NN

# -*- coding: utf-8 -*-
"""
Created on Thu Feb  8 08:25:54 2018

@author: schiejak
"""


from __future__ import division, print_function, absolute_import 
import tflearn 
from tflearn.layers.core import input_data, dropout, fully_connected 
from tflearn.layers.conv import conv_2d, max_pool_2d
from tflearn.layers.estimator import regression
from tflearn.data_utils import image_preloader
import numpy as np
import tensorflow as tf



cesta_k_datum = ''

#Loading images

X, Y = image_preloader(cesta_k_datum, image_shape=(160, 160), mode='folder', 
                       files_extension='.png') 

X = np.reshape(X, (-1, 160, 160,1))

#Defining neural network
tf.reset_default_graph()


sit = input_data(shape=[None, 160, 160, 1])

sit = conv_2d(sit, 64, 3, activation='relu') 
sit = conv_2d(sit, 64, 3, activation='relu') 
sit = max_pool_2d(sit, 2, strides=2)

sit = conv_2d(sit, 128, 3, activation='relu') 
sit = conv_2d(sit, 128, 3, activation='relu') 
sit = max_pool_2d(sit, 2, strides=2) 

sit = conv_2d(sit, 256, 3, activation='relu') 
sit = conv_2d(sit, 256, 3, activation='relu') 
sit = max_pool_2d(sit, 2, strides=2) 

sit = conv_2d(sit, 512, 3, activation='relu') 
sit = conv_2d(sit, 512, 3, activation='relu') 
sit = max_pool_2d(sit, 2, strides=2) 

sit = conv_2d(sit, 512, 3, activation='relu') 
sit = conv_2d(sit, 512, 3, activation='relu') 
sit = max_pool_2d(sit, 2, strides=2) 

 
sit = fully_connected(sit, 1024, activation='relu') 
sit = dropout(sit, 0.8) 
sit = fully_connected(sit, 2, activation='softmax') 


sit = regression(sit, optimizer='adam', 
                     loss='categorical_crossentropy', 
                     learning_rate=0.0001) 

# Training 
model = tflearn.DNN(sit, checkpoint_path='model_proj_Y', 
                    max_checkpoints=1, tensorboard_verbose=0) 


model.fit(X, Y, n_epoch=30, validation_set = 0.1, shuffle=True, 
          show_metric=True, batch_size=8, snapshot_step=20, 
          snapshot_epoch=False, run_id='projekt_Y')