GraphNeuralNetwork_Spektral_logP-regression-from-SDF.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Tutorial based on the following Spektral example<br />\n",
    "https://github.com/danielegrattarola/spektral/blob/master/examples/classification_delaunay.py\n",
    "\n",
    "Data set source : \"JPlogP: an improved logP predictor trained using predicted data\"<br />\n",
    "https://jcheminf.biomedcentral.com/articles/10.1186/s13321-018-0316-5\n",
    "\n",
    "Conda environment<br />\n",
    "File conda-env_spektral.txt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n",
      "/home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:516: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\n",
      "/home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:517: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\n",
      "/home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:518: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\n",
      "/home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:519: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\n",
      "/home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:520: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\n",
      "/home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:525: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\n",
      "/home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:541: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint8 = np.dtype([(\"qint8\", np.int8, 1)])\n",
      "/home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:542: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint8 = np.dtype([(\"quint8\", np.uint8, 1)])\n",
      "/home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:543: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint16 = np.dtype([(\"qint16\", np.int16, 1)])\n",
      "/home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:544: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_quint16 = np.dtype([(\"quint16\", np.uint16, 1)])\n",
      "/home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:545: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  _np_qint32 = np.dtype([(\"qint32\", np.int32, 1)])\n",
      "/home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/tensorboard/compat/tensorflow_stub/dtypes.py:550: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.\n",
      "  np_resource = np.dtype([(\"resource\", np.ubyte, 1)])\n"
     ]
    }
   ],
   "source": [
    "# Imports\n",
    "import numpy as np\n",
    "\n",
    "import pandas as pd\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "from keras.layers import Input, Dense\n",
    "from keras.regularizers import l2\n",
    "from keras.models import Model\n",
    "from keras.optimizers import Adam\n",
    "from keras.callbacks import EarlyStopping\n",
    "\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import confusion_matrix\n",
    "\n",
    "from spektral.layers import GraphAttention, GlobalAttentionPool\n",
    "from spektral.utils import label_to_one_hot, load_sdf, nx_to_numpy\n",
    "from spektral.chem import sdf_to_nx"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Reading SDF\n"
     ]
    }
   ],
   "source": [
    "# Molecules loading in the SDF internal format of Spektral (not true SD but JSON like)\n",
    "# (5k molecules from the Delaunay et al. article, filtered to contain only CHNO atoms)\n",
    "sdf_loaded = load_sdf(\"data/Trainingset_Delaunay-CHNO5000selected.sdf\", amount=None)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "5000\n"
     ]
    }
   ],
   "source": [
    "# Conversion from SDF to nx.Graphs with Spektral\n",
    "sdf_nx = sdf_to_nx(sdf_loaded, keep_hydrogen=True)\n",
    "print(len(sdf_nx))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Extraction from nx.Graphs of the adjency matrices and the nodes attributes of the molecules\n",
    "sdf_adj, sdf_node, _ = nx_to_numpy(sdf_nx,\n",
    "                                   nf_keys=['atomic_num'],\n",
    "                                   ef_keys=['type']\n",
    "                                  )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Node attributes preprocessing\n",
    "uniq_node = np.unique([v for x in sdf_node for v in np.unique(x)])\n",
    "node = [label_to_one_hot(x, uniq_node) for x in sdf_node]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(5000, 2)\n"
     ]
    }
   ],
   "source": [
    "# Labels import and preprocessing\n",
    "label_pd = pd.read_csv(\"data/Trainingset_Delaunay-CHNO5000selected_label.csv\")\n",
    "label = label_pd.to_numpy()\n",
    "\n",
    "print(label.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of nodes in the graphs :  58\n",
      "Features dimenssionality : 4\n",
      "Number of classes : 2\n"
     ]
    }
   ],
   "source": [
    "# Model parameters\n",
    "N = node[0].shape[-2]          # Number of nodes in the graphs\n",
    "F = node[0].shape[-1]          # Original feature dimensionality\n",
    "n_classes = label.shape[-1]    # Number of classes\n",
    "\n",
    "l2_reg = 5e-4                  # Regularization rate for l2\n",
    "learning_rate = 1e-4           # Learning rate for Adam\n",
    "epochs = 50                     # Number of training epochs\n",
    "batch_size = 32                # Batch size\n",
    "es_patience = 5                # Patience for early stopping\n",
    "es_delta = 0.001                # Min Delta for early stopping\n",
    "\n",
    "print('Number of nodes in the graphs : ',N)\n",
    "print('Features dimenssionality :',F)\n",
    "print('Number of classes :',n_classes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Train/test split : 10%\n",
    "A_train, A_test, \\\n",
    "X_train, X_test, \\\n",
    "y_train, y_test = train_test_split(sdf_adj, node, label, test_size=0.1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py:66: The name tf.get_default_graph is deprecated. Please use tf.compat.v1.get_default_graph instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py:541: The name tf.placeholder is deprecated. Please use tf.compat.v1.placeholder instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py:4432: The name tf.random_uniform is deprecated. Please use tf.random.uniform instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py:148: The name tf.placeholder_with_default is deprecated. Please use tf.compat.v1.placeholder_with_default instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py:3733: calling dropout (from tensorflow.python.ops.nn_ops) with keep_prob is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use `rate` instead of `keep_prob`. Rate should be set to `rate = 1 - keep_prob`.\n"
     ]
    }
   ],
   "source": [
    "# Model definition\n",
    "X_in = Input(shape=(N, F))\n",
    "A_in = Input((N, N))\n",
    "\n",
    "gc1 = GraphAttention(32, activation='relu', kernel_regularizer=l2(l2_reg))([X_in, A_in])\n",
    "gc2 = GraphAttention(64, activation='relu', kernel_regularizer=l2(l2_reg))([gc1, A_in])\n",
    "pool = GlobalAttentionPool(128)(gc2)\n",
    "\n",
    "output = Dense(n_classes, activation='softmax')(pool)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/keras/optimizers.py:793: The name tf.train.Optimizer is deprecated. Please use tf.compat.v1.train.Optimizer instead.\n",
      "\n",
      "WARNING:tensorflow:From /home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py:3576: The name tf.log is deprecated. Please use tf.math.log instead.\n",
      "\n",
      "Model: \"model_1\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_1 (InputLayer)            (None, 58, 4)        0                                            \n",
      "__________________________________________________________________________________________________\n",
      "input_2 (InputLayer)            (None, 58, 58)       0                                            \n",
      "__________________________________________________________________________________________________\n",
      "graph_attention_1 (GraphAttenti (None, 58, 32)       224         input_1[0][0]                    \n",
      "                                                                 input_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "graph_attention_2 (GraphAttenti (None, 58, 64)       2240        graph_attention_1[0][0]          \n",
      "                                                                 input_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "global_attention_pool_1 (Global (None, 128)          16640       graph_attention_2[0][0]          \n",
      "__________________________________________________________________________________________________\n",
      "dense_1 (Dense)                 (None, 2)            258         global_attention_pool_1[0][0]    \n",
      "==================================================================================================\n",
      "Total params: 19,362\n",
      "Trainable params: 19,362\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# Build model\n",
    "model = Model(inputs=[X_in, A_in], outputs=output)\n",
    "optimizer = Adam(lr=learning_rate)\n",
    "model.compile(optimizer=optimizer, loss='categorical_crossentropy', metrics=['acc'])\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /home/lionel/anaconda3/envs/spektral/lib/python3.6/site-packages/tensorflow/python/ops/math_grad.py:1250: add_dispatch_support.<locals>.wrapper (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use tf.where in 2.0, which has the same broadcast rule as np.where\n",
      "Train on 4050 samples, validate on 450 samples\n",
      "Epoch 1/50\n",
      "4050/4050 [==============================] - 44s 11ms/step - loss: 0.4901 - acc: 0.7859 - val_loss: 0.3404 - val_acc: 0.8667\n",
      "Epoch 2/50\n",
      "4050/4050 [==============================] - 43s 11ms/step - loss: 0.3677 - acc: 0.8462 - val_loss: 0.3110 - val_acc: 0.8822\n",
      "Epoch 3/50\n",
      "4050/4050 [==============================] - 45s 11ms/step - loss: 0.3568 - acc: 0.8496 - val_loss: 0.3125 - val_acc: 0.8844\n",
      "Epoch 4/50\n",
      "4050/4050 [==============================] - 42s 10ms/step - loss: 0.3450 - acc: 0.8560 - val_loss: 0.2925 - val_acc: 0.8956\n",
      "Epoch 5/50\n",
      "4050/4050 [==============================] - 44s 11ms/step - loss: 0.3387 - acc: 0.8573 - val_loss: 0.3062 - val_acc: 0.8822\n",
      "Epoch 6/50\n",
      "4050/4050 [==============================] - 46s 11ms/step - loss: 0.3341 - acc: 0.8580 - val_loss: 0.2811 - val_acc: 0.9022\n",
      "Epoch 7/50\n",
      "4050/4050 [==============================] - 41s 10ms/step - loss: 0.3267 - acc: 0.8640 - val_loss: 0.2904 - val_acc: 0.8978\n",
      "Epoch 8/50\n",
      "4050/4050 [==============================] - 44s 11ms/step - loss: 0.3277 - acc: 0.8610 - val_loss: 0.2762 - val_acc: 0.8933\n",
      "Epoch 9/50\n",
      "4050/4050 [==============================] - 43s 11ms/step - loss: 0.3206 - acc: 0.8627 - val_loss: 0.2736 - val_acc: 0.8978\n",
      "Epoch 10/50\n",
      "4050/4050 [==============================] - 47s 12ms/step - loss: 0.3171 - acc: 0.8677 - val_loss: 0.2709 - val_acc: 0.8978\n",
      "Epoch 11/50\n",
      "4050/4050 [==============================] - 45s 11ms/step - loss: 0.3143 - acc: 0.8686 - val_loss: 0.2800 - val_acc: 0.8956\n",
      "Epoch 12/50\n",
      "4050/4050 [==============================] - 47s 12ms/step - loss: 0.3115 - acc: 0.8681 - val_loss: 0.2675 - val_acc: 0.8956\n",
      "Epoch 13/50\n",
      "4050/4050 [==============================] - 46s 11ms/step - loss: 0.3096 - acc: 0.8674 - val_loss: 0.2661 - val_acc: 0.8956\n",
      "Epoch 14/50\n",
      "4050/4050 [==============================] - 44s 11ms/step - loss: 0.3091 - acc: 0.8686 - val_loss: 0.2698 - val_acc: 0.8956\n",
      "Epoch 15/50\n",
      "4050/4050 [==============================] - 41s 10ms/step - loss: 0.3041 - acc: 0.8719 - val_loss: 0.2634 - val_acc: 0.8978\n",
      "Epoch 16/50\n",
      "4050/4050 [==============================] - 42s 10ms/step - loss: 0.3029 - acc: 0.8716 - val_loss: 0.2653 - val_acc: 0.8956\n",
      "Epoch 17/50\n",
      "4050/4050 [==============================] - 41s 10ms/step - loss: 0.3017 - acc: 0.8709 - val_loss: 0.2801 - val_acc: 0.8889\n",
      "Epoch 18/50\n",
      "4050/4050 [==============================] - 39s 10ms/step - loss: 0.3044 - acc: 0.8728 - val_loss: 0.2611 - val_acc: 0.8978\n",
      "Epoch 19/50\n",
      "4050/4050 [==============================] - 42s 10ms/step - loss: 0.2985 - acc: 0.8716 - val_loss: 0.2590 - val_acc: 0.8956\n",
      "Epoch 20/50\n",
      "4050/4050 [==============================] - 42s 10ms/step - loss: 0.2977 - acc: 0.8746 - val_loss: 0.2591 - val_acc: 0.8911\n",
      "Epoch 21/50\n",
      "4050/4050 [==============================] - 40s 10ms/step - loss: 0.2966 - acc: 0.8733 - val_loss: 0.2849 - val_acc: 0.8867\n",
      "Epoch 22/50\n",
      "4050/4050 [==============================] - 43s 11ms/step - loss: 0.2970 - acc: 0.8726 - val_loss: 0.2566 - val_acc: 0.8933\n",
      "Epoch 23/50\n",
      "4050/4050 [==============================] - 39s 10ms/step - loss: 0.2973 - acc: 0.8756 - val_loss: 0.2563 - val_acc: 0.8933\n",
      "Epoch 24/50\n",
      "4050/4050 [==============================] - 43s 11ms/step - loss: 0.2954 - acc: 0.8746 - val_loss: 0.2578 - val_acc: 0.8911\n",
      "Epoch 25/50\n",
      "4050/4050 [==============================] - 43s 11ms/step - loss: 0.2944 - acc: 0.8719 - val_loss: 0.2566 - val_acc: 0.8889\n",
      "Epoch 26/50\n",
      "4050/4050 [==============================] - 41s 10ms/step - loss: 0.3002 - acc: 0.8741 - val_loss: 0.2530 - val_acc: 0.8933\n",
      "Epoch 27/50\n",
      "4050/4050 [==============================] - 45s 11ms/step - loss: 0.2919 - acc: 0.8775 - val_loss: 0.2574 - val_acc: 0.8956\n",
      "Epoch 28/50\n",
      "4050/4050 [==============================] - 44s 11ms/step - loss: 0.2930 - acc: 0.8756 - val_loss: 0.2574 - val_acc: 0.9000\n",
      "Epoch 29/50\n",
      "4050/4050 [==============================] - 46s 11ms/step - loss: 0.2937 - acc: 0.8760 - val_loss: 0.2524 - val_acc: 0.9000\n",
      "Epoch 30/50\n",
      "4050/4050 [==============================] - 43s 11ms/step - loss: 0.2898 - acc: 0.8790 - val_loss: 0.2625 - val_acc: 0.8978\n",
      "Epoch 31/50\n",
      "4050/4050 [==============================] - 42s 10ms/step - loss: 0.2915 - acc: 0.8753 - val_loss: 0.2529 - val_acc: 0.9000\n",
      "Restoring model weights from the end of the best epoch\n",
      "Epoch 00031: early stopping\n"
     ]
    }
   ],
   "source": [
    "# Train model : 10% validation split\n",
    "history = model.fit([X_train, A_train],\n",
    "              y_train,\n",
    "              batch_size=batch_size,\n",
    "              validation_split=0.1,\n",
    "              epochs=epochs,\n",
    "              callbacks=[\n",
    "                  EarlyStopping(monitor='val_loss', min_delta=es_delta, patience=es_patience, verbose=1, restore_best_weights=True)\n",
    "              ])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dict_keys(['val_loss', 'val_acc', 'loss', 'acc'])\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "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": [
    "# list all data in history\n",
    "print(history.history.keys())\n",
    "# summarize history for accuracy\n",
    "plt.plot(history.history['acc'])\n",
    "plt.plot(history.history['val_acc'])\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'validation'], loc='upper left')\n",
    "plt.show()\n",
    "# summarize history for loss\n",
    "plt.plot(history.history['loss'])\n",
    "plt.plot(history.history['val_loss'])\n",
    "plt.title('model loss')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'validation'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Evaluating model.\n",
      "500/500 [==============================] - 2s 3ms/step\n",
      "Done.\n",
      "Test loss: 0.28709950852394106\n",
      "Test accuracy: 0.8699999990463256\n"
     ]
    }
   ],
   "source": [
    "# Evaluate model\n",
    "print('Evaluating model.')\n",
    "eval_results = model.evaluate([X_test, A_test],\n",
    "                              y_test,\n",
    "                              batch_size=batch_size)\n",
    "print('Done.\\n'\n",
    "      'Test loss: {}\\n'\n",
    "      'Test accuracy: {}'.format(*eval_results))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Predictions calculation\n",
    "y_test_pred = model.predict([X_test, A_test])\n",
    "\n",
    "\n",
    "y_test_classes = y_test.argmax(axis=-1)\n",
    "y_pred_classes = y_test_pred.argmax(axis=-1)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[306,  19],\n",
       "       [ 46, 129]])"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Matrix confusion\n",
    "y_test_classes = y_test.argmax(axis=-1)\n",
    "y_pred_classes = y_test_pred.argmax(axis=-1)\n",
    "confusion_matrix(y_test_classes, y_pred_classes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}