Keras 简介 ¶
Keras 是建立和训练深度学习模型的高级 API。它被用于快速原型、高级研究和生产。Keras 具有三个主要优点:
- 用户友好
Keras API 简单、稳定、容易调试。 - 高度模块化
Keras API 可以像搭积木一样来构建深度学习系统。 - 易于扩展
可以很容易地实现研究过程中的各种新奇想法。
比如:创建新层、新的损失函数、提升state of art 模型的性能等。
1. 导入 tf.keras
¶
tf.keras
是 Keras API 在TensorFlow 里的实现。这是一个高级API,用于构建和训练模型,同时兼容 TensorFlow 的绝大部分功能,比如,eager execution, tf.data
模块,及Estimators。 tf.keras
使得 TensorFlow 更容易使用,同时牺牲灵活性和性能。
使用 tf.keras
,首先需要在您的代码开始时导入tf.keras
:
import tensorflow as tf
from tensorflow import keras
- 1
- 2
tf.keras
与 keras
绝对兼容,但请注意:
tf.keras
与keras
版本相同时,才绝对兼容。可以通过tf.keras.version
.来查看tf.keras
的版本。- 当保存一个模型的参数时,
tf.keras
默认保存成 checkpoint格式。可以通过设置save_format='h5'
来保存成 HDF5 格式。
2. 建立一个简单的模型 ¶
2.1 使用 Sequential API 构建模型 ¶
在 Keras 里,你用layers来搭建模型。一个模型(通常)是一个 layer 组成的图(Graph)。最常见的模型类型一般是由多个 layer 堆叠体:tf.keras.Sequential
模型
以构建一个简单的全连接网络(比如:多层感知器)为例:
model = keras.Sequential()
# Adds a densely-connected layer with 64 units to the model:
model.add(keras.layers.Dense(64, activation='relu'))
# Add another:
model.add(keras.layers.Dense(64, activation='relu'))
# Add a softmax layer with 10 output units:
model.add(keras.layers.Dense(10, activation='softmax'))
- 1
- 2
- 3
- 4
- 5
- 6
- 7
2.2 设置层的参数 ¶
在 tf.keras.layers
中有很多层,下面是一些通用的构造函数的参数:
activation
:设置层使用的激活函数。
指定方法:名称 或 可调用对象
默认为空。kernel_initializer
和bias_initializer
:设置层创建时,权重和偏差的初始化方法。
指定方法:名称 或 可调用对象
默认为"Glorot uniform"
initializer。kernel_regularizer
和bias_regularizer
:设置层的权重、偏差的正则化方法。比如:L1 或 L2 正则。
默认为空。
下面是一个实例,例子中对层的参数进行了指定:
# Create a sigmoid layer:
layers.Dense(64, activation='sigmoid')
# Or:
layers.Dense(64, activation=tf.sigmoid)
# A linear layer with L1 regularization of factor 0.01 applied to the kernel matrix:
layers.Dense(64, kernel_regularizer=keras.regularizers.l1(0.01))
# A linear layer with L2 regularization of factor 0.01 applied to the bias vector:
layers.Dense(64, bias_regularizer=keras.regularizers.l2(0.01))
# A linear layer with a kernel initialized to a random orthogonal matrix:
layers.Dense(64, kernel_initializer='orthogonal')
# A linear layer with a bias vector initialized to 2.0s:
layers.Dense(64, bias_initializer=keras.initializers.constant(2.0))
3. 训练和评估 ¶
3.1 配置训练过程 ¶
在模型构建完成后,通过调用 compile
方法来指定配置训练过程。
model.compile(optimizer=tf.train.AdamOptimizer(0.001),
loss='categorical_crossentropy',
metrics=['accuracy'])
- 1
- 2
- 3
tf.keras.Model.compile
有三个重要的参数:
optimizer
:训练过程使用的优化方法。此参数通过tf.train
模块的优化方法的实例来指定,比如:AdamOptimizer
,RMSPropOptimizer
,GradientDescentOptimizer
。loss
:训练过程中使用的损失函数(通过最小化损失函数来训练模型)。 常用的有:(mse
),categorical_crossentropy
和binary_crossentropy
。
指定方法:名称 或 可调用对象 from thetf.keras.losses
模块。metrics
:训练过程中,监测的指标(Used to monitor training)。
指定方法:名称 或 可调用对象 from thetf.keras.metrics
模块。
下面是配置模型训练过程的一个例子:
# Configure a model for mean-squared error regression.
model.compile(optimizer=tf.train.AdamOptimizer(0.01),
loss='mse', # mean squared error
metrics=['mae']) # mean absolute error
# Configure a model for categorical classification.
model.compile(optimizer=tf.train.RMSPropOptimizer(0.01),
loss=keras.losses.categorical_crossentropy,
metrics=[keras.metrics.categorical_accuracy])
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
3.2 输入 Numpy 数据 ¶
对于小的数据集,可以直接使用 NumPy 格式的数据进行训练、评估模型。模型使用 fit
方法直接使用 numpy 格式的数据训练模型:
import numpy as np
data = np.random.random((1000, 32))
labels = np.random.random((1000, 10))
model.fit(data, labels, epochs=10, batch_size=32)
- 1
- 2
- 3
- 4
- 5
- 6
tf.keras.Model.fit
有三个重要的参数:
epochs
: 训练多少个 epochs。batch_size
:当时用 NumPy 格式数据时,模型会将数据切片成许多的小 batches,然后迭代这些batches。
指定 batch size。
注意:如果数据不成正好分成如干个batch,最后一个 batch 可能会很小。validation_data
:当对模型进行原型研究时,您希望在某些验证数据上轻松地监视其性能。
通过这个参数(a tuple of inputs and labels)允许模型在每个 epoch 结束后,以推理模式在指定的数据集上计算并显示损失和评价指标
下面是一个使用 validatation_data
的例子:
import numpy as np
data = np.random.random((1000, 32))
labels = np.random.random((1000, 10))
val_data = np.random.random((100, 32))
val_labels = np.random.random((100, 10))
model.fit(data, labels, epochs=10, batch_size=32,
validation_data=(val_data, val_labels))
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
3.3 输入 tf.data.datasets
¶
使用 Datasets API 输入大型数据集 或 跨设备训练。 给 fit
方法传递一个 tf.data.Dataset
实例:
# Instantiates a toy dataset instance:
dataset = tf.data.Dataset.from_tensor_slices((data, labels))
dataset = dataset.batch(32)
dataset = dataset.repeat()
# Don't forget to specify `steps_per_epoch` when calling `fit` on a dataset.
model.fit(dataset, epochs=10, steps_per_epoch=30)
- 1
- 2
- 3
- 4
- 5
- 6
- 7
这里,fit
方法用 steps_per_epoch
参数来判断训练到第多少个 epoch 了。
Dataset API 也可以用于验证:
dataset = tf.data.Dataset.from_tensor_slices((data, labels))
dataset = dataset.batch(32).repeat()
val_dataset = tf.data.Dataset.from_tensor_slices((val_data, val_labels))
val_dataset = val_dataset.batch(32).repeat()
model.fit(dataset, epochs=10, steps_per_epoch=30,
validation_data=val_dataset,
validation_steps=3)
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
3.4 评估 和 预测 ¶
tf.keras.Model.evaluate
和 tf.keras.Model.predict
能够使用 NumPy 数据 和 tf.data.Dataset
数据。
以下面的方式进行评估:
model.evaluate(x, y, batch_size=32)
model.evaluate(dataset, steps=30)
- 1
- 2
- 3
以下面的方式进行预测:
model.predict(x, batch_size=32)
model.predict(dataset, steps=30)
- 1
- 2
- 3
4. 建立高级模型 ¶
4.1 使用 Function API 构建模型 ¶
tf.keras.Sequential
模型只适用于多层简单堆叠网络,不能表示复杂模型。使用 Keras functional API 可以构建有复杂拓扑结构的模型。比如:
- 多个输入的模型(Multi-input models)
- 多个输出的模型(Multi-output models)
- 有共享层的模型(Models with shared layers (the same layer called several times))
- 有 non-sequential 数据流的模型(Models with non-sequential data flows (e.g. residual connections))
函数式 API 特点:
- 层是可调用的,返回值是一个 tensor。
- 输入 tensors 和输出 tensors 被用来定义一个
tf.keras.Model
实例 - 函数式 API 构建的模型的训练同
Sequential
模型。
下面的代码使用函数式 API 构建了一个简单的全连接网络:
inputs = keras.Input(shape=(32,)) # Returns a placeholder tensor
# A layer instance is callable on a tensor, and returns a tensor.
x = keras.layers.Dense(64, activation='relu')(inputs)
x = keras.layers.Dense(64, activation='relu')(x)
predictions = keras.layers.Dense(10, activation='softmax')(x)
# Instantiate the model given inputs and outputs.
model = keras.Model(inputs=inputs, outputs=predictions)
# The compile step specifies the training configuration.
model.compile(optimizer=tf.train.RMSPropOptimizer(0.001),
loss='categorical_crossentropy',
metrics=['accuracy'])
# Trains for 5 epochs
model.fit(data, labels, batch_size=32, epochs=5)
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
- 16
- 17
4.2 编写 Model 的子类来构建模型(Model subclassing) ¶
通过编写 tf.keras.Model
的子类来构建一个自定义模型,并且定义你的模型的前向传播。在 __init__
方法里创建 layers,并且将 layers 设置为类的实例的属性。在 call
方法里定义前向传播过程。
当使用 eager execution 时,Model subclassing 方法特别有用(pytorch 里确实也是这么干的)。
提示: 根据工作的不同,请使用不同的 API。虽然 model subclassing 提供了灵活性,但也更复杂、更容易出错。如果可能,请尽量使用 function API。
下面是 model subclassing 例子:
class MyModel(keras.Model):
def __init__(self, num_classes=10):
super(MyModel, self).__init__(name='my_model')
self.num_classes = num_classes
# Define your layers here.
self.dense_1 = keras.layers.Dense(32, activation='relu')
self.dense_2 = keras.layers.Dense(num_classes, activation='sigmoid')
def call(self, inputs):
# Define your forward pass here,
# using layers you previously defined (in `__init__`).
x = self.dense_1(inputs)
return self.dense_2(x)
def compute_output_shape(self, input_shape):
# You need to override this function if you want to use the subclassed model
# as part of a functional-style model.
# Otherwise, this method is optional.
shape = tf.TensorShape(input_shape).as_list()
shape[-1] = self.num_classes
return tf.TensorShape(shape)
# Instantiates the subclassed model.
model = MyModel(num_classes=10)
# The compile step specifies the training configuration.
model.compile(optimizer=tf.train.RMSPropOptimizer(0.001),
loss='categorical_crossentropy',
metrics=['accuracy'])
# Trains for 5 epochs.
model.fit(data, labels, batch_size=32, epochs=5)
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
- 16
- 17
- 18
- 19
- 20
- 21
- 22
- 23
- 24
- 25
- 26
- 27
- 28
- 29
- 30
- 31
- 32
- 33
- 34
4.3 自定义 layers ¶
可以通过编写 tf.keras.layers.Layer
的子类来创建一个自定义 layer,该子类编写过程中需要编写下面的方法:
build
:创建层的参数。Add weights with theadd_weight
method.call
:定义前向传播过程。compute_output_shape
:指定怎么根据输入去计算 layer 的输出 shape。- 可选地,a layer can be serialized by implementing the
get_config
method and thefrom_config
class method.
这里有一个自定义 layer 的例子,该 layer 将输入和一个矩阵进行相乘:
class MyLayer(keras.layers.Layer):
def __init__(self, output_dim, **kwargs):
self.output_dim = output_dim
super(MyLayer, self).__init__(**kwargs)
def build(self, input_shape):
shape = tf.TensorShape((input_shape[1], self.output_dim))
# Create a trainable weight variable for this layer.
self.kernel = self.add_weight(name='kernel',
shape=shape,
initializer='uniform',
trainable=True)
# Be sure to call this at the end
super(MyLayer, self).build(input_shape)
def call(self, inputs):
return tf.matmul(inputs, self.kernel)
def compute_output_shape(self, input_shape):
shape = tf.TensorShape(input_shape).as_list()
shape[-1] = self.output_dim
return tf.TensorShape(shape)
def get_config(self):
base_config = super(MyLayer, self).get_config()
base_config['output_dim'] = self.output_dim
@classmethod
def from_config(cls, config):
return cls(**config)
# Create a model using the custom layer
model = keras.Sequential([MyLayer(10),
keras.layers.Activation('softmax')])
# The compile step specifies the training configuration
model.compile(optimizer=tf.train.RMSPropOptimizer(0.001),
loss='categorical_crossentropy',
metrics=['accuracy'])
# Trains for 5 epochs.
model.fit(data, targets, batch_size=32, epochs=5)
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
- 16
- 17
- 18
- 19
- 20
- 21
- 22
- 23
- 24
- 25
- 26
- 27
- 28
- 29
- 30
- 31
- 32
- 33
- 34
- 35
- 36
- 37
- 38
- 39
- 40
- 41
- 42
- 43
- 44
5. 回调(Callbacks) ¶
回调用来在训练过程中,自定义、扩展模型的行为(A callback is an object passed to a model to customize and extend its behavior during training)。你可以编写自定义 callback,也可以使用 tf.keras.callbacks
内置的 callback。
tf.keras.callbacks
内置的 callback 有:
tf.keras.callbacks.ModelCheckpoint
:定期保存checkpoints。tf.keras.callbacks.LearningRateScheduler
:动态改变学习速率。tf.keras.callbacks.EarlyStopping
:当验证集上的性能不再提高时,终止训练。tf.keras.callbacks.TensorBoard
:使用TensorBoard 监测模型的行为。
为了使用一个 tf.keras.callbacks.Callback
,将将它传递给模型的 fit
方法:
callbacks = [
# Interrupt training if `val_loss` stops improving for over 2 epochs
keras.callbacks.EarlyStopping(patience=2, monitor='val_loss'),
# Write TensorBoard logs to `./logs` directory
keras.callbacks.TensorBoard(log_dir='./logs')
]
model.fit(data, labels, batch_size=32, epochs=5, callbacks=callbacks,
validation_data=(val_data, val_targets))
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
6. 模型的保存和恢复 ¶
6.1 只保存参数(Weights only) ¶
使用 tf.keras.Model.save_weights
来保存和加载模型的 weights:
# Save weights to a TensorFlow Checkpoint file
model.save_weights('./my_model')
# Restore the model's state,
# this requires a model with the same architecture.
model.load_weights('my_model')
- 1
- 2
- 3
- 4
- 5
- 6
默认情况下,这会以 TensorFlow checkpoint 格式保存模型的 weights。weights 也可以保存为 HDF5 格式(Keras 默认的保存格式):
# Save weights to a HDF5 file
model.save_weights('my_model.h5', save_format='h5')
# Restore the model's state
model.load_weights('my_model.h5')
- 1
- 2
- 3
- 4
- 5
6.2 只保存模型(Configuration only) ¶
一个模型的 configuration 可以被保存,序列化过程中不包含任何 weights。保存的 configuration 可以用来重新创建、初始化出相同的模型,即使没有模型原始的定义代码。Keras 支持 JSON,YAML 序列化格式:
# Serialize a model to JSON format
json_string = model.to_json()
# Recreate the model (freshly initialized)
fresh_model = keras.models.from_json(json_string)
# Serializes a model to YAML format
yaml_string = model.to_yaml()
# Recreate the model
fresh_model = keras.models.from_yaml(yaml_string)
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
注意:Subclassed models 是不可序列化的,因为它们的结构是在 Python 代码内的 call 方法里定义的。
6.3 整个模型(Entire model) ¶
整个模型可以被保存成一个文件(同时包含weights、configuration,甚至optimizer’s configuration)。这允许你去 checkpoint 一个模型、从 checkpoint 中模型所处状态恢复训练,而不需要原始的代码:
# Create a trivial model
model = keras.Sequential([
keras.layers.Dense(10, activation='softmax', input_shape=(32,)),
keras.layers.Dense(10, activation='softmax')
])
model.compile(optimizer='rmsprop',
loss='categorical_crossentropy',
metrics=['accuracy'])
model.fit(data, targets, batch_size=32, epochs=5)
# Save entire model to a HDF5 file
model.save('my_model.h5')
# Recreate the exact same model, including weights and optimizer.
model = keras.models.load_model('my_model.h5')
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
- 9
- 10
- 11
- 12
- 13
- 14
- 15
- 16
7. Eager execution ¶
Eager execution 是一个即时执行的编程环境(图的定义和执行是同步的,可以动态的更改图)。Keras 不支持动态图,但 tf.keras
支持,并且在程序检查和调试过程中非常有用。
tf.keras
里所有的模型构建 API 兼容 eager execution。并且,在编写 model subclassing 和 custom layers 时使用 eager execution,好处多多。
请看 eager execution guide 里的例子:使用 Keras models with custom training loops and tf.GradientTape
。
8. 分布式 ¶
8.1 Estimators ¶
Estimators API 被用来在分布时环境训练模型。`Estimator` API 旨在大型数据集的分布式训练,该 API 能够导出工业生产可用的模型。 is used for training models
一个 tf.keras.Model
可以用 tf.estimator
API 来训练(通过 tf.keras.estimator.model_to_estimator
将模型转为一个 tf.estimator.Estimator
对象)。详情见 Creating Estimators from Keras models。
model = keras.Sequential([layers.Dense(10,activation='softmax'),
layers.Dense(10,activation='softmax')])
model.compile(optimizer=tf.train.RMSPropOptimizer(0.001),
loss='categorical_crossentropy',
metrics=['accuracy'])
estimator = keras.estimator.model_to_estimator(model)
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
提示: 可以通过开启 eager execution 来调试 Estimator input functions、检查数据。
8.2 多 GPU ¶
tf.keras
模型可以通过 tf.contrib.distribute.DistributionStrategy
在多个 GPU 上运行。这个 API 在几乎不需要更改代码的情况下,实现在多个 GPU 上的分布式训练。
当前,tf.contrib.distribute.MirroredStrategy
是唯一支持的分布式策略。MirroredStrategy
对图进行复制,以同步的方式训练,并且梯度最后聚集在一个机器上。 为了使用 DistributionStrategy
with Keras,首先用 tf.keras.estimator.model_to_estimator
将 tf.keras.Model
转化为一个 tf.estimator.Estimator
,然后训练转化来的estimator。
下面的例子在一个机器的多个 GPU 上实现了 `tf.keras.Model` 的训练
首先,定义一个简单的模型:
model = keras.Sequential()
model.add(keras.layers.Dense(16, activation='relu', input_shape=(10,)))
model.add(keras.layers.Dense(1, activation='sigmoid'))
optimizer = tf.train.GradientDescentOptimizer(0.2)
model.compile(loss='binary_crossentropy', optimizer=optimizer)
model.summary()
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
定义输入 pipeline。`input_fn` 返回一个 `tf.data.Dataset` 对象,该对象用来将数据传给多个设备,每一个设备处理一个 batch 的一个 slice。
def input_fn():
x = np.random.random((1024, 10))
y = np.random.randint(2, size=(1024, 1))
x = tf.cast(x, tf.float32)
dataset = tf.data.Dataset.from_tensor_slices((x, y))
dataset = dataset.repeat(10)
dataset = dataset.batch(32)
return dataset
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
接下来,创建一个 tf.estimator.RunConfig
并设置 train_distribute
参数为 tf.contrib.distribute.MirroredStrategy
实例。当创建 MirroredStrategy
时,你可以指定一个设备列表 或 通过 num_gpus
参数设置 GPU 的数量。默认使用所有的 GPU。
strategy = tf.contrib.distribute.MirroredStrategy()
config = tf.estimator.RunConfig(train_distribute=strategy)
- 1
- 2
将 Keras model 转为一个 tf.estimator.Estimator
实例。
keras_estimator = keras.estimator.model_to_estimator(
keras_model=model,
config=config,
model_dir='/tmp/model_dir')
- 1
- 2
- 3
- 4
最后,训练这个 Estimator
实例 :
keras_estimator.train(input_fn=input_fn, steps=10)
- 1
Keras 简介 ¶
Keras 是建立和训练深度学习模型的高级 API。它被用于快速原型、高级研究和生产。Keras 具有三个主要优点:
- 用户友好
Keras API 简单、稳定、容易调试。 - 高度模块化
Keras API 可以像搭积木一样来构建深度学习系统。 - 易于扩展
可以很容易地实现研究过程中的各种新奇想法。
比如:创建新层、新的损失函数、提升state of art 模型的性能等。
1. 导入 tf.keras
¶
tf.keras
是 Keras API 在TensorFlow 里的实现。这是一个高级API,用于构建和训练模型,同时兼容 TensorFlow 的绝大部分功能,比如,eager execution, tf.data
模块,及Estimators。 tf.keras
使得 TensorFlow 更容易使用,同时牺牲灵活性和性能。
使用 tf.keras
,首先需要在您的代码开始时导入tf.keras
:
import tensorflow as tf
from tensorflow import keras
- 1
- 2
tf.keras
与 keras
绝对兼容,但请注意:
tf.keras
与keras
版本相同时,才绝对兼容。可以通过tf.keras.version
.来查看tf.keras
的版本。- 当保存一个模型的参数时,
tf.keras
默认保存成 checkpoint格式。可以通过设置save_format='h5'
来保存成 HDF5 格式。
2. 建立一个简单的模型 ¶
2.1 使用 Sequential API 构建模型 ¶
在 Keras 里,你用layers来搭建模型。一个模型(通常)是一个 layer 组成的图(Graph)。最常见的模型类型一般是由多个 layer 堆叠体:tf.keras.Sequential
模型
以构建一个简单的全连接网络(比如:多层感知器)为例:
model = keras.Sequential()
# Adds a densely-connected layer with 64 units to the model:
model.add(keras.layers.Dense(64, activation='relu'))
# Add another:
model.add(keras.layers.Dense(64, activation='relu'))
# Add a softmax layer with 10 output units:
model.add(keras.layers.Dense(10, activation='softmax'))
- 1
- 2
- 3
- 4
- 5
- 6
- 7
2.2 设置层的参数 ¶
在 tf.keras.layers
中有很多层,下面是一些通用的构造函数的参数:
activation
:设置层使用的激活函数。
指定方法:名称 或 可调用对象
默认为空。kernel_initializer
和bias_initializer
:设置层创建时,权重和偏差的初始化方法。
指定方法:名称 或 可调用对象
默认为"Glorot uniform"
initializer。kernel_regularizer
和bias_regularizer
:设置层的权重、偏差的正则化方法。比如:L1 或 L2 正则。
默认为空。
下面是一个实例,例子中对层的参数进行了指定: