吴裕雄--天生自然TensorFlow2教程：手写数字问题实战

import tensorflow as tf
from tensorflow import keras
from keras import Sequential,datasets, layers, optimizers, metrics

def preprocess(x, y):
    """数据处理函数"""
    x = tf.cast(x, dtype=tf.float32) / 255.
    y = tf.cast(y, dtype=tf.int32)
    return x, y

# 加载数据
(x, y), (x_test, y_test) = datasets.fashion_mnist.load_data()
print(x.shape, y.shape)

# 处理train数据
batch_size = 128
db = tf.data.Dataset.from_tensor_slices((x, y))
db = db.map(preprocess).shuffle(10000).batch(batch_size)

# 处理test数据
db_test = tf.data.Dataset.from_tensor_slices((x_test, y_test))
db_test = db_test.map(preprocess).batch(batch_size)

# # 生成train数据的迭代器
db_iter = iter(db)
sample = next(db_iter)
print(f'batch: {sample[0].shape,sample[1].shape}')

# 设计网络结构
model = Sequential([
    layers.Dense(256, activation=tf.nn.relu),  # [b,784] --> [b,256]
    layers.Dense(128, activation=tf.nn.relu),  # [b,256] --> [b,128]
    layers.Dense(64, activation=tf.nn.relu),  # [b,128] --> [b,64]
    layers.Dense(32, activation=tf.nn.relu),  # [b,64] --> [b,32]
    layers.Dense(10) # [b,32] --> [b,10], 330=32*10+10
])

model.build(input_shape=[None, 28 * 28])
model.summary()  # 调试
# w = w - lr*grad
optimizer = optimizers.Adam(lr=1e-3)  # 优化器，加快训练速度

def main():
    """主运行函数"""
    for epoch in range(10):
        for step, (x, y) in enumerate(db):
            # x:[b,28,28] --> [b,784]
            # y:[b]
            x = tf.reshape(x, [-1, 28 * 28])
            with tf.GradientTape() as tape:
                # [b,784] --> [b,10]
                logits = model(x)
                y_onehot = tf.one_hot(y, depth=10)
                # [b]
                loss_mse = tf.reduce_mean(tf.losses.MSE(y_onehot, logits))
                loss_ce = tf.reduce_mean(tf.losses.categorical_crossentropy(y_onehot,logits,from_logits=True))
            grads = tape.gradient(loss_ce, model.trainable_variables)
            optimizer.apply_gradients(zip(grads, model.trainable_variables))
            if step % 100 == 0:
                print(epoch, step, f'loss: {float(loss_ce),float(loss_mse)}')

        # test
        total_correct = 0
        total_num = 0
        for x, y in db_test:
            # x:[b,28,28] --> [b,784]
            # y:[b]
            x = tf.reshape(x, [-1, 28 * 28])
            # [b,10]
            logits = model(x)
            # logits --> prob [b,10]
            prob = tf.nn.softmax(logits, axis=1)
            # [b,10] --> [b], int32
            pred = tf.argmax(prob, axis=1)
            pred = tf.cast(pred, dtype=tf.int32)
            # pred:[b]
            # y:[b]
            # correct: [b], True: equal; False: not equal
            correct = tf.equal(pred, y)
            correct = tf.reduce_sum(tf.cast(correct, dtype=tf.int32))
            total_correct += int(correct)
            total_num += x.shape[0]
        acc = total_correct / total_num
        print(epoch, f'test acc: {acc}')

if __name__ == '__main__':
    main()