From 084f01974c15140fa84a76d4b531d9791b84ee1c Mon Sep 17 00:00:00 2001
From: talyanskaya <49466926+talyanskaya@users.noreply.github.com>
Date: Tue, 11 Jun 2019 00:14:50 +0300
Subject: [PATCH] Add files via upload

---
 Talyanskaya/Data collection.ipynb |   40 ++
 Talyanskaya/Link to model.txt     |    1 +
 Talyanskaya/Training model.ipynb  | 1076 +++++++++++++++++++++++++++++
 Talyanskaya/predictImages.py      |   67 ++
 Talyanskaya/requirements.txt      |   78 +++
 5 files changed, 1262 insertions(+)
 create mode 100644 Talyanskaya/Data collection.ipynb
 create mode 100644 Talyanskaya/Link to model.txt
 create mode 100644 Talyanskaya/Training model.ipynb
 create mode 100644 Talyanskaya/predictImages.py
 create mode 100644 Talyanskaya/requirements.txt

diff --git a/Talyanskaya/Data collection.ipynb b/Talyanskaya/Data collection.ipynb
new file mode 100644
index 0000000..3e1be8a
--- /dev/null
+++ b/Talyanskaya/Data collection.ipynb	
@@ -0,0 +1,40 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Data collection\n",
+    "As it was allowed to use common dataset for the whole group, the duties were ditributed in accordance with each others capabilities.\n",
+    "All technical part of data collection was processed by Alexandr Marinsky and Anton Anisimov (they used a little bit different approach and their code is available in their repositories), but the final cleaning of dataset was conducted by the forces of the rest of the group who manually checked and deleted all inappropriate images. Many thanks to all who took part in that.\n",
+    "\n",
+    "In general, dataset consisted of almost 50 000 images of two classes of approximately the same size: \n",
+    "- iphones (all existing models since 3rd)\n",
+    "- non-iphones (all other phone trademarks ditrubuted in accordance with their popularity)\n",
+    "\n",
+    "All the images were downloaded from Avito.ru website as a it's the largest source of user-generated content in Russian segment of Internet."
+   ]
+  }
+ ],
+ "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": "3.6.3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}
diff --git a/Talyanskaya/Link to model.txt b/Talyanskaya/Link to model.txt
new file mode 100644
index 0000000..8fabc47
--- /dev/null
+++ b/Talyanskaya/Link to model.txt	
@@ -0,0 +1 @@
+https://drive.google.com/file/d/1E0BhElLCY2BEqd2BCqA3RdCqaVP7_KN1/view?usp=sharing
\ No newline at end of file
diff --git a/Talyanskaya/Training model.ipynb b/Talyanskaya/Training model.ipynb
new file mode 100644
index 0000000..594951a
--- /dev/null
+++ b/Talyanskaya/Training model.ipynb	
@@ -0,0 +1,1076 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Training model\n",
+    "\n",
+    "Note: I have tried several models, starting with handmade 8-layer NN, ending with Resnet18, used by my colleagues, however, all of them failed to perform for different reasons (died VMs, some overtraining, other unknown problems). I do admit using different sources in the web as well as a help of my colleague Alexandr Marinsky and other people, thanks to them.\n",
+    "\n",
+    "Finally, I ended with this IncenptionV3-based model with some changes as this model showed not very high but stable results. Originally, this model comes from standard Keras models and was trained on well-known Imagenet datadase. \n",
+    "The algorithm is rather simple: traing the \"Inception part\" of NN on our dataset, when save the results as Numpy-arrays, train with this Numpy-arrays the upper layer and than complile and tune model with blicked inner layers."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {
+    "scrolled": true
+   },
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Using TensorFlow backend.\n"
+     ]
+    }
+   ],
+   "source": [
+    "from keras.preprocessing.image import ImageDataGenerator\n",
+    "from keras.models import Sequential, Model\n",
+    "from keras.applications.inception_v3 import InceptionV3\n",
+    "from keras.callbacks import ModelCheckpoint\n",
+    "from keras.optimizers import SGD\n",
+    "\n",
+    "from keras import backend as K\n",
+    "K.set_image_dim_ordering('th')\n",
+    "\n",
+    "import numpy as np\n",
+    "import pandas as pd\n",
+    "import h5py"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "inc_model=InceptionV3(include_top=False, \n",
+    "                      weights='imagenet', \n",
+    "                      input_shape=(3, 150, 150))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 126,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import zipfile\n",
+    "\n",
+    "local_zip = '//jet/prs/workspace/Dataset_iphone_new.zip'\n",
+    "zip_ref = zipfile.ZipFile(local_zip, 'r')\n",
+    "zip_ref.extractall('/home/talyanskaya_marina/Dataset_iphone_new')\n",
+    "zip_ref.close()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Found 24000 images belonging to 2 classes.\n",
+      "Found 24000 images belonging to 2 classes.\n"
+     ]
+    }
+   ],
+   "source": [
+    "bottleneck_datagen = ImageDataGenerator(rescale=1./255)\n",
+    "    \n",
+    "train_generator = bottleneck_datagen.flow_from_directory('/home/talyanskaya_marina/Dataset_iphone_new/Dataset_iphone_new/train',\n",
+    "                                        target_size=(150, 150),\n",
+    "                                        batch_size=32,\n",
+    "                                        class_mode=None,\n",
+    "                                        shuffle=False)\n",
+    "\n",
+    "validation_generator = bottleneck_datagen.flow_from_directory('/home/talyanskaya_marina/Dataset_iphone_new/Dataset_iphone_new/validation',\n",
+    "                                                               target_size=(150, 150),\n",
+    "                                                               batch_size=32,\n",
+    "                                                               class_mode=None,\n",
+    "                                                               shuffle=False)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os\n",
+    "import shutil"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "try:\n",
+    "    os.mkdir('/home/talyanskaya_marina/bottleneck_features')\n",
+    "except OSError:\n",
+    "    shutil.rmtree('/home/talyanskaya_marina/bottleneck_features')\n",
+    "    os.mkdir('/home/talyanskaya_marina/bottleneck_features')\n",
+    "\n",
+    "bottleneck_features_train = inc_model.predict_generator(train_generator, steps = 24000/32)\n",
+    "np.save(open('/home/talyanskaya_marina/bottleneck_features/bn_features_train.npy', 'wb+'), bottleneck_features_train)\n",
+    "bottleneck_features_validation = inc_model.predict_generator(validation_generator, steps = 24000/32)\n",
+    "np.save(open('/home/talyanskaya_marina/bottleneck_features/bn_features_validation.npy', 'wb+'), bottleneck_features_validation)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "train_data = np.load(open('/home/talyanskaya_marina/bottleneck_features/bn_features_train.npy', 'rb'))\n",
+    "train_labels = np.array([0] * 12000 + [1] * 12000) \n",
+    "\n",
+    "validation_data = np.load(open('/home/talyanskaya_marina/bottleneck_features/bn_features_validation.npy', 'rb'))\n",
+    "validation_labels = np.array([0] * 12000 + [1] * 12000)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from keras.layers.convolutional import Conv2D, MaxPooling2D\n",
+    "from keras.models import Sequential\n",
+    "from keras.layers import Dense, Dropout, Flatten"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 9,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "fc_model = Sequential()\n",
+    "fc_model.add(Flatten(input_shape=train_data.shape[1:]))\n",
+    "fc_model.add(Dense(64, activation='relu', name='dense_one'))\n",
+    "fc_model.add(Dropout(0.5, name='dropout_one'))\n",
+    "fc_model.add(Dense(64, activation='relu', name='dense_two'))\n",
+    "fc_model.add(Dropout(0.5, name='dropout_two'))\n",
+    "fc_model.add(Dense(1, activation='sigmoid', name='output'))\n",
+    "\n",
+    "fc_model.compile(optimizer='rmsprop', \n",
+    "              loss='binary_crossentropy', \n",
+    "              metrics=['accuracy'])"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 10,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "try:\n",
+    "    os.mkdir('/home/talyanskaya_marina/bottleneck_features_and_weights/')\n",
+    "except OSError:\n",
+    "    shutil.rmtree('/home/talyanskaya_marina/bottleneck_features_and_weights/')\n",
+    "    os.mkdir('/home/talyanskaya_marina/bottleneck_features_and_weights/')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 11,
+   "metadata": {
+    "scrolled": true
+   },
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/jet/var/python/lib/python3.6/site-packages/ipykernel_launcher.py:3: UserWarning: The `nb_epoch` argument in `fit` has been renamed `epochs`.\n",
+      "  This is separate from the ipykernel package so we can avoid doing imports until\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Train on 24000 samples, validate on 24000 samples\n",
+      "Epoch 1/50\n",
+      "24000/24000 [==============================] - 9s 359us/step - loss: 1.1693 - acc: 0.5064 - val_loss: 0.6886 - val_acc: 0.5092\n",
+      "Epoch 2/50\n",
+      "24000/24000 [==============================] - 8s 315us/step - loss: 0.7021 - acc: 0.5082 - val_loss: 0.6766 - val_acc: 0.5526\n",
+      "Epoch 3/50\n",
+      "24000/24000 [==============================] - 8s 314us/step - loss: 0.7030 - acc: 0.5096 - val_loss: 0.6861 - val_acc: 0.5312\n",
+      "Epoch 4/50\n",
+      "24000/24000 [==============================] - 8s 332us/step - loss: 0.7024 - acc: 0.5175 - val_loss: 0.6780 - val_acc: 0.5518\n",
+      "Epoch 5/50\n",
+      "24000/24000 [==============================] - 8s 335us/step - loss: 0.6988 - acc: 0.5293 - val_loss: 0.6633 - val_acc: 0.5620\n",
+      "Epoch 6/50\n",
+      "24000/24000 [==============================] - 7s 310us/step - loss: 0.6852 - acc: 0.5499 - val_loss: 0.6066 - val_acc: 0.5982\n",
+      "Epoch 7/50\n",
+      "24000/24000 [==============================] - 8s 314us/step - loss: 0.6269 - acc: 0.5981 - val_loss: 0.5505 - val_acc: 0.7205\n",
+      "Epoch 8/50\n",
+      "24000/24000 [==============================] - 8s 315us/step - loss: 0.5859 - acc: 0.6480 - val_loss: 0.5463 - val_acc: 0.7182\n",
+      "Epoch 9/50\n",
+      "24000/24000 [==============================] - 7s 312us/step - loss: 0.5437 - acc: 0.6846 - val_loss: 0.5141 - val_acc: 0.7465\n",
+      "Epoch 10/50\n",
+      "24000/24000 [==============================] - 8s 316us/step - loss: 0.5200 - acc: 0.7010 - val_loss: 0.4996 - val_acc: 0.7474\n",
+      "Epoch 11/50\n",
+      "24000/24000 [==============================] - 8s 313us/step - loss: 0.4992 - acc: 0.7110 - val_loss: 0.5119 - val_acc: 0.7413\n",
+      "Epoch 12/50\n",
+      "24000/24000 [==============================] - 8s 315us/step - loss: 0.4856 - acc: 0.7175 - val_loss: 0.4963 - val_acc: 0.7568\n",
+      "Epoch 13/50\n",
+      "24000/24000 [==============================] - 8s 316us/step - loss: 0.4790 - acc: 0.7211 - val_loss: 0.4861 - val_acc: 0.7515\n",
+      "Epoch 14/50\n",
+      "24000/24000 [==============================] - 8s 314us/step - loss: 0.4767 - acc: 0.7253 - val_loss: 0.4939 - val_acc: 0.7565\n",
+      "Epoch 15/50\n",
+      "24000/24000 [==============================] - 8s 316us/step - loss: 0.4654 - acc: 0.7290 - val_loss: 0.4920 - val_acc: 0.7580\n",
+      "Epoch 16/50\n",
+      "24000/24000 [==============================] - 8s 317us/step - loss: 0.4569 - acc: 0.7361 - val_loss: 0.4903 - val_acc: 0.7572\n",
+      "Epoch 17/50\n",
+      "24000/24000 [==============================] - 8s 318us/step - loss: 0.4520 - acc: 0.7333 - val_loss: 0.4845 - val_acc: 0.7648\n",
+      "Epoch 18/50\n",
+      "24000/24000 [==============================] - 8s 315us/step - loss: 0.4482 - acc: 0.7382 - val_loss: 0.5104 - val_acc: 0.7603\n",
+      "Epoch 19/50\n",
+      "24000/24000 [==============================] - 8s 314us/step - loss: 0.4467 - acc: 0.7360 - val_loss: 0.5332 - val_acc: 0.7476\n",
+      "Epoch 20/50\n",
+      "24000/24000 [==============================] - 8s 317us/step - loss: 0.4431 - acc: 0.7361 - val_loss: 0.4996 - val_acc: 0.7609\n",
+      "Epoch 21/50\n",
+      "24000/24000 [==============================] - 7s 308us/step - loss: 0.4422 - acc: 0.7422 - val_loss: 0.4990 - val_acc: 0.7588\n",
+      "Epoch 22/50\n",
+      "24000/24000 [==============================] - 8s 317us/step - loss: 0.4398 - acc: 0.7377 - val_loss: 0.5043 - val_acc: 0.7545\n",
+      "Epoch 23/50\n",
+      "24000/24000 [==============================] - 8s 315us/step - loss: 0.4352 - acc: 0.7577 - val_loss: 0.4944 - val_acc: 0.7608\n",
+      "Epoch 24/50\n",
+      "24000/24000 [==============================] - 8s 318us/step - loss: 0.4251 - acc: 0.7613 - val_loss: 0.5256 - val_acc: 0.7556\n",
+      "Epoch 25/50\n",
+      "24000/24000 [==============================] - 7s 312us/step - loss: 0.4244 - acc: 0.7680 - val_loss: 0.5180 - val_acc: 0.7597\n",
+      "Epoch 26/50\n",
+      "24000/24000 [==============================] - 8s 319us/step - loss: 0.4189 - acc: 0.7709 - val_loss: 0.5167 - val_acc: 0.7631\n",
+      "Epoch 27/50\n",
+      "24000/24000 [==============================] - 8s 315us/step - loss: 0.4203 - acc: 0.7712 - val_loss: 0.5160 - val_acc: 0.7606\n",
+      "Epoch 28/50\n",
+      "24000/24000 [==============================] - 8s 316us/step - loss: 0.4111 - acc: 0.7780 - val_loss: 0.5377 - val_acc: 0.7627\n",
+      "Epoch 29/50\n",
+      "24000/24000 [==============================] - 8s 314us/step - loss: 0.4109 - acc: 0.7800 - val_loss: 0.5652 - val_acc: 0.7664\n",
+      "Epoch 30/50\n",
+      "24000/24000 [==============================] - 8s 316us/step - loss: 0.4089 - acc: 0.7808 - val_loss: 0.5344 - val_acc: 0.7570\n",
+      "Epoch 31/50\n",
+      "24000/24000 [==============================] - 8s 313us/step - loss: 0.4074 - acc: 0.7869 - val_loss: 0.5374 - val_acc: 0.7595\n",
+      "Epoch 32/50\n",
+      "24000/24000 [==============================] - 8s 317us/step - loss: 0.4029 - acc: 0.7855 - val_loss: 0.6612 - val_acc: 0.7658\n",
+      "Epoch 33/50\n",
+      "24000/24000 [==============================] - 8s 315us/step - loss: 0.3991 - acc: 0.7882 - val_loss: 0.6722 - val_acc: 0.7573\n",
+      "Epoch 34/50\n",
+      "24000/24000 [==============================] - 8s 314us/step - loss: 0.3963 - acc: 0.7907 - val_loss: 0.6105 - val_acc: 0.7577\n",
+      "Epoch 35/50\n",
+      "24000/24000 [==============================] - 8s 316us/step - loss: 0.3939 - acc: 0.7902 - val_loss: 0.5849 - val_acc: 0.7646\n",
+      "Epoch 36/50\n",
+      "24000/24000 [==============================] - 8s 326us/step - loss: 0.3891 - acc: 0.7923 - val_loss: 0.7010 - val_acc: 0.7629\n",
+      "Epoch 37/50\n",
+      "24000/24000 [==============================] - 8s 317us/step - loss: 0.3886 - acc: 0.7943 - val_loss: 0.6281 - val_acc: 0.7662\n",
+      "Epoch 38/50\n",
+      "24000/24000 [==============================] - 8s 317us/step - loss: 0.3764 - acc: 0.7992 - val_loss: 0.6210 - val_acc: 0.7489\n",
+      "Epoch 39/50\n",
+      "24000/24000 [==============================] - 8s 317us/step - loss: 0.3810 - acc: 0.7967 - val_loss: 0.6037 - val_acc: 0.7420\n",
+      "Epoch 40/50\n",
+      "24000/24000 [==============================] - 8s 318us/step - loss: 0.3807 - acc: 0.8000 - val_loss: 0.6978 - val_acc: 0.7586\n",
+      "Epoch 41/50\n",
+      "24000/24000 [==============================] - 8s 316us/step - loss: 0.3782 - acc: 0.7996 - val_loss: 0.5917 - val_acc: 0.7577\n",
+      "Epoch 42/50\n",
+      "24000/24000 [==============================] - 8s 314us/step - loss: 0.3814 - acc: 0.7963 - val_loss: 0.7144 - val_acc: 0.7621\n",
+      "Epoch 43/50\n",
+      "24000/24000 [==============================] - 8s 316us/step - loss: 0.3806 - acc: 0.7942 - val_loss: 0.6483 - val_acc: 0.7520\n",
+      "Epoch 44/50\n",
+      "24000/24000 [==============================] - 7s 304us/step - loss: 0.3775 - acc: 0.8007 - val_loss: 0.7409 - val_acc: 0.7648\n",
+      "Epoch 45/50\n",
+      "24000/24000 [==============================] - 8s 333us/step - loss: 0.3743 - acc: 0.8003 - val_loss: 0.5713 - val_acc: 0.7562\n",
+      "Epoch 46/50\n",
+      "24000/24000 [==============================] - 8s 334us/step - loss: 0.3717 - acc: 0.8038 - val_loss: 0.6715 - val_acc: 0.7549\n",
+      "Epoch 47/50\n",
+      "24000/24000 [==============================] - 8s 318us/step - loss: 0.3713 - acc: 0.8055 - val_loss: 0.6999 - val_acc: 0.7581\n",
+      "Epoch 48/50\n",
+      "24000/24000 [==============================] - 8s 318us/step - loss: 0.3752 - acc: 0.8033 - val_loss: 0.7368 - val_acc: 0.7597\n",
+      "Epoch 49/50\n",
+      "24000/24000 [==============================] - 7s 312us/step - loss: 0.3629 - acc: 0.8091 - val_loss: 0.7648 - val_acc: 0.7570\n",
+      "Epoch 50/50\n",
+      "24000/24000 [==============================] - 8s 314us/step - loss: 0.3680 - acc: 0.8044 - val_loss: 0.7187 - val_acc: 0.7558\n"
+     ]
+    },
+    {
+     "data": {
+      "text/plain": [
+       "<keras.callbacks.History at 0x7f4de0cb9e10>"
+      ]
+     },
+     "execution_count": 11,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "fc_model.fit(train_data, train_labels,\n",
+    "            nb_epoch=50, batch_size=32,\n",
+    "            validation_data=(validation_data, validation_labels))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 12,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "fc_model.save_weights('/home/talyanskaya_marina/bottleneck_features_and_weights/fc_inception_iphone_250.hdf5') # сохраняем веса"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 13,
+   "metadata": {
+    "scrolled": true
+   },
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "24000/24000 [==============================] - 3s 128us/step\n"
+     ]
+    },
+    {
+     "data": {
+      "text/plain": [
+       "[0.71865298407276468, 0.75579166666666664]"
+      ]
+     },
+     "execution_count": 13,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "fc_model.evaluate(validation_data, validation_labels)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 59,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/jet/var/python/lib/python3.6/site-packages/ipykernel_launcher.py:9: UserWarning: Update your `Model` call to the Keras 2 API: `Model(inputs=Tensor(\"in..., outputs=Tensor(\"ou...)`\n",
+      "  if __name__ == '__main__':\n"
+     ]
+    }
+   ],
+   "source": [
+    "weights_filename='/home/talyanskaya_marina/bottleneck_features_and_weights/fc_inception_iphone_250.hdf5'\n",
+    "\n",
+    "x = Flatten()(inc_model.output)\n",
+    "x = Dense(64, activation='relu', name='dense_one')(x)\n",
+    "x = Dropout(0.5, name='dropout_one')(x)\n",
+    "x = Dense(64, activation='relu', name='dense_two')(x)\n",
+    "x = Dropout(0.5, name='dropout_two')(x)\n",
+    "top_model=Dense(1, activation='sigmoid', name='output')(x)\n",
+    "finalModel = Model(input=inc_model.input, output=top_model)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 60,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "weights_filename='/home/talyanskaya_marina/bottleneck_features_and_weights/fc_inception_iphone_250.hdf5'\n",
+    "finalModel.load_weights(weights_filename, by_name=True)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 61,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "for layer in inc_model.layers[:205]:\n",
+    "    layer.trainable = False"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 62,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import tensorflow as tf\n",
+    "from sklearn.metrics import average_precision_score\n",
+    "\n",
+    "def AP(y_true, y_pred):\n",
+    "    return tf.py_func(average_precision_score, (y_true, y_pred), tf.double)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 63,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "finalModel.compile(loss='binary_crossentropy',\n",
+    "              optimizer=SGD(lr=1e-4, momentum=0.9),\n",
+    "              metrics=['accuracy', AP])"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 64,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "try:\n",
+    "    os.mkdir('/home/talyanskaya_marina/new_model_weights/')\n",
+    "except OSError:\n",
+    "    shutil.rmtree('/home/talyanskaya_marina/new_model_weights/')\n",
+    "    os.mkdir('/home/talyanskaya_marina/new_model_weights/')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 65,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "filepath=\"/home/talyanskaya_marina/new_model_weights/weights-improvement-{epoch:02d}-{val_acc:.2f}.hdf5\"\n",
+    "checkpoint = ModelCheckpoint(filepath, monitor='val_acc', verbose=1, save_best_only=True, mode='max')\n",
+    "callbacks_list = [checkpoint]"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 66,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Found 24000 images belonging to 2 classes.\n",
+      "Found 24000 images belonging to 2 classes.\n",
+      "Found 24000 images belonging to 2 classes.\n"
+     ]
+    }
+   ],
+   "source": [
+    "train_datagen = ImageDataGenerator(\n",
+    "        rescale=1./255,\n",
+    "        rotation_range = 90,\n",
+    "        width_shift_range = 0.25,\n",
+    "        height_shift_range = 0.25,\n",
+    "        brightness_range = [0.3, 1.5],    \n",
+    "        shear_range = 0.4,\n",
+    "        zoom_range = 0.2,\n",
+    "        vertical_flip = True,\n",
+    "        horizontal_flip = True)\n",
+    "\n",
+    "test_datagen = ImageDataGenerator(rescale=1./255)\n",
+    "\n",
+    "\n",
+    "train_generator = train_datagen.flow_from_directory(\n",
+    "        '/home/talyanskaya_marina/Dataset_iphone_new/Dataset_iphone_new/train',#'/home/talyanskaya_marina/final_dataset/train',\n",
+    "        target_size=(150, 150),\n",
+    "        batch_size=32,\n",
+    "        class_mode='binary')\n",
+    "\n",
+    "validation_generator = test_datagen.flow_from_directory(\n",
+    "        '/home/talyanskaya_marina/Dataset_iphone_new/Dataset_iphone_new/validation',#'/home/talyanskaya_marina/final_dataset/validation/',\n",
+    "        target_size=(150, 150),\n",
+    "        batch_size=32,\n",
+    "        class_mode='binary')\n",
+    "\n",
+    "\n",
+    "pred_generator=test_datagen.flow_from_directory('/home/talyanskaya_marina/Dataset_iphone_new/Dataset_iphone_new/validation',#'/home/talyanskaya_marina/final_dataset/validation/',\n",
+    "                                                     target_size=(150,150),\n",
+    "                                                     batch_size=100,\n",
+    "                                                     class_mode='binary')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 67,
+   "metadata": {
+    "scrolled": true
+   },
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/jet/var/python/lib/python3.6/site-packages/ipykernel_launcher.py:7: UserWarning: The semantics of the Keras 2 argument `steps_per_epoch` is not the same as the Keras 1 argument `samples_per_epoch`. `steps_per_epoch` is the number of batches to draw from the generator at each epoch. Basically steps_per_epoch = samples_per_epoch/batch_size. Similarly `nb_val_samples`->`validation_steps` and `val_samples`->`steps` arguments have changed. Update your method calls accordingly.\n",
+      "  import sys\n",
+      "/jet/var/python/lib/python3.6/site-packages/ipykernel_launcher.py:7: UserWarning: Update your `fit_generator` call to the Keras 2 API: `fit_generator(<keras_pre..., validation_data=<keras_pre..., callbacks=[<keras.ca..., steps_per_epoch=750, epochs=100, validation_steps=24000)`\n",
+      "  import sys\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Epoch 1/100\n",
+      "750/750 [==============================] - 250s 333ms/step - loss: 0.5139 - acc: 0.7418 - AP: 0.8427 - val_loss: 0.4572 - val_acc: 0.7794 - val_AP: 0.8848\n",
+      "\n",
+      "Epoch 00001: val_acc improved from -inf to 0.77942, saving model to /home/talyanskaya_marina/new_model_weights/weights-improvement-01-0.78.hdf5\n",
+      "Epoch 2/100\n",
+      "750/750 [==============================] - 229s 306ms/step - loss: 0.5099 - acc: 0.7442 - AP: 0.8460 - val_loss: 0.4759 - val_acc: 0.7759 - val_AP: 0.8860\n",
+      "\n",
+      "Epoch 00002: val_acc did not improve from 0.77942\n",
+      "Epoch 3/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.5011 - acc: 0.7524 - AP: 0.8560 - val_loss: 0.4587 - val_acc: 0.7810 - val_AP: 0.8869\n",
+      "\n",
+      "Epoch 00003: val_acc improved from 0.77942 to 0.78104, saving model to /home/talyanskaya_marina/new_model_weights/weights-improvement-03-0.78.hdf5\n",
+      "Epoch 4/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4912 - acc: 0.7587 - AP: 0.8586 - val_loss: 0.4786 - val_acc: 0.7704 - val_AP: 0.8879\n",
+      "\n",
+      "Epoch 00004: val_acc did not improve from 0.78104\n",
+      "Epoch 5/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.4933 - acc: 0.7535 - AP: 0.8553 - val_loss: 0.4468 - val_acc: 0.7772 - val_AP: 0.8867\n",
+      "\n",
+      "Epoch 00005: val_acc did not improve from 0.78104\n",
+      "Epoch 6/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.4909 - acc: 0.7546 - AP: 0.8569 - val_loss: 0.4795 - val_acc: 0.7820 - val_AP: 0.8922\n",
+      "\n",
+      "Epoch 00006: val_acc improved from 0.78104 to 0.78204, saving model to /home/talyanskaya_marina/new_model_weights/weights-improvement-06-0.78.hdf5\n",
+      "Epoch 7/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.4934 - acc: 0.7576 - AP: 0.8612 - val_loss: 0.4664 - val_acc: 0.7762 - val_AP: 0.8884\n",
+      "\n",
+      "Epoch 00007: val_acc did not improve from 0.78204\n",
+      "Epoch 8/100\n",
+      "750/750 [==============================] - 228s 303ms/step - loss: 0.4846 - acc: 0.7630 - AP: 0.8618 - val_loss: 0.4656 - val_acc: 0.7890 - val_AP: 0.8909\n",
+      "\n",
+      "Epoch 00008: val_acc improved from 0.78204 to 0.78896, saving model to /home/talyanskaya_marina/new_model_weights/weights-improvement-08-0.79.hdf5\n",
+      "Epoch 9/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.4788 - acc: 0.7665 - AP: 0.8651 - val_loss: 0.4964 - val_acc: 0.7772 - val_AP: 0.8907\n",
+      "\n",
+      "Epoch 00009: val_acc did not improve from 0.78896\n",
+      "Epoch 10/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.4786 - acc: 0.7655 - AP: 0.8677 - val_loss: 0.4775 - val_acc: 0.7565 - val_AP: 0.8890\n",
+      "\n",
+      "Epoch 00010: val_acc did not improve from 0.78896\n",
+      "Epoch 11/100\n",
+      "750/750 [==============================] - 229s 305ms/step - loss: 0.4727 - acc: 0.7735 - AP: 0.8701 - val_loss: 0.4625 - val_acc: 0.7866 - val_AP: 0.8923\n",
+      "\n",
+      "Epoch 00011: val_acc did not improve from 0.78896\n",
+      "Epoch 12/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4673 - acc: 0.7740 - AP: 0.8728 - val_loss: 0.4715 - val_acc: 0.7842 - val_AP: 0.8926\n",
+      "\n",
+      "Epoch 00012: val_acc did not improve from 0.78896\n",
+      "Epoch 13/100\n",
+      "750/750 [==============================] - 227s 302ms/step - loss: 0.4718 - acc: 0.7736 - AP: 0.8718 - val_loss: 0.4394 - val_acc: 0.7954 - val_AP: 0.8937\n",
+      "\n",
+      "Epoch 00013: val_acc improved from 0.78896 to 0.79538, saving model to /home/talyanskaya_marina/new_model_weights/weights-improvement-13-0.80.hdf5\n",
+      "Epoch 14/100\n",
+      "750/750 [==============================] - 226s 302ms/step - loss: 0.4636 - acc: 0.7761 - AP: 0.8759 - val_loss: 0.4398 - val_acc: 0.7817 - val_AP: 0.8918\n",
+      "\n",
+      "Epoch 00014: val_acc did not improve from 0.79538\n",
+      "Epoch 15/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.4678 - acc: 0.7732 - AP: 0.8728 - val_loss: 0.4579 - val_acc: 0.7978 - val_AP: 0.8945\n",
+      "\n",
+      "Epoch 00015: val_acc improved from 0.79538 to 0.79783, saving model to /home/talyanskaya_marina/new_model_weights/weights-improvement-15-0.80.hdf5\n",
+      "Epoch 16/100\n",
+      "750/750 [==============================] - 229s 306ms/step - loss: 0.4626 - acc: 0.7793 - AP: 0.8765 - val_loss: 0.4520 - val_acc: 0.7980 - val_AP: 0.8949\n",
+      "\n",
+      "Epoch 00016: val_acc improved from 0.79783 to 0.79804, saving model to /home/talyanskaya_marina/new_model_weights/weights-improvement-16-0.80.hdf5\n",
+      "Epoch 17/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4643 - acc: 0.7807 - AP: 0.8780 - val_loss: 0.4435 - val_acc: 0.7944 - val_AP: 0.8961\n",
+      "\n",
+      "Epoch 00017: val_acc did not improve from 0.79804\n",
+      "Epoch 18/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.4608 - acc: 0.7838 - AP: 0.8813 - val_loss: 0.4275 - val_acc: 0.7910 - val_AP: 0.8973\n",
+      "\n",
+      "Epoch 00018: val_acc did not improve from 0.79804\n",
+      "Epoch 19/100\n",
+      "750/750 [==============================] - 228s 305ms/step - loss: 0.4534 - acc: 0.7808 - AP: 0.8799 - val_loss: 0.4763 - val_acc: 0.7955 - val_AP: 0.9001\n",
+      "\n",
+      "Epoch 00019: val_acc did not improve from 0.79804\n",
+      "Epoch 20/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4533 - acc: 0.7865 - AP: 0.8808 - val_loss: 0.4417 - val_acc: 0.7714 - val_AP: 0.8932\n",
+      "\n",
+      "Epoch 00020: val_acc did not improve from 0.79804\n",
+      "Epoch 21/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4483 - acc: 0.7863 - AP: 0.8839 - val_loss: 0.4275 - val_acc: 0.7970 - val_AP: 0.8970\n",
+      "\n",
+      "Epoch 00021: val_acc did not improve from 0.79804\n",
+      "Epoch 22/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.4516 - acc: 0.7873 - AP: 0.8828 - val_loss: 0.4364 - val_acc: 0.8035 - val_AP: 0.8959\n",
+      "\n",
+      "Epoch 00022: val_acc improved from 0.79804 to 0.80354, saving model to /home/talyanskaya_marina/new_model_weights/weights-improvement-22-0.80.hdf5\n",
+      "Epoch 23/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4517 - acc: 0.7837 - AP: 0.8844 - val_loss: 0.4464 - val_acc: 0.7952 - val_AP: 0.8948\n",
+      "\n",
+      "Epoch 00023: val_acc did not improve from 0.80354\n",
+      "Epoch 24/100\n",
+      "750/750 [==============================] - 228s 305ms/step - loss: 0.4503 - acc: 0.7891 - AP: 0.8865 - val_loss: 0.4428 - val_acc: 0.7893 - val_AP: 0.8968\n",
+      "\n",
+      "Epoch 00024: val_acc did not improve from 0.80354\n",
+      "Epoch 25/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4413 - acc: 0.7898 - AP: 0.8906 - val_loss: 0.4637 - val_acc: 0.7921 - val_AP: 0.8975\n",
+      "\n",
+      "Epoch 00025: val_acc did not improve from 0.80354\n",
+      "Epoch 26/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4412 - acc: 0.7898 - AP: 0.8908 - val_loss: 0.4530 - val_acc: 0.7969 - val_AP: 0.8967\n",
+      "\n",
+      "Epoch 00026: val_acc did not improve from 0.80354\n",
+      "Epoch 27/100\n",
+      "750/750 [==============================] - 228s 305ms/step - loss: 0.4374 - acc: 0.7943 - AP: 0.8888 - val_loss: 0.4636 - val_acc: 0.8064 - val_AP: 0.9009\n",
+      "\n",
+      "Epoch 00027: val_acc improved from 0.80354 to 0.80642, saving model to /home/talyanskaya_marina/new_model_weights/weights-improvement-27-0.81.hdf5\n",
+      "Epoch 28/100\n",
+      "750/750 [==============================] - 230s 307ms/step - loss: 0.4403 - acc: 0.7905 - AP: 0.8919 - val_loss: 0.4278 - val_acc: 0.7989 - val_AP: 0.9003\n",
+      "\n",
+      "Epoch 00028: val_acc did not improve from 0.80642\n",
+      "Epoch 29/100\n",
+      "750/750 [==============================] - 228s 305ms/step - loss: 0.4307 - acc: 0.7981 - AP: 0.8956 - val_loss: 0.4953 - val_acc: 0.8008 - val_AP: 0.8972\n",
+      "\n",
+      "Epoch 00029: val_acc did not improve from 0.80642\n",
+      "Epoch 30/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4278 - acc: 0.7982 - AP: 0.8980 - val_loss: 0.4825 - val_acc: 0.7950 - val_AP: 0.8969\n",
+      "\n",
+      "Epoch 00030: val_acc did not improve from 0.80642\n",
+      "Epoch 31/100\n",
+      "750/750 [==============================] - 228s 305ms/step - loss: 0.4274 - acc: 0.8006 - AP: 0.8965 - val_loss: 0.4449 - val_acc: 0.7927 - val_AP: 0.8959\n",
+      "\n",
+      "Epoch 00031: val_acc did not improve from 0.80642\n",
+      "Epoch 32/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4318 - acc: 0.8026 - AP: 0.8973 - val_loss: 0.4842 - val_acc: 0.8030 - val_AP: 0.8979\n",
+      "\n",
+      "Epoch 00032: val_acc did not improve from 0.80642\n",
+      "Epoch 33/100\n",
+      "750/750 [==============================] - 226s 301ms/step - loss: 0.4361 - acc: 0.7950 - AP: 0.8922 - val_loss: 0.4746 - val_acc: 0.7870 - val_AP: 0.8943\n",
+      "\n",
+      "Epoch 00033: val_acc did not improve from 0.80642\n",
+      "Epoch 34/100\n",
+      "750/750 [==============================] - 229s 305ms/step - loss: 0.4274 - acc: 0.8005 - AP: 0.8978 - val_loss: 0.4479 - val_acc: 0.8001 - val_AP: 0.8970\n",
+      "\n",
+      "Epoch 00034: val_acc did not improve from 0.80642\n",
+      "Epoch 35/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.4270 - acc: 0.7986 - AP: 0.8962 - val_loss: 0.4520 - val_acc: 0.7974 - val_AP: 0.8968\n",
+      "\n",
+      "Epoch 00035: val_acc did not improve from 0.80642\n",
+      "Epoch 36/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4263 - acc: 0.8017 - AP: 0.8973 - val_loss: 0.4809 - val_acc: 0.7960 - val_AP: 0.8984\n",
+      "\n",
+      "Epoch 00036: val_acc did not improve from 0.80642\n",
+      "Epoch 37/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.4237 - acc: 0.8022 - AP: 0.8982 - val_loss: 0.4509 - val_acc: 0.7725 - val_AP: 0.8929\n",
+      "\n",
+      "Epoch 00037: val_acc did not improve from 0.80642\n",
+      "Epoch 38/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4232 - acc: 0.8045 - AP: 0.9005 - val_loss: 0.4754 - val_acc: 0.8043 - val_AP: 0.9002\n",
+      "\n",
+      "Epoch 00038: val_acc did not improve from 0.80642\n",
+      "Epoch 39/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4227 - acc: 0.8059 - AP: 0.9006 - val_loss: 0.4247 - val_acc: 0.7912 - val_AP: 0.8995\n",
+      "\n",
+      "Epoch 00039: val_acc did not improve from 0.80642\n",
+      "Epoch 40/100\n",
+      "750/750 [==============================] - 229s 305ms/step - loss: 0.4201 - acc: 0.8098 - AP: 0.9018 - val_loss: 0.4620 - val_acc: 0.7816 - val_AP: 0.8972\n",
+      "\n",
+      "Epoch 00040: val_acc did not improve from 0.80642\n",
+      "Epoch 41/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4161 - acc: 0.8109 - AP: 0.9025 - val_loss: 0.4234 - val_acc: 0.7886 - val_AP: 0.8980\n",
+      "\n",
+      "Epoch 00041: val_acc did not improve from 0.80642\n",
+      "Epoch 42/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4124 - acc: 0.8065 - AP: 0.9059 - val_loss: 0.4238 - val_acc: 0.8027 - val_AP: 0.9019\n",
+      "\n",
+      "Epoch 00042: val_acc did not improve from 0.80642\n",
+      "Epoch 43/100\n",
+      "750/750 [==============================] - 226s 302ms/step - loss: 0.4178 - acc: 0.8094 - AP: 0.9014 - val_loss: 0.4229 - val_acc: 0.7955 - val_AP: 0.8981\n",
+      "\n",
+      "Epoch 00043: val_acc did not improve from 0.80642\n",
+      "Epoch 44/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.4145 - acc: 0.8071 - AP: 0.9032 - val_loss: 0.4363 - val_acc: 0.7953 - val_AP: 0.8985\n",
+      "\n",
+      "Epoch 00044: val_acc did not improve from 0.80642\n",
+      "Epoch 45/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4122 - acc: 0.8088 - AP: 0.9059 - val_loss: 0.4224 - val_acc: 0.8003 - val_AP: 0.9001\n",
+      "\n",
+      "Epoch 00045: val_acc did not improve from 0.80642\n",
+      "Epoch 46/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4132 - acc: 0.8069 - AP: 0.9046 - val_loss: 0.4412 - val_acc: 0.8000 - val_AP: 0.8990\n",
+      "\n",
+      "Epoch 00046: val_acc did not improve from 0.80642\n",
+      "Epoch 47/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4094 - acc: 0.8085 - AP: 0.9074 - val_loss: 0.4392 - val_acc: 0.8057 - val_AP: 0.8998\n",
+      "\n",
+      "Epoch 00047: val_acc did not improve from 0.80642\n",
+      "Epoch 48/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.4046 - acc: 0.8139 - AP: 0.9100 - val_loss: 0.4467 - val_acc: 0.8014 - val_AP: 0.9029\n",
+      "\n",
+      "Epoch 00048: val_acc did not improve from 0.80642\n",
+      "Epoch 49/100\n",
+      "750/750 [==============================] - 229s 305ms/step - loss: 0.4047 - acc: 0.8141 - AP: 0.9096 - val_loss: 0.4219 - val_acc: 0.8044 - val_AP: 0.9018\n",
+      "\n",
+      "Epoch 00049: val_acc did not improve from 0.80642\n",
+      "Epoch 50/100\n",
+      "750/750 [==============================] - 228s 303ms/step - loss: 0.4059 - acc: 0.8180 - AP: 0.9083 - val_loss: 0.4420 - val_acc: 0.8040 - val_AP: 0.9011\n",
+      "\n",
+      "Epoch 00050: val_acc did not improve from 0.80642\n",
+      "Epoch 51/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.4007 - acc: 0.8139 - AP: 0.9119 - val_loss: 0.4532 - val_acc: 0.7904 - val_AP: 0.8985\n",
+      "\n",
+      "Epoch 00051: val_acc did not improve from 0.80642\n",
+      "Epoch 52/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.4021 - acc: 0.8185 - AP: 0.9135 - val_loss: 0.4696 - val_acc: 0.8017 - val_AP: 0.8993\n",
+      "\n",
+      "Epoch 00052: val_acc did not improve from 0.80642\n",
+      "Epoch 53/100\n",
+      "750/750 [==============================] - 224s 299ms/step - loss: 0.3972 - acc: 0.8185 - AP: 0.9131 - val_loss: 0.4419 - val_acc: 0.7968 - val_AP: 0.9008\n",
+      "\n",
+      "Epoch 00053: val_acc did not improve from 0.80642\n",
+      "Epoch 54/100\n",
+      "750/750 [==============================] - 227s 302ms/step - loss: 0.4021 - acc: 0.8180 - AP: 0.9120 - val_loss: 0.4215 - val_acc: 0.8007 - val_AP: 0.9012\n",
+      "\n",
+      "Epoch 00054: val_acc did not improve from 0.80642\n",
+      "Epoch 55/100\n",
+      "750/750 [==============================] - 235s 314ms/step - loss: 0.3982 - acc: 0.8229 - AP: 0.9137 - val_loss: 0.4391 - val_acc: 0.8055 - val_AP: 0.8996\n",
+      "\n",
+      "Epoch 00055: val_acc did not improve from 0.80642\n",
+      "Epoch 56/100\n",
+      "750/750 [==============================] - 237s 316ms/step - loss: 0.4037 - acc: 0.8165 - AP: 0.9107 - val_loss: 0.4348 - val_acc: 0.7971 - val_AP: 0.8993\n",
+      "\n",
+      "Epoch 00056: val_acc did not improve from 0.80642\n",
+      "Epoch 57/100\n",
+      "750/750 [==============================] - 230s 307ms/step - loss: 0.3947 - acc: 0.8200 - AP: 0.9131 - val_loss: 0.4387 - val_acc: 0.8079 - val_AP: 0.9012\n",
+      "\n",
+      "Epoch 00057: val_acc improved from 0.80642 to 0.80792, saving model to /home/talyanskaya_marina/new_model_weights/weights-improvement-57-0.81.hdf5\n",
+      "Epoch 58/100\n",
+      "750/750 [==============================] - 229s 305ms/step - loss: 0.3989 - acc: 0.8192 - AP: 0.9120 - val_loss: 0.4329 - val_acc: 0.8007 - val_AP: 0.9038\n",
+      "\n",
+      "Epoch 00058: val_acc did not improve from 0.80792\n",
+      "Epoch 59/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.3963 - acc: 0.8220 - AP: 0.9131 - val_loss: 0.4789 - val_acc: 0.8097 - val_AP: 0.9026\n",
+      "\n",
+      "Epoch 00059: val_acc improved from 0.80792 to 0.80967, saving model to /home/talyanskaya_marina/new_model_weights/weights-improvement-59-0.81.hdf5\n",
+      "Epoch 60/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.3937 - acc: 0.8196 - AP: 0.9150 - val_loss: 0.4388 - val_acc: 0.8044 - val_AP: 0.9028\n",
+      "\n",
+      "Epoch 00060: val_acc did not improve from 0.80967\n",
+      "Epoch 61/100\n",
+      "750/750 [==============================] - 232s 309ms/step - loss: 0.3928 - acc: 0.8233 - AP: 0.9148 - val_loss: 0.4472 - val_acc: 0.8066 - val_AP: 0.9009\n",
+      "\n",
+      "Epoch 00061: val_acc did not improve from 0.80967\n",
+      "Epoch 62/100\n",
+      "750/750 [==============================] - 230s 306ms/step - loss: 0.3971 - acc: 0.8209 - AP: 0.9134 - val_loss: 0.4562 - val_acc: 0.8035 - val_AP: 0.9006\n",
+      "\n",
+      "Epoch 00062: val_acc did not improve from 0.80967\n",
+      "Epoch 63/100\n",
+      "750/750 [==============================] - 231s 308ms/step - loss: 0.3912 - acc: 0.8228 - AP: 0.9157 - val_loss: 0.4293 - val_acc: 0.8060 - val_AP: 0.9001\n",
+      "\n",
+      "Epoch 00063: val_acc did not improve from 0.80967\n",
+      "Epoch 64/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.3880 - acc: 0.8220 - AP: 0.9161 - val_loss: 0.4603 - val_acc: 0.8113 - val_AP: 0.9013\n",
+      "\n",
+      "Epoch 00064: val_acc improved from 0.80967 to 0.81125, saving model to /home/talyanskaya_marina/new_model_weights/weights-improvement-64-0.81.hdf5\n",
+      "Epoch 65/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.3873 - acc: 0.8223 - AP: 0.9167 - val_loss: 0.4635 - val_acc: 0.8043 - val_AP: 0.9012\n",
+      "\n",
+      "Epoch 00065: val_acc did not improve from 0.81125\n",
+      "Epoch 66/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.3874 - acc: 0.8280 - AP: 0.9184 - val_loss: 0.4335 - val_acc: 0.8006 - val_AP: 0.9029\n",
+      "\n",
+      "Epoch 00066: val_acc did not improve from 0.81125\n",
+      "Epoch 67/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.3872 - acc: 0.8260 - AP: 0.9180 - val_loss: 0.4298 - val_acc: 0.7970 - val_AP: 0.9005\n",
+      "\n",
+      "Epoch 00067: val_acc did not improve from 0.81125\n",
+      "Epoch 68/100\n",
+      "750/750 [==============================] - 226s 302ms/step - loss: 0.3862 - acc: 0.8266 - AP: 0.9192 - val_loss: 0.4452 - val_acc: 0.8109 - val_AP: 0.9054\n",
+      "\n",
+      "Epoch 00068: val_acc did not improve from 0.81125\n",
+      "Epoch 69/100\n",
+      "750/750 [==============================] - 227s 302ms/step - loss: 0.3841 - acc: 0.8291 - AP: 0.9195 - val_loss: 0.4446 - val_acc: 0.7977 - val_AP: 0.8996\n",
+      "\n",
+      "Epoch 00069: val_acc did not improve from 0.81125\n",
+      "Epoch 70/100\n"
+     ]
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.3856 - acc: 0.8282 - AP: 0.9192 - val_loss: 0.4786 - val_acc: 0.8095 - val_AP: 0.8999\n",
+      "\n",
+      "Epoch 00070: val_acc did not improve from 0.81125\n",
+      "Epoch 71/100\n",
+      "750/750 [==============================] - 225s 300ms/step - loss: 0.3864 - acc: 0.8276 - AP: 0.9190 - val_loss: 0.4501 - val_acc: 0.8095 - val_AP: 0.9046\n",
+      "\n",
+      "Epoch 00071: val_acc did not improve from 0.81125\n",
+      "Epoch 72/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.3810 - acc: 0.8245 - AP: 0.9205 - val_loss: 0.4365 - val_acc: 0.8054 - val_AP: 0.9019\n",
+      "\n",
+      "Epoch 00072: val_acc did not improve from 0.81125\n",
+      "Epoch 73/100\n",
+      "750/750 [==============================] - 228s 303ms/step - loss: 0.3811 - acc: 0.8251 - AP: 0.9186 - val_loss: 0.4793 - val_acc: 0.8050 - val_AP: 0.8994\n",
+      "\n",
+      "Epoch 00073: val_acc did not improve from 0.81125\n",
+      "Epoch 74/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.3750 - acc: 0.8315 - AP: 0.9226 - val_loss: 0.4352 - val_acc: 0.8058 - val_AP: 0.8994\n",
+      "\n",
+      "Epoch 00074: val_acc did not improve from 0.81125\n",
+      "Epoch 75/100\n",
+      "750/750 [==============================] - 229s 305ms/step - loss: 0.3757 - acc: 0.8317 - AP: 0.9219 - val_loss: 0.4895 - val_acc: 0.7990 - val_AP: 0.9010\n",
+      "\n",
+      "Epoch 00075: val_acc did not improve from 0.81125\n",
+      "Epoch 76/100\n",
+      "750/750 [==============================] - 235s 313ms/step - loss: 0.3792 - acc: 0.8289 - AP: 0.9206 - val_loss: 0.4686 - val_acc: 0.8027 - val_AP: 0.9004\n",
+      "\n",
+      "Epoch 00076: val_acc did not improve from 0.81125\n",
+      "Epoch 77/100\n",
+      "750/750 [==============================] - 233s 311ms/step - loss: 0.3725 - acc: 0.8354 - AP: 0.9258 - val_loss: 0.4697 - val_acc: 0.8102 - val_AP: 0.9000\n",
+      "\n",
+      "Epoch 00077: val_acc did not improve from 0.81125\n",
+      "Epoch 78/100\n",
+      "750/750 [==============================] - 233s 311ms/step - loss: 0.3749 - acc: 0.8331 - AP: 0.9231 - val_loss: 0.4576 - val_acc: 0.8127 - val_AP: 0.9021\n",
+      "\n",
+      "Epoch 00078: val_acc improved from 0.81125 to 0.81275, saving model to /home/talyanskaya_marina/new_model_weights/weights-improvement-78-0.81.hdf5\n",
+      "Epoch 79/100\n",
+      "750/750 [==============================] - 229s 305ms/step - loss: 0.3707 - acc: 0.8354 - AP: 0.9258 - val_loss: 0.4560 - val_acc: 0.8136 - val_AP: 0.9041\n",
+      "\n",
+      "Epoch 00079: val_acc improved from 0.81275 to 0.81363, saving model to /home/talyanskaya_marina/new_model_weights/weights-improvement-79-0.81.hdf5\n",
+      "Epoch 80/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.3767 - acc: 0.8310 - AP: 0.9218 - val_loss: 0.4510 - val_acc: 0.8017 - val_AP: 0.8998\n",
+      "\n",
+      "Epoch 00080: val_acc did not improve from 0.81363\n",
+      "Epoch 81/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.3725 - acc: 0.8326 - AP: 0.9249 - val_loss: 0.4572 - val_acc: 0.8076 - val_AP: 0.8998\n",
+      "\n",
+      "Epoch 00081: val_acc did not improve from 0.81363\n",
+      "Epoch 82/100\n",
+      "750/750 [==============================] - 231s 308ms/step - loss: 0.3743 - acc: 0.8295 - AP: 0.9218 - val_loss: 0.4514 - val_acc: 0.8115 - val_AP: 0.9043\n",
+      "\n",
+      "Epoch 00082: val_acc did not improve from 0.81363\n",
+      "Epoch 83/100\n",
+      "750/750 [==============================] - 232s 309ms/step - loss: 0.3642 - acc: 0.8360 - AP: 0.9262 - val_loss: 0.4420 - val_acc: 0.8072 - val_AP: 0.9027\n",
+      "\n",
+      "Epoch 00083: val_acc did not improve from 0.81363\n",
+      "Epoch 84/100\n",
+      "750/750 [==============================] - 230s 307ms/step - loss: 0.3701 - acc: 0.8372 - AP: 0.9256 - val_loss: 0.4545 - val_acc: 0.8102 - val_AP: 0.9003\n",
+      "\n",
+      "Epoch 00084: val_acc did not improve from 0.81363\n",
+      "Epoch 85/100\n",
+      "750/750 [==============================] - 230s 306ms/step - loss: 0.3703 - acc: 0.8342 - AP: 0.9252 - val_loss: 0.4585 - val_acc: 0.7995 - val_AP: 0.9008\n",
+      "\n",
+      "Epoch 00085: val_acc did not improve from 0.81363\n",
+      "Epoch 86/100\n",
+      "750/750 [==============================] - 227s 302ms/step - loss: 0.3710 - acc: 0.8337 - AP: 0.9249 - val_loss: 0.4479 - val_acc: 0.8047 - val_AP: 0.9024\n",
+      "\n",
+      "Epoch 00086: val_acc did not improve from 0.81363\n",
+      "Epoch 87/100\n",
+      "750/750 [==============================] - 229s 305ms/step - loss: 0.3647 - acc: 0.8370 - AP: 0.9266 - val_loss: 0.4416 - val_acc: 0.8076 - val_AP: 0.9039\n",
+      "\n",
+      "Epoch 00087: val_acc did not improve from 0.81363\n",
+      "Epoch 88/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.3680 - acc: 0.8345 - AP: 0.9257 - val_loss: 0.4616 - val_acc: 0.8116 - val_AP: 0.9029\n",
+      "\n",
+      "Epoch 00088: val_acc did not improve from 0.81363\n",
+      "Epoch 89/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.3653 - acc: 0.8404 - AP: 0.9273 - val_loss: 0.4458 - val_acc: 0.8122 - val_AP: 0.9052\n",
+      "\n",
+      "Epoch 00089: val_acc did not improve from 0.81363\n",
+      "Epoch 90/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.3550 - acc: 0.8415 - AP: 0.9331 - val_loss: 0.4564 - val_acc: 0.8140 - val_AP: 0.9041\n",
+      "\n",
+      "Epoch 00090: val_acc improved from 0.81363 to 0.81400, saving model to /home/talyanskaya_marina/new_model_weights/weights-improvement-90-0.81.hdf5\n",
+      "Epoch 91/100\n",
+      "750/750 [==============================] - 230s 306ms/step - loss: 0.3587 - acc: 0.8390 - AP: 0.9289 - val_loss: 0.4524 - val_acc: 0.8095 - val_AP: 0.9028\n",
+      "\n",
+      "Epoch 00091: val_acc did not improve from 0.81400\n",
+      "Epoch 92/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.3595 - acc: 0.8409 - AP: 0.9285 - val_loss: 0.4600 - val_acc: 0.8116 - val_AP: 0.9075\n",
+      "\n",
+      "Epoch 00092: val_acc did not improve from 0.81400\n",
+      "Epoch 93/100\n",
+      "750/750 [==============================] - 226s 302ms/step - loss: 0.3647 - acc: 0.8360 - AP: 0.9281 - val_loss: 0.4696 - val_acc: 0.8080 - val_AP: 0.9003\n",
+      "\n",
+      "Epoch 00093: val_acc did not improve from 0.81400\n",
+      "Epoch 94/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.3588 - acc: 0.8421 - AP: 0.9299 - val_loss: 0.4484 - val_acc: 0.8090 - val_AP: 0.9040\n",
+      "\n",
+      "Epoch 00094: val_acc did not improve from 0.81400\n",
+      "Epoch 95/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.3589 - acc: 0.8387 - AP: 0.9289 - val_loss: 0.4430 - val_acc: 0.8126 - val_AP: 0.9044\n",
+      "\n",
+      "Epoch 00095: val_acc did not improve from 0.81400\n",
+      "Epoch 96/100\n",
+      "750/750 [==============================] - 227s 303ms/step - loss: 0.3596 - acc: 0.8386 - AP: 0.9274 - val_loss: 0.4870 - val_acc: 0.8156 - val_AP: 0.9032\n",
+      "\n",
+      "Epoch 00096: val_acc improved from 0.81400 to 0.81563, saving model to /home/talyanskaya_marina/new_model_weights/weights-improvement-96-0.82.hdf5\n",
+      "Epoch 97/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.3621 - acc: 0.8404 - AP: 0.9290 - val_loss: 0.4139 - val_acc: 0.8125 - val_AP: 0.9056\n",
+      "\n",
+      "Epoch 00097: val_acc did not improve from 0.81563\n",
+      "Epoch 98/100\n",
+      "750/750 [==============================] - 229s 305ms/step - loss: 0.3600 - acc: 0.8397 - AP: 0.9294 - val_loss: 0.4225 - val_acc: 0.8122 - val_AP: 0.9050\n",
+      "\n",
+      "Epoch 00098: val_acc did not improve from 0.81563\n",
+      "Epoch 99/100\n",
+      "750/750 [==============================] - 227s 302ms/step - loss: 0.3568 - acc: 0.8392 - AP: 0.9296 - val_loss: 0.4410 - val_acc: 0.8113 - val_AP: 0.9072\n",
+      "\n",
+      "Epoch 00099: val_acc did not improve from 0.81563\n",
+      "Epoch 100/100\n",
+      "750/750 [==============================] - 228s 304ms/step - loss: 0.3546 - acc: 0.8436 - AP: 0.9318 - val_loss: 0.4489 - val_acc: 0.8107 - val_AP: 0.9059\n",
+      "\n",
+      "Epoch 00100: val_acc did not improve from 0.81563\n"
+     ]
+    }
+   ],
+   "source": [
+    "finalModel.fit_generator(\n",
+    "        train_generator,\n",
+    "        samples_per_epoch=24000,\n",
+    "        nb_epoch=100,\n",
+    "        validation_data=validation_generator,\n",
+    "        nb_val_samples=24000,\n",
+    "    callbacks=callbacks_list)\n",
+    "\n",
+    "try:\n",
+    "    os.mkdir('/home/talyanskaya_marina/lastModel')\n",
+    "except OSError:\n",
+    "    shutil.rmtree('/home/talyanskaya_marina/lastModel')\n",
+    "    os.mkdir('/home/talyanskaya_marina/lastModel')\n",
+    "finalModel.save('/home/talyanskaya_marina/lastModel/model.hdf5')"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Input - output funcion"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 57,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Found 750 images belonging to 1 classes.\n"
+     ]
+    }
+   ],
+   "source": [
+    "import argparse\n",
+    "import keras\n",
+    "import os\n",
+    "from keras.preprocessing.image import ImageDataGenerator\n",
+    "import pandas as pd\n",
+    "    \n",
+    "localPath = os.getenv(\"PWD\")\n",
+    "#os.chdir('/home/talyanskaya_marina/finalFolder')\n",
+    "#localPath = os.getcwd()\n",
+    "\n",
+    "def makeYourPredictions(inPath, modelPath, outPath):\n",
+    "\n",
+    "    model = keras.models.load_model(modelPath)\n",
+    "    \n",
+    "    finalTest_datagen = ImageDataGenerator(rescale=1./255)\n",
+    "\n",
+    "    finalTest_generator = finalTest_datagen.flow_from_directory(inPath,\n",
+    "                                            target_size=(150, 150),\n",
+    "                                            batch_size=32,\n",
+    "                                            class_mode='binary',\n",
+    "                                            shuffle=False)\n",
+    "    \n",
+    "    inFiles = finalTest_generator.filenames;\n",
+    "    \n",
+    "    finalPredictions = model.predict_generator(finalTest_generator, steps=len(inFiles)/32).flatten();\n",
+    "      \n",
+    "    out_df = pd.DataFrame()\n",
+    "    out_df['image_name'] = inFiles\n",
+    "    out_df['iphone_probability'] = finalPredictions\n",
+    "    out_df.to_csv(outPath + 'yourPredictions.csv', index=False)\n",
+    "\n",
+    "if __name__ == '__main__':\n",
+    "\n",
+    "    parser = argparse.ArgumentParser(description='Iphone detector')\n",
+    "    parser.add_argument('-i', type=str, default=localPath + '/in_folder')#)default='/home/talyanskaya_marina/data main/data/final test/test')\n",
+    "    parser.add_argument('-m', type=str, default=localPath + '/model.hdf5') #default='/home/talyanskaya_marina/savedModel-82/model-82.hdf5')\n",
+    "    parser.add_argument('-o', type=str, default=localPath + '/')#default='/home/talyanskaya_marina/YourPredictionsOutput')\n",
+    "    parser.add_argument('-f', type=str, default='nthg')\n",
+    "    args = parser.parse_args()\n",
+    "\n",
+    "    makeYourPredictions(args.i, args.m, args.o)\n",
+    "    \n",
+    "    \n",
+    "    #If we want to write down results in .txt file:\n",
+    "    #try:\n",
+    "    #    os.mkdir(outPath)\n",
+    "    #except OSError:\n",
+    "    #    shutil.rmtree(outPath)\n",
+    "    #    os.mkdir(outPath)\n",
+    "    #finalPath = outPath + '/yourPredictions.txt';\n",
+    "\n",
+    "    #f = open(finalPath,'w+');\n",
+    "    \n",
+    "    #for i in range(finalPredictions.shape[0]):\n",
+    "    #    f.write('Probability of ' + inFiles[i] + ' img is iphone is ' + str(round(finalPredictions[i], 2)) + '\\n');\n",
+    "        \n",
+    "    #close(f)\n",
+    "        "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 58,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "145.0"
+      ]
+     },
+     "execution_count": 58,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "df = pd.read_csv('/home/talyanskaya_marina/finalFolder/yourPredictions.csv')\n",
+    "df['decision']=round(df['iphone_probability'])\n",
+    "sum(df['decision'])"
+   ]
+  }
+ ],
+ "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": "3.6.3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}
diff --git a/Talyanskaya/predictImages.py b/Talyanskaya/predictImages.py
new file mode 100644
index 0000000..7df9a40
--- /dev/null
+++ b/Talyanskaya/predictImages.py
@@ -0,0 +1,67 @@
+#!/usr/bin/env 
+import argparse
+import keras
+import os
+from keras.preprocessing.image import ImageDataGenerator
+import pandas as pd
+
+from keras import backend as K
+K.set_image_dim_ordering('th')
+
+def AP(y_true, y_pred):
+    import tensorflow as tf
+    from sklearn.metrics import average_precision_score
+    return tf.py_func(average_precision_score, (y_true, y_pred), tf.double)
+    
+localPath = os.getenv("PWD")
+#os.chdir('/home/talyanskaya_marina/finalFolder')
+#localPath = os.getcwd()
+
+def makeYourPredictions(inPath, modelPath, outPath):
+
+    model = keras.models.load_model(modelPath, custom_objects={'AP': AP})
+    
+    finalTest_datagen = ImageDataGenerator(rescale=1./255)
+
+    finalTest_generator = finalTest_datagen.flow_from_directory(inPath,
+                                            target_size=(150, 150),
+                                            batch_size=32,
+                                            class_mode='binary',
+                                            shuffle=False)
+    
+    inFiles = finalTest_generator.filenames;
+    
+    finalPredictions = model.predict_generator(finalTest_generator, steps=len(inFiles)/32).flatten();
+      
+    out_df = pd.DataFrame()
+    out_df['image_name'] = inFiles
+    out_df['iphone_probability'] = 1 - finalPredictions
+    out_df.to_csv(outPath + 'yourPredictions.csv', index=False)
+
+if __name__ == '__main__':
+
+    parser = argparse.ArgumentParser(description='Iphone detector')
+    parser.add_argument('-i', type=str, default=localPath + '/in_folder')#)default='/home/talyanskaya_marina/data main/data/final test/test')
+    parser.add_argument('-m', type=str, default=localPath + '/model.hdf5') #default='/home/talyanskaya_marina/savedModel-82/model-82.hdf5')
+    parser.add_argument('-o', type=str, default=localPath + '/')#default='/home/talyanskaya_marina/YourPredictionsOutput')
+    parser.add_argument('-f', type=str, default='nthg')
+    args = parser.parse_args()
+
+    makeYourPredictions(args.i, args.m, args.o)
+    
+    
+    #If we want to write down results in .txt file:
+    #try:
+    #    os.mkdir(outPath)
+    #except OSError:
+    #    shutil.rmtree(outPath)
+    #    os.mkdir(outPath)
+    #finalPath = outPath + '/yourPredictions.txt';
+
+    #f = open(finalPath,'w+');
+    
+    #for i in range(finalPredictions.shape[0]):
+    #    f.write('Probability of ' + inFiles[i] + ' img is iphone is ' + str(round(finalPredictions[i], 2)) + '\n');
+        
+    #close(f)
+        
\ No newline at end of file
diff --git a/Talyanskaya/requirements.txt b/Talyanskaya/requirements.txt
new file mode 100644
index 0000000..2cc26b4
--- /dev/null
+++ b/Talyanskaya/requirements.txt
@@ -0,0 +1,78 @@
+astor==0.6.2
+backports.weakref==1.0rc1
+bleach==1.5.0
+certifi==2019.3.9
+chardet==3.0.4
+cycler==0.10.0
+Cython==0.25.2
+decorator==4.0.11
+docopt==0.6.2
+entrypoints==0.2.3
+enum34==1.1.6
+futures==3.1.1
+gast==0.2.0
+grpcio==1.10.0
+h5py==2.7.1
+html5lib==0.9999999
+idna==2.8
+ipykernel==4.6.1
+ipython==6.1.0
+ipython-genutils==0.2.0
+ipywidgets==6.0.0
+jedi==0.10.2
+Jinja2==2.9.6
+joblib==0.13.2
+jsonschema==2.6.0
+jupyter==1.0.0
+jupyter-client==5.1.0
+jupyter-console==5.1.0
+jupyter-core==4.3.0
+Keras==2.2.2
+Keras-Applications==1.0.4
+Keras-Preprocessing==1.0.2
+kiwisolver==1.1.0
+Markdown==2.6.8
+MarkupSafe==1.0
+matplotlib==3.1.0
+mistune==0.7.4
+nbconvert==5.2.1
+nbformat==4.3.0
+nose==1.3.7
+notebook==5.1.0
+numpy==1.13.3
+pandas==0.20.2
+pandocfilters==1.4.1
+pexpect==4.2.0
+pickleshare==0.7.4
+Pillow==6.0.0
+pipreqs==0.4.9
+prompt-toolkit==1.0.14
+protobuf==3.6.1
+ptyprocess==0.5.2
+pydot==1.2.3
+Pygments==2.2.0
+pyparsing==2.2.0
+python-dateutil==2.6.0
+pytz===2011k
+PyYAML==3.12
+pyzmq==16.0.2
+qtconsole==4.3.0
+requests==2.22.0
+scikit-learn==0.21.2
+scipy==1.0.0
+simplegeneric==0.8.1
+six==1.11.0
+sklearn==0.0
+tensorboard==1.10.0
+tensorflow==1.10.0
+termcolor==1.1.0
+terminado==0.6
+testpath==0.3.1
+tornado==4.5.1
+traitlets==4.3.2
+urllib3==1.25.3
+wcwidth==0.1.7
+webencodings==0.5.1
+Werkzeug==0.12.2
+widgetsnbextension==2.0.0
+yarg==0.1.9
\ No newline at end of file