TF2 Keras (8) : 自己实现训练过程

本文是对官方文档 的学习笔记。

---

Keras 从三个层次上提供了训练，评估模型的接口

• 内建模型，这是最上层也是最简单，但是最缺乏定制能力的层 参考这里
• 使用内建fit ，但是希望自己实现部分定式化，需要自己重载 train_step方法 参考这里
• 自己从头（白纸）写起， 这是最low level 也是最灵活的一种方法。 就是本文要将的方法。

GradientTape ： 第一个端到端模型

在GradientTape 内部调用模型可以获得 loss value 的梯度， 优化器会利用这些梯度来更新参数。（利用 model.trainable_weights 获取可训练参数）

从一个最简单的 MINST 模型开始:

inputs = keras.Input(shape=(784,), name="digits")
x1 = layers.Dense(64, activation="relu")(inputs)
x2 = layers.Dense(64, activation="relu")(x1)
outputs = layers.Dense(10, name="predictions")(x2)
model = keras.Model(inputs=inputs, outputs=outputs)

首先，我们需要一个优化器，损失函数和一个数据集。

# Instantiate an optimizer.
optimizer = keras.optimizers.SGD(learning_rate=1e-3)
# Instantiate a loss function.
loss_fn = keras.losses.SparseCategoricalCrossentropy(from_logits=True)

# Prepare the training dataset.
batch_size = 64
(x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data()
x_train = np.reshape(x_train, (-1, 784))
x_test = np.reshape(x_test, (-1, 784))

# Reserve 10,000 samples for validation.
x_val = x_train[-10000:]
y_val = y_train[-10000:]
x_train = x_train[:-10000]
y_train = y_train[:-10000]

# Prepare the training dataset.
train_dataset = tf.data.Dataset.from_tensor_slices((x_train, y_train))
train_dataset = train_dataset.shuffle(buffer_size=1024).batch(batch_size)

# Prepare the validation dataset.
val_dataset = tf.data.Dataset.from_tensor_slices((x_val, y_val))
val_dataset = val_dataset.batch(batch_size)

训练循环 （training loop）

这里包括

• 一个 for 循环去遍历所有的epochs
• 在每个epoch 里， 用for训练遍历所有要训练的 batchs
• 在每个batch里，使用 GradientTape() Scope
• 在每个Scope 里， 利用 model (forward pass) 计算 损失值 (loss)
• 在Scope 外部， 或者各个权重针对上述 loss 的梯度 (gradients)
• 最后， 优化器(Optimizer) 根据梯度来更新权重。

epochs = 2
for epoch in range(epochs):
    print("\nStart of epoch %d" % (epoch,))

    # Iterate over the batches of the dataset.
    for step, (x_batch_train, y_batch_train) in enumerate(train_dataset):

        # Open a GradientTape to record the operations run
        # during the forward pass, which enables auto-differentiation.
        with tf.GradientTape() as tape:

            # Run the forward pass of the layer.
            # The operations that the layer applies
            # to its inputs are going to be recorded
            # on the GradientTape.
            logits = model(x_batch_train, training=True)  # Logits for this minibatch

            # Compute the loss value for this minibatch.
            loss_value = loss_fn(y_batch_train, logits)

        # Use the gradient tape to automatically retrieve
        # the gradients of the trainable variables with respect to the loss.
        grads = tape.gradient(loss_value, model.trainable_weights)

        # Run one step of gradient descent by updating
        # the value of the variables to minimize the loss.
        optimizer.apply_gradients(zip(grads, model.trainable_weights))

        # Log every 200 batches.
        if step % 200 == 0:
            print(
                "Training loss (for one batch) at step %d: %.4f"
                % (step, float(loss_value))
            )
            print("Seen so far: %s samples" % ((step + 1) * 64))

Low-level metrics

在上面的训练代码中， 可以使用内建的 Metric ， 流程是：

• 创建 Metric 实例
• 在每个 batch执行完毕后， 调用 metric.update_state()
• 当需要显示 metric 的时候， 调用 metric.result()
• 在需要重置 metric (每个 epoch 结束的时候)， 调用 metric.reset_states()

利用上诉方法， 在每个epoch 在验证集上计算 SparseCategoricalAccuracy

# Get model
inputs = keras.Input(shape=(784,), name="digits")
x = layers.Dense(64, activation="relu", name="dense_1")(inputs)
x = layers.Dense(64, activation="relu", name="dense_2")(x)
outputs = layers.Dense(10, name="predictions")(x)
model = keras.Model(inputs=inputs, outputs=outputs)

# Instantiate an optimizer to train the model.
optimizer = keras.optimizers.SGD(learning_rate=1e-3)
# Instantiate a loss function.
loss_fn = keras.losses.SparseCategoricalCrossentropy(from_logits=True)

# Prepare the metrics.
train_acc_metric = keras.metrics.SparseCategoricalAccuracy()
val_acc_metric = keras.metrics.SparseCategoricalAccuracy()

Training & evaluation loop:

import time

epochs = 2
for epoch in range(epochs):
    print("\nStart of epoch %d" % (epoch,))
    start_time = time.time()

    # Iterate over the batches of the dataset.
    for step, (x_batch_train, y_batch_train) in enumerate(train_dataset):
        with tf.GradientTape() as tape:
            logits = model(x_batch_train, training=True)
            loss_value = loss_fn(y_batch_train, logits)
        grads = tape.gradient(loss_value, model.trainable_weights)
        optimizer.apply_gradients(zip(grads, model.trainable_weights))

        # Update training metric.
        train_acc_metric.update_state(y_batch_train, logits)

        # Log every 200 batches.
        if step % 200 == 0:
            print(
                "Training loss (for one batch) at step %d: %.4f"
                % (step, float(loss_value))
            )
            print("Seen so far: %d samples" % ((step + 1) * 64))

    # Display metrics at the end of each epoch.
    train_acc = train_acc_metric.result()
    print("Training acc over epoch: %.4f" % (float(train_acc),))

    # Reset training metrics at the end of each epoch
    train_acc_metric.reset_states()

    # Run a validation loop at the end of each epoch.
    for x_batch_val, y_batch_val in val_dataset:
        val_logits = model(x_batch_val, training=False)
        # Update val metrics
        val_acc_metric.update_state(y_batch_val, val_logits)
    val_acc = val_acc_metric.result()
    val_acc_metric.reset_states()
    print("Validation acc: %.4f" % (float(val_acc),))
    print("Time taken: %.2fs" % (time.time() - start_time))

利用 tf.function加速训练

TensorFlow 2.0中的默认运行时是 eager execution。 这对于调试非常有用，但是 graph compilation 具有一定的性能优势。将计算描述为静态图可使框架应用全局性能优化。当框架被迫一个接一个地执行另一个操作，而又不知道接下来会发生什么时，优化不可能的。

可以将以Tensor为输入的任何函数编译为静态图。只需在其上添加一个@ tf.function装饰器，如下所示：

@tf.function
def train_step(x, y):
    with tf.GradientTape() as tape:
        logits = model(x, training=True)
        loss_value = loss_fn(y, logits)
    grads = tape.gradient(loss_value, model.trainable_weights)
    optimizer.apply_gradients(zip(grads, model.trainable_weights))
    train_acc_metric.update_state(y, logits)
    return loss_value

@tf.function
def test_step(x, y):
    val_logits = model(x, training=False)
    val_acc_metric.update_state(y, val_logits)

import time

epochs = 2
for epoch in range(epochs):
    print("\nStart of epoch %d" % (epoch,))
    start_time = time.time()

    # Iterate over the batches of the dataset.
    for step, (x_batch_train, y_batch_train) in enumerate(train_dataset):
        loss_value = train_step(x_batch_train, y_batch_train)

        # Log every 200 batches.
        if step % 200 == 0:
            print(
                "Training loss (for one batch) at step %d: %.4f"
                % (step, float(loss_value))
            )
            print("Seen so far: %d samples" % ((step + 1) * 64))

    # Display metrics at the end of each epoch.
    train_acc = train_acc_metric.result()
    print("Training acc over epoch: %.4f" % (float(train_acc),))

    # Reset training metrics at the end of each epoch
    train_acc_metric.reset_states()

    # Run a validation loop at the end of each epoch.
    for x_batch_val, y_batch_val in val_dataset:
        test_step(x_batch_val, y_batch_val)

    val_acc = val_acc_metric.result()
    val_acc_metric.reset_states()
    print("Validation acc: %.4f" % (float(val_acc),))
    print("Time taken: %.2fs" % (time.time() - start_time))

Low-level losses 追踪

Layer 和 Model 递归地跟踪调用self.add_loss（value）的 layer 在前向传递过程中删除时的任何 loss。loss 的结果列表可通过 model.losses 获得。如果要使用这些 loss 组件，则应将它们求和并将其添加到训练步骤的 loss 中。

一个 Regularization Layer

class ActivityRegularizationLayer(layers.Layer):
    def call(self, inputs):
        self.add_loss(1e-2 * tf.reduce_sum(inputs))
        return inputs

使用上面 Layer 的一个简单例子

inputs = keras.Input(shape=(784,), name="digits")
x = layers.Dense(64, activation="relu")(inputs)
# Insert activity regularization as a layer
x = ActivityRegularizationLayer()(x)
x = layers.Dense(64, activation="relu")(x)
outputs = layers.Dense(10, name="predictions")(x)

model = keras.Model(inputs=inputs, outputs=outputs)

训练函数

@tf.function
def train_step(x, y):
    with tf.GradientTape() as tape:
        logits = model(x, training=True)
        loss_value = loss_fn(y, logits)
        # Add any extra losses created during the forward pass.
        loss_value += sum(model.losses)
    grads = tape.gradient(loss_value, model.trainable_weights)
    optimizer.apply_gradients(zip(grads, model.trainable_weights))
    train_acc_metric.update_state(y, logits)
    return loss_value

总结

上面就是关于所有的细节部分。 下面的例子展示的是如何利用这些知识来创建一个面向 MINST 的 DCGAN

端到端的例子： 从头创建一个 GAN 训练循环

训练GAN 大致步骤：

• 训练鉴别器。 -在潜在空间中采样一批随机点。 -通过“生成器”模型将点转换为伪图像。 -获取一批真实图像，并将其与生成的图像合并。 -训练“判别器”模型对生成的图像与真实图像进行分类。
• 训练生成器。 -在潜在空间中采样随机点。 -通过“生成器”网络将点转换为伪图像。 -获取一批真实图像，并将其与生成的图像合并。 -训练“生成器”模型以“欺骗”鉴别器，并将伪造图像分类为真实图像。
• [关于 GAN 更多内容](Deep Learning with Python.
  )

创建鉴别器

discriminator = keras.Sequential(
    [
        keras.Input(shape=(28, 28, 1)),
        layers.Conv2D(64, (3, 3), strides=(2, 2), padding="same"),
        layers.LeakyReLU(alpha=0.2),
        layers.Conv2D(128, (3, 3), strides=(2, 2), padding="same"),
        layers.LeakyReLU(alpha=0.2),
        layers.GlobalMaxPooling2D(),
        layers.Dense(1),
    ],
    name="discriminator",
)
discriminator.summary()

创建一个生成器生成数据维度为 (28, 28, 1)

latent_dim = 128

generator = keras.Sequential(
    [
        keras.Input(shape=(latent_dim,)),
        # We want to generate 128 coefficients to reshape into a 7x7x128 map
        layers.Dense(7 * 7 * 128),
        layers.LeakyReLU(alpha=0.2),
        layers.Reshape((7, 7, 128)),
        layers.Conv2DTranspose(128, (4, 4), strides=(2, 2), padding="same"),
        layers.LeakyReLU(alpha=0.2),
        layers.Conv2DTranspose(128, (4, 4), strides=(2, 2), padding="same"),
        layers.LeakyReLU(alpha=0.2),
        layers.Conv2D(1, (7, 7), padding="same", activation="sigmoid"),
    ],
    name="generator",
)

重点， 训练循环

# Instantiate one optimizer for the discriminator and another for the generator.
d_optimizer = keras.optimizers.Adam(learning_rate=0.0003)
g_optimizer = keras.optimizers.Adam(learning_rate=0.0004)

# Instantiate a loss function.
loss_fn = keras.losses.BinaryCrossentropy(from_logits=True)


@tf.function
def train_step(real_images):
    # Sample random points in the latent space
    random_latent_vectors = tf.random.normal(shape=(batch_size, latent_dim))
    # Decode them to fake images
    generated_images = generator(random_latent_vectors)
    # Combine them with real images
    combined_images = tf.concat([generated_images, real_images], axis=0)

    # Assemble labels discriminating real from fake images
    labels = tf.concat(
        [tf.ones((batch_size, 1)), tf.zeros((real_images.shape[0], 1))], axis=0
    )
    # Add random noise to the labels - important trick!
    labels += 0.05 * tf.random.uniform(labels.shape)

    # Train the discriminator
    with tf.GradientTape() as tape:
        predictions = discriminator(combined_images)
        d_loss = loss_fn(labels, predictions)
    grads = tape.gradient(d_loss, discriminator.trainable_weights)
    d_optimizer.apply_gradients(zip(grads, discriminator.trainable_weights))

    # Sample random points in the latent space
    random_latent_vectors = tf.random.normal(shape=(batch_size, latent_dim))
    # Assemble labels that say "all real images"
    misleading_labels = tf.zeros((batch_size, 1))

    # Train the generator (note that we should *not* update the weights
    # of the discriminator)!
    with tf.GradientTape() as tape:
        predictions = discriminator(generator(random_latent_vectors))
        g_loss = loss_fn(misleading_labels, predictions)
    grads = tape.gradient(g_loss, generator.trainable_weights)
    g_optimizer.apply_gradients(zip(grads, generator.trainable_weights))
    return d_loss, g_loss, generated_images

训练GAN， 不挺的调用 train_step

import os

# Prepare the dataset. We use both the training & test MNIST digits.
batch_size = 64
(x_train, _), (x_test, _) = keras.datasets.mnist.load_data()
all_digits = np.concatenate([x_train, x_test])
all_digits = all_digits.astype("float32") / 255.0
all_digits = np.reshape(all_digits, (-1, 28, 28, 1))
dataset = tf.data.Dataset.from_tensor_slices(all_digits)
dataset = dataset.shuffle(buffer_size=1024).batch(batch_size)

epochs = 1  # In practice you need at least 20 epochs to generate nice digits.
save_dir = "./"

for epoch in range(epochs):
    print("\nStart epoch", epoch)

    for step, real_images in enumerate(dataset):
        # Train the discriminator & generator on one batch of real images.
        d_loss, g_loss, generated_images = train_step(real_images)

        # Logging.
        if step % 200 == 0:
            # Print metrics
            print("discriminator loss at step %d: %.2f" % (step, d_loss))
            print("adversarial loss at step %d: %.2f" % (step, g_loss))

            # Save one generated image
            img = tf.keras.preprocessing.image.array_to_img(
                generated_images[0] * 255.0, scale=False
            )
            img.save(os.path.join(save_dir, "generated_img" + str(step) + ".png"))

        # To limit execution time we stop after 10 steps.
        # Remove the lines below to actually train the model!
        if step > 10:
            break