【TF2】记录一次完整的深度学习代码

# 导入库
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers, datasets, optimizers, Sequential, metrics

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

# 定义batch大小
batch_size = 128

# 定义预处理函数
def preprocess(x, y):
    # 把x转换为float32，并把值压缩到[0, 1]
    x = tf.cast(x, dtype=tf.float32) / 255.
    # 把y转换为int32
    y = tf.cast(y, dtype=tf.int32)
    return x, y

# db数据集
# 将x,y转换成Dataset对象，才能进行随机打散，设置batch等操作
db = tf.data.Dataset.from_tensor_slices((x, y))
# 调用预处理函数，返回新的x，y的值。并随机打散，设置batch为128
db = db.map(preprocess).shuffle(10000).batch(batch_size)

# db_test测试数据集
# 将x_test,y_test转换成Dataset对象，才能进行设置batch等操作
db_test = tf.data.Dataset.from_tensor_slices((x_test, y_test))
# 调用预处理函数，返回新的x，y的值。并设置batch为128
db_test = db_test.map(preprocess).batch(batch_size)

# 查看样本
# db_iter = iter(db)
# spmple = next(db_iter)
# print(spmple[0].shape, spmple[1].shape)

# 定义网络模型
# 5个全连接层
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]
])

# 用于初始化层内的参数和变量（权重w和偏置值b），构建权值矩阵
# 通过build 函数完成内部张量（权值）的创建，其中None为任意设置的batch 数量，28*28为输入特征长度
model.build(input_shape=[None, 28*28])

# 查看网络结构
model.summary()

# 设置梯度优化器
# 对所有w、b进行更新操作，即w = w - lr * grad
optimizer = optimizers.Adam(lr=1e-3)

# 主函数
def main():
    # epoch = 30
    for epoch in range(30):
        for step, (x, y) in enumerate(db):
            # x: [b, 28, 28]
            # y: [b]
            x = tf.reshape(x, [-1, 784])    # [b, 28, 28] => [b, 784]

            with tf.GradientTape() as tape:
                # [b, 784] => [b, 10]
                # 完成前向传播
                logits = model(x)
                y_onehot = tf.one_hot(y, depth=10)
                # [b]
                # 均方差
                # loss_mse = tf.reduce_mean(tf.losses.MSE(y_onehot, logits))
                # 交叉熵
                loss_ce = tf.reduce_mean(tf.losses.categorical_crossentropy(y_onehot, logits, from_logits=True))

            # 梯度计算
            grads = tape.gradient(loss_ce, model.trainable_variables)
            # 梯度更新，对所有trainable_variables更新
            optimizer.apply_gradients(zip(grads, model.trainable_variables))

            # 每100个step打印一次信息
            if step % 100 == 0:
                print(epoch, step, 'loss:', float(loss_ce))

        # 测试数据集test
        total_correct, total_num = 0, 0
        for x,y in db_test:
            # x: [b, 28, 28]
            # y: [b]
            x = tf.reshape(x, [-1, 784])

            logits = model(x)

            # logits => prod    [b, 10]
            prod = tf.nn.softmax(logits, axis=1)

            # [b, 10] => [b]
            pred = tf.argmax(prod, axis=1)
            pred = tf.cast(pred, dtype=tf.int32)

            # pred: [b]
            # y: [b]
            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, 'test acc', acc)


if __name__ == '__main__':
    main()