mnist_conv_using

# -*- coding: utf-8 -*-

import numpy
from keras.preprocessing import image
from keras.models import model_from_json
from matplotlib import pyplot as plt
from keras.backend import set_image_data_format

# Загрузка и компиляции модели
with open("mnist_conv.json", "r") as json_file:
    loaded_model_json = json_file.read()
loaded_model = model_from_json(loaded_model_json)
loaded_model.load_weights("mnist_conv.h5")
loaded_model.compile(loss="categorical_crossentropy", optimizer="adam",
                     metrics=["accuracy"])

# Какова размерность входных данных для загруженной нейронной сети?
first_layer_config = loaded_model.get_config()['layers'][0]['config']
batch_input_shape = first_layer_config['batch_input_shape']
data_format = first_layer_config['data_format']
img_shape = (batch_input_shape[2:] if
             data_format == 'channels_first' else
             batch_input_shape[1:-1])

# Загружаем свою картинку
img_path = '7py.png'
img = image.load_img(img_path, target_size=img_shape, color_mode='grayscale')
plt.imshow(img, cmap='gray')
plt.show()

# Преобразуем картинку в массив и нормализуем
set_image_data_format(data_format)
x = image.img_to_array(img)
x = 255 - x
x /= 255
x = numpy.expand_dims(x, axis=0)

prediction = loaded_model.predict(x)
print(prediction)
prediction = numpy.argmax(prediction, axis=1)
print(prediction)