本文是對(duì)官方文檔 的學(xué)習(xí)筆記疆液。

Layer 類(lèi)

在 Keras 中 layer 是一個(gè)核心組件吃谣， 它主要包含了連接權(quán)重以及算法或南。 如下是一個(gè)全連接層 Densely-connected.

class Linear(keras.layers.Layer):
    def __init__(self, units=32, input_dim=32):
        super(Linear, self).__init__()
        w_init = tf.random_normal_initializer()
        self.w = tf.Variable(
            initial_value=w_init(shape=(input_dim, units), dtype="float32"),
            trainable=True,
        )
        b_init = tf.zeros_initializer()
        self.b = tf.Variable(
            initial_value=b_init(shape=(units,), dtype="float32"), trainable=True
        )

    def call(self, inputs):
        return tf.matmul(inputs, self.w) + self.b

調(diào)用上面的 Layer

x = tf.ones((2, 2))
linear_layer = Linear(4, 2)
y = linear_layer(x)
print(y)

利用 add_weight() 初始化 weight

class Linear(keras.layers.Layer):
    def __init__(self, units=32, input_dim=32):
        super(Linear, self).__init__()
        self.w = self.add_weight(
            shape=(input_dim, units), initializer="random_normal", trainable=True
        )
        self.b = self.add_weight(shape=(units,), initializer="zeros", trainable=True)

    def call(self, inputs):
        return tf.matmul(inputs, self.w) + self.b


x = tf.ones((2, 2))
linear_layer = Linear(4, 2)
y = linear_layer(x)
print(y)

不可訓(xùn)練權(quán)重 （non-trainable weights）

開(kāi)發(fā)者可以在自己的 Layer 中加入一些不可變權(quán)重， 這些權(quán)重在反向傳播（backpropagation） 中不會(huì)被改變南缓。

class ComputeSum(keras.layers.Layer):
    def __init__(self, input_dim):
        super(ComputeSum, self).__init__()
        self.total = tf.Variable(initial_value=tf.zeros((input_dim,)), trainable=False)

    def call(self, inputs):
        self.total.assign_add(tf.reduce_sum(inputs, axis=0))
        return self.total


x = tf.ones((2, 2))
my_sum = ComputeSum(2)
y = my_sum(x)
print(y.numpy())
y = my_sum(x)
print(y.numpy())

print("weights:", len(my_sum.weights))
print("non-trainable weights:", len(my_sum.non_trainable_weights))

# It's not included in the trainable weights:
print("trainable_weights:", my_sum.trainable_weights)

最佳實(shí)踐（Best practice）：知道Input shape 之后再創(chuàng)建權(quán)重

如下呢岗，不要再 init 中創(chuàng)建weight限府， 而是將其放在 build 函數(shù)中本昏。 在模型第一次被使用時(shí)，call 會(huì)自動(dòng)調(diào)用 build 函數(shù)枪汪， 因而weight 就會(huì)被創(chuàng)建涌穆。

class Linear(keras.layers.Layer):
    def __init__(self, units=32):
        super(Linear, self).__init__()
        self.units = units

    def build(self, input_shape):
        self.w = self.add_weight(
            shape=(input_shape[-1], self.units),
            initializer="random_normal",
            trainable=True,
        )
        self.b = self.add_weight(
            shape=(self.units,), initializer="random_normal", trainable=True
        )

    def call(self, inputs):
        return tf.matmul(inputs, self.w) + self.b

# At instantiation, we don't know on what inputs this is going to get called
linear_layer = Linear(32)

# The layer's weights are created dynamically the first time the layer is called
y = linear_layer(x)

Layer 可以嵌套 Layer

外層Layer 會(huì)tracking 內(nèi)部layer 的權(quán)重怔昨， 創(chuàng)建子層的操作最好放在 init 函數(shù)中：

# Let's assume we are reusing the Linear class
# with a `build` method that we defined above.

class MLPBlock(keras.layers.Layer):
    def __init__(self):
        super(MLPBlock, self).__init__()
        self.linear_1 = Linear(32)
        self.linear_2 = Linear(32)
        self.linear_3 = Linear(1)

    def call(self, inputs):
        x = self.linear_1(inputs)
        x = tf.nn.relu(x)
        x = self.linear_2(x)
        x = tf.nn.relu(x)
        return self.linear_3(x)


mlp = MLPBlock()
y = mlp(tf.ones(shape=(3, 64)))  # The first call to the `mlp` will create the weights
print("weights:", len(mlp.weights))
print("trainable weights:", len(mlp.trainable_weights))

add_loss() 函數(shù)

可以利用 add_loss() 創(chuàng)建一個(gè) Regularization Layer ， 專(zhuān)門(mén)計(jì)算Regularization宿稀。

# A layer that creates an activity regularization loss
class ActivityRegularizationLayer(keras.layers.Layer):
    def __init__(self, rate=1e-2):
        super(ActivityRegularizationLayer, self).__init__()
        self.rate = rate

    def call(self, inputs):
        self.add_loss(self.rate * tf.reduce_sum(inputs))
        return inputs

class OuterLayer(keras.layers.Layer):
    def __init__(self):
        super(OuterLayer, self).__init__()
        self.activity_reg = ActivityRegularizationLayer(1e-2)

    def call(self, inputs):
        return self.activity_reg(inputs)


layer = OuterLayer()
assert len(layer.losses) == 0  # No losses yet since the layer has never been called

_ = layer(tf.zeros(1, 1))
assert len(layer.losses) == 1  # We created one loss value

# `layer.losses` gets reset at the start of each __call__
_ = layer(tf.zeros(1, 1))
assert len(layer.losses) == 1  # This is the loss created during the call above

所有子layer 的 Regularization 都會(huì)被最外層計(jì)算

class OuterLayerWithKernelRegularizer(keras.layers.Layer):
    def __init__(self):
        super(OuterLayerWithKernelRegularizer, self).__init__()
        self.dense = keras.layers.Dense(
            32, kernel_regularizer=tf.keras.regularizers.l2(1e-3)
        )

    def call(self, inputs):
        return self.dense(inputs)


layer = OuterLayerWithKernelRegularizer()
_ = layer(tf.zeros((1, 1)))

# This is `1e-3 * sum(layer.dense.kernel ** 2)`,
# created by the `kernel_regularizer` above.
print(layer.losses)

新加的Loss在訓(xùn)練的時(shí)候趁舀， 會(huì)被記入損失值

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

# Iterate over the batches of a dataset.
for x_batch_train, y_batch_train in train_dataset:
  with tf.GradientTape() as tape:
    logits = layer(x_batch_train)  # Logits for this minibatch
    # Loss value for this minibatch
    loss_value = loss_fn(y_batch_train, logits)
    # Add extra losses created during this forward pass:
    loss_value += sum(model.losses)

  grads = tape.gradient(loss_value, model.trainable_weights)
  optimizer.apply_gradients(zip(grads, model.trainable_weights))

這里涉及到自己實(shí)現(xiàn)訓(xùn)練，具體信息可以參考 guide to writing a training loop from scratch.

這些 Loss 函數(shù)也可以配合 Keras 內(nèi)建的 fit

import numpy as np

inputs = keras.Input(shape=(3,))
outputs = ActivityRegularizationLayer()(inputs)
model = keras.Model(inputs, outputs)

# If there is a loss passed in `compile`, thee regularization
# losses get added to it
model.compile(optimizer="adam", loss="mse")
model.fit(np.random.random((2, 3)), np.random.random((2, 3)))

# It's also possible not to pass any loss in `compile`,
# since the model already has a loss to minimize, via the `add_loss`
# call during the forward pass!
model.compile(optimizer="adam")
model.fit(np.random.random((2, 3)), np.random.random((2, 3)))

add_metric

類(lèi)似 loss 函數(shù)祝沸， metric 也可以自行定義矮烹。

class LogisticEndpoint(keras.layers.Layer):
    def __init__(self, name=None):
        super(LogisticEndpoint, self).__init__(name=name)
        self.loss_fn = keras.losses.BinaryCrossentropy(from_logits=True)
        self.accuracy_fn = keras.metrics.BinaryAccuracy()

    def call(self, targets, logits, sample_weights=None):
        # Compute the training-time loss value and add it
        # to the layer using `self.add_loss()`.
        loss = self.loss_fn(targets, logits, sample_weights)
        self.add_loss(loss)

        # Log accuracy as a metric and add it
        # to the layer using `self.add_metric()`.
        acc = self.accuracy_fn(targets, logits, sample_weights)
        self.add_metric(acc, name="accuracy")

        # Return the inference-time prediction tensor (for `.predict()`).
        return tf.nn.softmax(logits)

這樣使用 Metric 的好處是易于跟中其在訓(xùn)練中的變化

layer = LogisticEndpoint()

targets = tf.ones((2, 2))
logits = tf.ones((2, 2))
y = layer(targets, logits)

print("layer.metrics:", layer.metrics)
print("current accuracy value:", float(layer.metrics[0].result()))

在 fit 中也是同樣

inputs = keras.Input(shape=(3,), name="inputs")
targets = keras.Input(shape=(10,), name="targets")
logits = keras.layers.Dense(10)(inputs)
predictions = LogisticEndpoint(name="predictions")(logits, targets)

model = keras.Model(inputs=[inputs, targets], outputs=predictions)
model.compile(optimizer="adam")

data = {
    "inputs": np.random.random((3, 3)),
    "targets": np.random.random((3, 10)),
}
model.fit(data)

在 Layer 中開(kāi)啟序列化

如果在Layer 類(lèi)中實(shí)現(xiàn) get_config 那么就可以實(shí)現(xiàn)對(duì) layer 的序列化， 進(jìn)而單獨(dú) save/load 一層

class Linear(keras.layers.Layer):
    def __init__(self, units=32):
        super(Linear, self).__init__()
        self.units = units

    def build(self, input_shape):
        self.w = self.add_weight(
            shape=(input_shape[-1], self.units),
            initializer="random_normal",
            trainable=True,
        )
        self.b = self.add_weight(
            shape=(self.units,), initializer="random_normal", trainable=True
        )

    def call(self, inputs):
        return tf.matmul(inputs, self.w) + self.b

    def get_config(self):
        return {"units": self.units}


# Now you can recreate the layer from its config:
layer = Linear(64)
config = layer.get_config()
print(config)
new_layer = Linear.from_config(config)

如果打算序列化層罩锐， 那么在 __init__ 函數(shù)里奉狈，最好把輸入的重要參數(shù)都保存到 config 中，以便使用涩惑。

class Linear(keras.layers.Layer):
    def __init__(self, units=32, **kwargs):
        super(Linear, self).__init__(**kwargs)
        self.units = units

    def build(self, input_shape):
        self.w = self.add_weight(
            shape=(input_shape[-1], self.units),
            initializer="random_normal",
            trainable=True,
        )
        self.b = self.add_weight(
            shape=(self.units,), initializer="random_normal", trainable=True
        )

    def call(self, inputs):
        return tf.matmul(inputs, self.w) + self.b

    def get_config(self):
        config = super(Linear, self).get_config()
        config.update({"units": self.units})
        return config


layer = Linear(64)
config = layer.get_config()
print(config)
new_layer = Linear.from_config(config)

如果想定制反序列化仁期， 可以重載 from_config :

def from_config(cls, config):
  return cls(**config)

在 Call 函數(shù)中使用 Training 參數(shù)

有些 Layer 在訓(xùn)練和預(yù)測(cè)時(shí)表現(xiàn)不一樣， 所以就需要在Call 函數(shù)中得知當(dāng)前是在訓(xùn)練還是在預(yù)測(cè)竭恬。

class CustomDropout(keras.layers.Layer):
    def __init__(self, rate, **kwargs):
        super(CustomDropout, self).__init__(**kwargs)
        self.rate = rate

    def call(self, inputs, training=None):
        if training:
            return tf.nn.dropout(inputs, rate=self.rate)
        return inputs

在 Call 函數(shù)中使用 Mask 參數(shù)

簡(jiǎn)介沒(méi)看明白跛蛋， 詳情請(qǐng)移步 "understanding padding and masking".

Model 類(lèi)

對(duì)于定制化模型，一般來(lái)說(shuō)痊硕，用Layer 來(lái)實(shí)現(xiàn)內(nèi)部結(jié)果赊级， 然后用Model 類(lèi)來(lái)作為外層封裝。 Model類(lèi)和 Layer 類(lèi)的 API 格式相同岔绸， 但有如下區(qū)別：

• 會(huì)暴露內(nèi)建的訓(xùn)練函數(shù)理逊，比如：model.fit(), model.evaluate(), model.predict()
• 會(huì)通過(guò) model.layers 暴露內(nèi)部的層結(jié)構(gòu)
• 會(huì)暴露save 和序列化函數(shù) ：save(), save_weights() ...

什么時(shí)候用 Model 什么時(shí)候用 Layer ？
如果這個(gè)類(lèi)會(huì)用到 fit 亭螟， save 則需要用 Model 類(lèi)挡鞍， 否則用 fit。

class ResNet(tf.keras.Model):

    def __init__(self):
        super(ResNet, self).__init__()
        self.block_1 = ResNetBlock()
        self.block_2 = ResNetBlock()
        self.global_pool = layers.GlobalAveragePooling2D()
        self.classifier = Dense(num_classes)

    def call(self, inputs):
        x = self.block_1(inputs)
        x = self.block_2(x)
        x = self.global_pool(x)
        return self.classifier(x)


resnet = ResNet()
dataset = ...
resnet.fit(dataset, epochs=10)
resnet.save(filepath)

一個(gè)End-To-End 的例子

這里用一個(gè) AutoEncoder 來(lái)展示如何用 Layer 预烙，Model 來(lái)完成一個(gè) 端到端 (End-To-End ) 的模型：

from tensorflow.keras import layers


class Sampling(layers.Layer):
    """Uses (z_mean, z_log_var) to sample z, the vector encoding a digit."""

    def call(self, inputs):
        z_mean, z_log_var = inputs
        batch = tf.shape(z_mean)[0]
        dim = tf.shape(z_mean)[1]
        epsilon = tf.keras.backend.random_normal(shape=(batch, dim))
        return z_mean + tf.exp(0.5 * z_log_var) * epsilon


class Encoder(layers.Layer):
    """Maps MNIST digits to a triplet (z_mean, z_log_var, z)."""

    def __init__(self, latent_dim=32, intermediate_dim=64, name="encoder", **kwargs):
        super(Encoder, self).__init__(name=name, **kwargs)
        self.dense_proj = layers.Dense(intermediate_dim, activation="relu")
        self.dense_mean = layers.Dense(latent_dim)
        self.dense_log_var = layers.Dense(latent_dim)
        self.sampling = Sampling()

    def call(self, inputs):
        x = self.dense_proj(inputs)
        z_mean = self.dense_mean(x)
        z_log_var = self.dense_log_var(x)
        z = self.sampling((z_mean, z_log_var))
        return z_mean, z_log_var, z


class Decoder(layers.Layer):
    """Converts z, the encoded digit vector, back into a readable digit."""

    def __init__(self, original_dim, intermediate_dim=64, name="decoder", **kwargs):
        super(Decoder, self).__init__(name=name, **kwargs)
        self.dense_proj = layers.Dense(intermediate_dim, activation="relu")
        self.dense_output = layers.Dense(original_dim, activation="sigmoid")

    def call(self, inputs):
        x = self.dense_proj(inputs)
        return self.dense_output(x)


class VariationalAutoEncoder(keras.Model):
    """Combines the encoder and decoder into an end-to-end model for training."""

    def __init__(
        self,
        original_dim,
        intermediate_dim=64,
        latent_dim=32,
        name="autoencoder",
        **kwargs
    ):
        super(VariationalAutoEncoder, self).__init__(name=name, **kwargs)
        self.original_dim = original_dim
        self.encoder = Encoder(latent_dim=latent_dim, intermediate_dim=intermediate_dim)
        self.decoder = Decoder(original_dim, intermediate_dim=intermediate_dim)

    def call(self, inputs):
        z_mean, z_log_var, z = self.encoder(inputs)
        reconstructed = self.decoder(z)
        # Add KL divergence regularization loss.
        kl_loss = -0.5 * tf.reduce_mean(
            z_log_var - tf.square(z_mean) - tf.exp(z_log_var) + 1
        )
        self.add_loss(kl_loss)
        return reconstructed

利用這個(gè)模型來(lái)完成 MINST 訓(xùn)練

original_dim = 784
vae = VariationalAutoEncoder(original_dim, 64, 32)

optimizer = tf.keras.optimizers.Adam(learning_rate=1e-3)
mse_loss_fn = tf.keras.losses.MeanSquaredError()

loss_metric = tf.keras.metrics.Mean()

(x_train, _), _ = tf.keras.datasets.mnist.load_data()
x_train = x_train.reshape(60000, 784).astype("float32") / 255

train_dataset = tf.data.Dataset.from_tensor_slices(x_train)
train_dataset = train_dataset.shuffle(buffer_size=1024).batch(64)

epochs = 2

# Iterate over epochs.
for epoch in range(epochs):
    print("Start of epoch %d" % (epoch,))

    # Iterate over the batches of the dataset.
    for step, x_batch_train in enumerate(train_dataset):
        with tf.GradientTape() as tape:
            reconstructed = vae(x_batch_train)
            # Compute reconstruction loss
            loss = mse_loss_fn(x_batch_train, reconstructed)
            loss += sum(vae.losses)  # Add KLD regularization loss

        grads = tape.gradient(loss, vae.trainable_weights)
        optimizer.apply_gradients(zip(grads, vae.trainable_weights))

        loss_metric(loss)

        if step % 100 == 0:
            print("step %d: mean loss = %.4f" % (step, loss_metric.result()))

因?yàn)閂AE 類(lèi)是 Model 類(lèi)的子類(lèi)墨微， 它有內(nèi)建的訓(xùn)練功能， 所以也可以這樣寫(xiě)：

vae = VariationalAutoEncoder(784, 64, 32)

optimizer = tf.keras.optimizers.Adam(learning_rate=1e-3)

vae.compile(optimizer, loss=tf.keras.losses.MeanSquaredError())
vae.fit(x_train, x_train, epochs=2, batch_size=64)

用 Functional API 來(lái)實(shí)現(xiàn)

上面的例子使用了面向?qū)ο螅╫bject-oriented） 的方式來(lái)完成扁掸， 同樣的也可以用流程化的變成方式：

original_dim = 784
intermediate_dim = 64
latent_dim = 32

# Define encoder model.
original_inputs = tf.keras.Input(shape=(original_dim,), name="encoder_input")
x = layers.Dense(intermediate_dim, activation="relu")(original_inputs)
z_mean = layers.Dense(latent_dim, name="z_mean")(x)
z_log_var = layers.Dense(latent_dim, name="z_log_var")(x)
z = Sampling()((z_mean, z_log_var))
encoder = tf.keras.Model(inputs=original_inputs, outputs=z, name="encoder")

# Define decoder model.
latent_inputs = tf.keras.Input(shape=(latent_dim,), name="z_sampling")
x = layers.Dense(intermediate_dim, activation="relu")(latent_inputs)
outputs = layers.Dense(original_dim, activation="sigmoid")(x)
decoder = tf.keras.Model(inputs=latent_inputs, outputs=outputs, name="decoder")

# Define VAE model.
outputs = decoder(z)
vae = tf.keras.Model(inputs=original_inputs, outputs=outputs, name="vae")

# Add KL divergence regularization loss.
kl_loss = -0.5 * tf.reduce_mean(z_log_var - tf.square(z_mean) - tf.exp(z_log_var) + 1)
vae.add_loss(kl_loss)

# Train.
optimizer = tf.keras.optimizers.Adam(learning_rate=1e-3)
vae.compile(optimizer, loss=tf.keras.losses.MeanSquaredError())
vae.fit(x_train, x_train, epochs=3, batch_size=64)

這里也用了 Sampling 翘县， 形式不重要，實(shí)戰(zhàn)中可以使用混合形式谴分， 最高的目標(biāo)是達(dá)到你的目標(biāo)锈麸。