Java Web学习平台   

import struct

import tensorflow as tf

import numpy as np

def get_img_datas(file_path):

    image_file = None

    with open(file_path, 'rb') as file:

        image_file = file.read()

    image_datas = []

    file_index = 0

    magic_num, data_num, rows, cols = struct.unpack_from('>IIII', image_file, file_index)

    file_index += struct.calcsize('>IIII')

    read_formate = ">{}B".format(rows * cols)

    for image_index in range(0, data_num):

        data = struct.unpack_from(read_formate, image_file, file_index)

        data = np.reshape(data, (28, 28))

        # print(data)

        image_datas.append(data)

        file_index += struct.calcsize(read_formate)

    print('read image datas {}/{}'.format(len(image_datas), data_num))

    return image_datas

def get_label_datas(file_path):

    with open(file_path, 'rb') as file:

        label_file = file.read()

    label_datas = []

    file_index = 0

    magic_num, data_num = struct.unpack_from('>II', label_file, file_index)

    file_index += struct.calcsize(">II")

    for label_index in range(0, data_num):

        data = struct.unpack_from(">B", label_file, file_index)

        label_datas.append(data)

        file_index += struct.calcsize(">B")

    print('read label datas {}/{}'.format(len(label_datas), data_num))

    return label_datas

def pre_process(image_data, label_data):

    image_data = np.array(image_data)

    image_data = image_data / 255.

    #image_data = np.ceil(image_data)

    # print(image_data[0])

    label_data = np.reshape(label_data, len(label_data))  # (60000,1)===>(60000,)

    # print(np.shape(image_data))

    # image_data = np.reshape(image_data, (6000, (28, 28)))

    # print(np.shape(image_data))

    return image_data, label_data

models = tf.keras.models

layers = tf.keras.layers

model = models.Sequential()

model.add(layers.Flatten(input_shape=(28, 28)))

model.add(layers.Dense(128, activation='relu'))

model.add(layers.Dense(128, activation='relu'))

model.add(layers.Dense(10, activation='softmax'))

model.compile(optimizer='adam',

              loss='sparse_categorical_crossentropy',

              metrics=['accuracy'])

model.summary()

train_image = get_img_datas("D:/images-idx3-ubyte/train-images-idx3-ubyte")

train_label = get_label_datas("D:/images-idx3-ubyte/train-labels-idx1-ubyte")

train_image, train_label = pre_process(train_image, train_label)

history = model.fit(x=train_image, y=train_label, epochs=15)

loss = history.history['loss']

acc = history.history['accuracy']

''' plt.figure(figsize=(15, 5), frameon=True) plt.subplot(1, 2, 1) # plt.plot(loss,label="Loss") plt.plot(acc, label='Training Accuracy') # plt.plot(val_acc, label='Validation Accuracy') plt.title('Training and Validation Accuracy') plt.legend() plt.grid() plt.subplot(1, 2, 2) plt.plot(loss, label='Training Loss') # plt.plot(val_loss, label='Validation Loss') plt.title('Training and Validation Loss') plt.legend() plt.show() '''

test_image = get_img_datas("D:/images-idx3-ubyte/t10k-images-idx3-ubyte")

test_label = get_label_datas("D:/images-idx3-ubyte/t10k-labels-idx1-ubyte")

test_image, test_label = pre_process(train_image, train_label)

print("===========================")

print(model.evaluate(test_image, test_label, verbose="2"))

print("===========================")