# from flask import Flask
# from keras.preprocessing.image import ImageDataGenerator
import numpy
import os
import scipy
from PIL import Image
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense
from keras.utils import np_utils
img_size = 66, 36

def rgb2gray(rgb):
    return numpy.dot(rgb[...,:3], [0.299, 0.587, 0.114])

def load_image( infilename ) :
    img = Image.open( infilename )
    img.load()
    data = numpy.asarray( img, dtype="int32" )
    return data

def loadDataFromFile(file_path):
    data = numpy.empty([0,66,36])
    img = mpimg.imread(file_path)
    gray_data = numpy.array([rgb2gray(img)])
    numpy.append(data, numpy.array(gray_data),axis = 0)
    data = numpy.append(data, numpy.array(gray_data),axis = 0)
    return data

def loadTrainData(dir_path):
    x_data = numpy.empty([0,66,36])
    y_data = numpy.array([])

    for folder in os.scandir(dir_path):
        if folder.is_dir():
            subdir_name = folder.name
            for img_file in os.scandir(folder.path):
                img = mpimg.imread(img_file.path)
                resized = scipy.misc.imresize(img,img_size)
                gray_data = numpy.array([rgb2gray(resized)])
                x_data = numpy.append(x_data, numpy.array(gray_data),axis = 0)
                y_data = numpy.append(y_data, numpy.array([int(subdir_name)]))

    return x_data, y_data

numpy.random.seed(42)

img_width, img_height = 36, 66
train_data_dir = 'C:\\Users\\Daniil\\Documents\\Visual Studio 2017\\Projects\\PythonApplication3\\data\\train'
test_data_dir = 'C:\\Users\\Daniil\\Documents\\Visual Studio 2017\\Projects\\PythonApplication3\\data\\test'
epochs = 40
batch_size = 16

(X_train, y_train) = loadTrainData(train_data_dir)
(X_test, y_test) = loadTrainData(test_data_dir)
X_train = X_train.reshape(X_train.shape[0], 2376)
X_test = X_test.reshape(X_test.shape[0], 2376)
X_train = X_train.astype('float32')
X_test = X_test.astype('float32')
X_train /= 255
X_test /= 255
y_test = np_utils.to_categorical(y_test, 10)
y_train = np_utils.to_categorical(y_train, 10)


model = Sequential()
# Добавляем уровни сети
# Входной слой: 800 нейронов, у каждого 784 входа, веса инициализируются случайно по нормальному распределению,
# функция активации ReLU
model.add(Dense(800, input_dim=2376, kernel_initializer="normal", activation="relu"))
#model.add(Dense(600, activation="relu", kernel_initializer="normal"))
# Выходной слой: 10 нейронов, веса инициализируются случайно по нормальному распределению, функция активации SoftMax
model.add(Dense(10, activation="softmax", kernel_initializer="normal"))

model.compile(loss="categorical_crossentropy", optimizer="SGD", metrics=["accuracy"])

model.fit(X_train, y_train, batch_size=batch_size, epochs=epochs, verbose=1)

scores = model.evaluate(X_test, y_test, verbose=1)

print("Точность работы на тестовых данных: %.2f%%" % (scores[1]*100))