\n",
      "Epoch 246/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5466\n",
      "Epoch 247/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5443\n",
      "Epoch 248/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5482\n",
      "Epoch 249/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5453\n",
      "Epoch 250/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5416\n",
      "Epoch 251/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5504\n",
      "Epoch 252/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5376\n",
      "Epoch 253/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5265\n",
      "Epoch 254/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5423\n",
      "Epoch 255/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5219\n",
      "Epoch 256/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5332\n",
      "Epoch 257/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5313\n",
      "Epoch 258/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5347\n",
      "Epoch 259/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5331\n",
      "Epoch 260/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5272\n",
      "Epoch 261/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5258\n",
      "Epoch 262/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5194\n",
      "Epoch 263/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5137\n",
      "Epoch 264/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5172\n",
      "Epoch 265/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5287\n",
      "Epoch 266/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5121\n",
      "Epoch 267/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5180\n",
      "Epoch 268/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5225\n",
      "Epoch 269/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5241\n",
      "Epoch 270/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5148\n",
      "Epoch 271/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5167\n",
      "Epoch 272/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5187\n",
      "Epoch 273/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5153\n",
      "Epoch 274/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5194\n",
      "Epoch 275/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5135\n",
      "Epoch 276/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5252\n",
      "Epoch 277/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5208\n",
      "Epoch 278/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5107\n",
      "Epoch 279/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5034\n",
      "Epoch 280/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5103\n",
      "Epoch 281/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5084\n",
      "Epoch 282/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5043\n",
      "Epoch 283/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5132\n",
      "Epoch 284/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5136\n",
      "Epoch 285/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5048\n",
      "Epoch 286/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4983\n",
      "Epoch 287/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5036\n",
      "Epoch 288/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4994\n",
      "Epoch 289/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5038\n",
      "Epoch 290/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.5001\n",
      "Epoch 291/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4921\n",
      "Epoch 292/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4939\n",
      "Epoch 293/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4953\n",
      "Epoch 294/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4923\n",
      "Epoch 295/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4912\n",
      "Epoch 296/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4844\n",
      "Epoch 297/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4961\n",
      "Epoch 298/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4998\n",
      "Epoch 299/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4827\n",
      "Epoch 300/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4975\n",
      "Epoch 301/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4971\n",
      "Epoch 302/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4843\n",
      "Epoch 303/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4973\n",
      "Epoch 304/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4971\n",
      "Epoch 305/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4852\n",
      "Epoch 306/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4804\n",
      "Epoch 307/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4848\n",
      "Epoch 308/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4752\n",
      "Epoch 309/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4919\n",
      "Epoch 310/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4941\n",
      "Epoch 311/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4875\n",
      "Epoch 312/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4866\n",
      "Epoch 313/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4820\n",
      "Epoch 314/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4857\n",
      "Epoch 315/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4841\n",
      "Epoch 316/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4829\n",
      "Epoch 317/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4794\n",
      "Epoch 318/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4737\n",
      "Epoch 319/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4736\n",
      "Epoch 320/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4689\n",
      "Epoch 321/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4824\n",
      "Epoch 322/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4673\n",
      "Epoch 323/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4723\n",
      "Epoch 324/500\n",
      "100/100 [==============================] - 0s 2ms/sample - loss: 0.4713\n",
      "Epoch 325/500\n",
      "100/100 [==============================] - 0s 2ms/sample - loss: 0.4753\n",
      "Epoch 326/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4630\n",
      "Epoch 327/500\n",
      "100/100 [==============================] - 0s 2ms/sample - loss: 0.4776\n",
      "Epoch 328/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4634\n",
      "Epoch 329/500\n",
      "100/100 [==============================] - 0s 2ms/sample - loss: 0.4719\n",
      "Epoch 330/500\n",
      "100/100 [==============================] - 0s 2ms/sample - loss: 0.4710\n",
      "Epoch 331/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4618\n",
      "Epoch 332/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4653\n",
      "Epoch 333/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4647\n",
      "Epoch 334/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4723\n",
      "Epoch 335/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4557\n",
      "Epoch 336/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4640\n",
      "Epoch 337/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4694\n",
      "Epoch 338/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4601\n",
      "Epoch 339/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4630\n",
      "Epoch 340/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4581\n",
      "Epoch 341/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4524\n",
      "Epoch 342/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4587\n",
      "Epoch 343/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4518\n",
      "Epoch 344/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4593\n",
      "Epoch 345/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4555\n",
      "Epoch 346/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4575\n",
      "Epoch 347/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4507\n",
      "Epoch 348/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4550\n",
      "Epoch 349/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4619\n",
      "Epoch 350/500\n",
      "100/100 [==============================] - 1s 7ms/sample - loss: 0.4518\n",
      "Epoch 351/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4499\n",
      "Epoch 352/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4582\n",
      "Epoch 353/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4578\n",
      "Epoch 354/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4528\n",
      "Epoch 355/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4551\n",
      "Epoch 356/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4449\n",
      "Epoch 357/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4530\n",
      "Epoch 358/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4441\n",
      "Epoch 359/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4388\n",
      "Epoch 360/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4489\n",
      "Epoch 361/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4576\n",
      "Epoch 362/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4522\n",
      "Epoch 363/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4404\n",
      "Epoch 364/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4380\n",
      "Epoch 365/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4306\n",
      "Epoch 366/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4412\n",
      "Epoch 367/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4435\n",
      "Epoch 368/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4342\n",
      "Epoch 369/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4367\n",
      "Epoch 370/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4478\n",
      "Epoch 371/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4327\n",
      "Epoch 372/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4348\n",
      "Epoch 373/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4368\n",
      "Epoch 374/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4480\n",
      "Epoch 375/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4364\n",
      "Epoch 376/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4356\n",
      "Epoch 377/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4374\n",
      "Epoch 378/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4424\n",
      "Epoch 379/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4534\n",
      "Epoch 380/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4290\n",
      "Epoch 381/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4344\n",
      "Epoch 382/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4199\n",
      "Epoch 383/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4361\n",
      "Epoch 384/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4225\n",
      "Epoch 385/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4319\n",
      "Epoch 386/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4155\n",
      "Epoch 387/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4172\n",
      "Epoch 388/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4194\n",
      "Epoch 389/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4314\n",
      "Epoch 390/500\n",
      "100/100 [==============================] - 0s 990us/sample - loss: 0.4276\n",
      "Epoch 391/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4306\n",
      "Epoch 392/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4280\n",
      "Epoch 393/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4376\n",
      "Epoch 394/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4249\n",
      "Epoch 395/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4435\n",
      "Epoch 396/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4255\n",
      "Epoch 397/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4231\n",
      "Epoch 398/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4246\n",
      "Epoch 399/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4242\n",
      "Epoch 400/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4304\n",
      "Epoch 401/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4272\n",
      "Epoch 402/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4259\n",
      "Epoch 403/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4224\n",
      "Epoch 404/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4238\n",
      "Epoch 405/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4175\n",
      "Epoch 406/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4211\n",
      "Epoch 407/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4247\n",
      "Epoch 408/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4141\n",
      "Epoch 409/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4234\n",
      "Epoch 410/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4296\n",
      "Epoch 411/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4210\n",
      "Epoch 412/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4347\n",
      "Epoch 413/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4192\n",
      "Epoch 414/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4216\n",
      "Epoch 415/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4144\n",
      "Epoch 416/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4160\n",
      "Epoch 417/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4180\n",
      "Epoch 418/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4220\n",
      "Epoch 419/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4192\n",
      "Epoch 420/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4075\n",
      "Epoch 421/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4173\n",
      "Epoch 422/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4107\n",
      "Epoch 423/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4161\n",
      "Epoch 424/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4255\n",
      "Epoch 425/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4181\n",
      "Epoch 426/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4114\n",
      "Epoch 427/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4216\n",
      "Epoch 428/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4223\n",
      "Epoch 429/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4015\n",
      "Epoch 430/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4141\n",
      "Epoch 431/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4135\n",
      "Epoch 432/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4197\n",
      "Epoch 433/500\n",
      "100/100 [==============================] - 0s 2ms/sample - loss: 0.4105\n",
      "Epoch 434/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4081\n",
      "Epoch 435/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4181\n",
      "Epoch 436/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4146\n",
      "Epoch 437/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4077\n",
      "Epoch 438/500\n",
      "100/100 [==============================] - 0s 2ms/sample - loss: 0.4113\n",
      "Epoch 439/500\n",
      "100/100 [==============================] - 0s 4ms/sample - loss: 0.4196\n",
      "Epoch 440/500\n",
      "100/100 [==============================] - 0s 3ms/sample - loss: 0.3981\n",
      "Epoch 441/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4099\n",
      "Epoch 442/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4139\n",
      "Epoch 443/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4098\n",
      "Epoch 444/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4021\n",
      "Epoch 445/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4092\n",
      "Epoch 446/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4201\n",
      "Epoch 447/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4152\n",
      "Epoch 448/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3953\n",
      "Epoch 449/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4131\n",
      "Epoch 450/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4071\n",
      "Epoch 451/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4076\n",
      "Epoch 452/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4101\n",
      "Epoch 453/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4137\n",
      "Epoch 454/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4083\n",
      "Epoch 455/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4031\n",
      "Epoch 456/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3971\n",
      "Epoch 457/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3999\n",
      "Epoch 458/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4012\n",
      "Epoch 459/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3993\n",
      "Epoch 460/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3929\n",
      "Epoch 461/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4037\n",
      "Epoch 462/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3984\n",
      "Epoch 463/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4033\n",
      "Epoch 464/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4167\n",
      "Epoch 465/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4021\n",
      "Epoch 466/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4140\n",
      "Epoch 467/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4188\n",
      "Epoch 468/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4044\n",
      "Epoch 469/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4005\n",
      "Epoch 470/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3909\n",
      "Epoch 471/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4075\n",
      "Epoch 472/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4043\n",
      "Epoch 473/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4002\n",
      "Epoch 474/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3964\n",
      "Epoch 475/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3914\n",
      "Epoch 476/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4065\n",
      "Epoch 477/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3952\n",
      "Epoch 478/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4121\n",
      "Epoch 479/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4052\n",
      "Epoch 480/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4005\n",
      "Epoch 481/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4082\n",
      "Epoch 482/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3977\n",
      "Epoch 483/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3945\n",
      "Epoch 484/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4046\n",
      "Epoch 485/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4048\n",
      "Epoch 486/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3997\n",
      "Epoch 487/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3877\n",
      "Epoch 488/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3984\n",
      "Epoch 489/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4102\n",
      "Epoch 490/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3931\n",
      "Epoch 491/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4114\n",
      "Epoch 492/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3974\n",
      "Epoch 493/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3867\n",
      "Epoch 494/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.4008\n",
      "Epoch 495/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3882\n",
      "Epoch 496/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3915\n",
      "Epoch 497/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3936\n",
      "Epoch 498/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3961\n",
      "Epoch 499/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3962\n",
      "Epoch 500/500\n",
      "100/100 [==============================] - 0s 1ms/sample - loss: 0.3927\n"
     ]
    }
   ],
   "source": [
    "# Compile the model\n",
    "model.compile(loss = 'categorical_crossentropy', optimizer='adam')\n",
    "\n",
    "# Fit the model\n",
    "history = model.fit(X, Y, epochs = 500, batch_size = 256, verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "parallel-theory",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history.history[\"loss\"], '-o', label=\"Loss\")\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "unavailable-stanley",
   "metadata": {},
   "source": [
    "### Compare the predictions with the training data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "coated-craft",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Average success rate on training set: 84.5 %\n"
     ]
    }
   ],
   "source": [
    "# Calculate predictions\n",
    "predictions = model.predict(X, verbose=0)\n",
    "\n",
    "# Compare to correct result\n",
    "train_res = np.argmax(Y,axis=2)-np.argmax(predictions,axis=2)\n",
    "\n",
    "# Count correct and incorrect predictions\n",
    "no_false = np.count_nonzero(train_res)\n",
    "no_true = len(Y)*embed-no_false\n",
    "print(\"Average success rate on training set: %s %%\" %str(np.round(100*no_true/(embed*len(Y)),2)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "handed-insured",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "C(O)C(=N)O|| C(=)C(=N)O||\n",
      "C1(C)CC1|||| C((C)CC1||||\n",
      "C1(C)CC1O||| C((C)CC1||||\n",
      "OC(C)=O||||| CCCC)=O|||||\n",
      "C(=O)C(=N)N| C(=O)C(CN)N|\n",
      "#CCC(C)C|||| CCC#(C)C||||\n",
      "OCC#CC#N|||| CCC(CC#N||||\n",
      "#CCC(=N)O||| CCC#(CN)O|||\n",
      "=C(N)C(=O)O| =C(O)C(=O)O|\n",
      "C(=O)C(C)C|| C(=O)C(C)=||\n"
     ]
    }
   ],
   "source": [
    "# Compare the model predictions and the \"ground truth\" training set side by sidie\n",
    "\n",
    "max_display = min(10, len(train_data))   # can be a pre-defined value, say 10 molecules at most\n",
    "\n",
    "for i in range(max_display):\n",
    "    v = model.predict(X[i:i+1]) \n",
    "    idxs = np.argmax(v, axis=2)\n",
    "    pred=  \"\".join([int_to_char[h] for h in idxs[0]])\n",
    "    \n",
    "    # Note that here we use the argmax and do not sample using the model output\n",
    "    idxs2 = np.argmax(Y[i:i+1], axis=2)\n",
    "    true =  \"\".join([int_to_char[k] for k in idxs2[0]])\n",
    "    if true != pred:\n",
    "        print (true, pred)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "surrounded-sympathy",
   "metadata": {},
   "source": [
    "## Molecule Generation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "aboriginal-transport",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "N=\n",
      "N=C\n",
      "N=CC\n",
      "N=CCN\n",
      "N=CCNC\n",
      "N=CCNCC\n",
      "N=CCNCCC\n",
      "N=CCNCCCC\n",
      "N=CCNCCCCC\n",
      "N=CCNCCCCCC\n",
      "N=CCNCCCCCCC\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<rdkit.Chem.rdchem.Mol at 0x2c3df705170>"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Define a given starting seed character of the molecule\n",
    "seed = 'N'\n",
    "\n",
    "# Set the size to be the same as the training set\n",
    "size = Y.shape[-1]\n",
    "\n",
    "# Initialize array of probabilities and character\n",
    "pred = []\n",
    "preds = []\n",
    "next_char = []\n",
    "\n",
    "molecule = seed\n",
    "molecule_n = np.zeros((1,1,size))\n",
    "molecule_n[0, 0, char_to_int[seed]] = 1\n",
    " \n",
    "# Continue as long as we do not add '|'\n",
    "\n",
    "while next_char !='|' and len(molecule) < embed:    \n",
    "    # use the model to predict the probability of the next character\n",
    "    pred = model.predict(np.expand_dims(molecule_n[:,-1,:], axis = 1)).flatten()\n",
    "    preds.append(pred) \n",
    "    \n",
    "    # generate the next character with the highest probability\n",
    "#    index = np.argmax(pred).astype(int)\n",
    "    # generate the next character \"randomly\" based on the predicted probability         \n",
    "    index = np.random.choice(np.arange(0,len(char_to_int)), size=None, replace=True, p=pred)\n",
    "    next_char = int_to_char[index]      \n",
    "    molecule += next_char\n",
    "    print(molecule)\n",
    "    \n",
    "    molecule_new = np.zeros((1, 1, size))\n",
    "    molecule_new[0, 0, char_to_int[next_char]] = 1\n",
    "    molecule_n = np.append(molecule_n, molecule_new, axis=1)     \n",
    "    \n",
    "preds = np.array(preds)\n",
    "Chem.MolFromSmiles(molecule)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "outer-clock",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# As the sequence gets longer, the probability of '|' changes.\n",
    "plt.bar(range(len(preds[:,-1])), preds[:,-1], color='g')\n",
    "plt.xlabel('Sequence length')\n",
    "plt.ylabel('Probability')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "perceived-surprise",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot probability distributions over the sequence\n",
    "fig, ax = plt.subplots()\n",
    "\n",
    "for i in range(len(preds)):\n",
    "    ax.plot(preds[i,:], 'c', alpha=min(i * 0.1, 1))       \n",
    "\n",
    "ax.set_xticks(np.arange(0,len(char_to_int)))    \n",
    "ax.set_xticklabels(list(char_to_int.keys()))\n",
    "plt.ylabel('Probability')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "bottom-midwest",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}