TensorFlow2实战-系列教程15：Resnet实战3

🧡💛💚TensorFlow2实战-系列教程总目录

有任何问题欢迎在下面留言
本篇文章的代码运行界面均在Jupyter Notebook中进行
本篇文章配套的代码资源已经上传

Resnet实战1
Resnet实战2
Resnet实战3

7、训练脚本train.py解读------配置训练参数

# create modelmodel = get_model()# define loss and optimizerloss_object = tf.keras.losses.SparseCategoricalCrossentropy()optimizer = tf.keras.optimizers.Adam(lr=0.001)train_loss = tf.keras.metrics.Mean(name='train_loss')train_accuracy = tf.keras.metrics.SparseCategoricalAccuracy(name='train_accuracy')valid_loss = tf.keras.metrics.Mean(name='valid_loss')valid_accuracy = tf.keras.metrics.SparseCategoricalAccuracy(name='valid_accuracy')

前面已经解析了模型的构建
loss_object，多元交叉熵损失函数
optimizer ，Adam优化器，学习率为0.001
train_loss ，返回的是batch的平均损失
train_accuracy ，loss计算方法对应的准确率计算方法
valid_loss 和valid_accuracy 是验证集的平均损失和准确率计算方法

8、训练脚本train.py解读------模型训练

@tf.function
def train_step(images, labels):with tf.GradientTape() as tape:predictions = model(images, training=True)loss = loss_object(y_true=labels, y_pred=predictions)gradients = tape.gradient(loss, model.trainable_variables)optimizer.apply_gradients(grads_and_vars=zip(gradients, model.trainable_variables))train_loss(loss)train_accuracy(labels, predictions)@tf.function
def valid_step(images, labels):predictions = model(images, training=False)v_loss = loss_object(labels, predictions)valid_loss(v_loss)valid_accuracy(labels, predictions)

train_step(images, labels) 函数：

装饰器 @tf.function 将这个函数转换为 TensorFlow 图，这可以提高执行效率。
with tf.GradientTape() as tape：这是一个自动微分的上下文管理器，用于记录在其内部执行的所有操作，以便于后续计算梯度
predictions = model(images, training=True)：通过模型传递输入图像，得到预测结果。training=True 表示模型在训练模式下运行
loss = loss_object(y_true=labels, y_pred=predictions)：计算真实标签和预测标签之间的损失。
gradients = tape.gradient(loss, model.trainable_variables)：计算损失相对于模型可训练变量的梯度。
optimizer.apply_gradients(grads_and_vars=zip(gradients, model.trainable_variables))：应用梯度下降算法来更新模型的权重。
train_loss(loss) 和 train_accuracy(labels, predictions)：更新训练损失和准确率的指标。

valid_step(images, labels) 函数，不需要计算梯度，其他都一样

    for epoch in range(config.EPOCHS):train_loss.reset_states()train_accuracy.reset_states()valid_loss.reset_states()valid_accuracy.reset_states()step = 0for images, labels in train_dataset:step += 1train_step(images, labels)print("Epoch: {}/{}, step: {}/{}, loss: {:.5f}, accuracy: {:.5f}".format(epoch + 1, config.EPOCHS, step, math.ceil(train_count / config.BATCH_SIZE), train_loss.result(), train_accuracy.result()))for valid_images, valid_labels in valid_dataset:valid_step(valid_images, valid_labels)print("Epoch: {}/{}, train loss: {:.5f}, train accuracy: {:.5f}, ""valid loss: {:.5f}, valid accuracy: {:.5f}".format(epoch + 1, config.EPOCHS, train_loss.result(), train_accuracy.result(), valid_loss.result(), valid_accuracy.result()))model.save_weights(filepath=config.save_model_dir, save_format='tf')