tf.layers

tf.layers

定义在tensorflow/python/layers/layers.py, 为我们提供了一些高层次的构建神经网络的接口。

函数

提供的主要API包含以下几个方面：

• 卷积
• pooling
• 全连接
• 批归一化BN
• dropout

卷积

tf.layers.conv1d

conv1d(
inputs,
filters,
kernel_size,
strides=1,
padding=’valid’,
data_format=’channels_last’,
dilation_rate=1,
activation=None,
use_bias=True,
kernel_initializer=None,
bias_initializer=tf.zeros_initializer(),
kernel_regularizer=None,
bias_regularizer=None,
activity_regularizer=None,
trainable=True,
name=None,
reuse=None
)

tf.layers.conv2d

conv2d(
inputs,
filters,
kernel_size,
strides=(1, 1),
padding=’valid’,
data_format=’channels_last’,
dilation_rate=(1, 1),
activation=None,
use_bias=True,
kernel_initializer=None,
bias_initializer=tf.zeros_initializer(),
kernel_regularizer=None,
bias_regularizer=None,
activity_regularizer=None,
trainable=True,
name=None,
reuse=None
)

tf.layers.conv3d

conv3d(
inputs,
filters,
kernel_size,
strides=(1, 1, 1),
padding=’valid’,
data_format=’channels_last’,
dilation_rate=(1, 1, 1),
activation=None,
use_bias=True,
kernel_initializer=None,
bias_initializer=tf.zeros_initializer(),
kernel_regularizer=None,
bias_regularizer=None,
activity_regularizer=None,
trainable=True,
name=None,
reuse=None
)

以上3个函数重要的设置函数主要是：

• inputs 为你输入的tensor
• filters: 整数，指定了你输出的channel数目
• kernel_size: 卷积核大小，一个整数/3个整数的列表/元组， 指定[depth, height, width]
• strides, [depth, height, width]上移动的步长
• padding：same 输入输出的[height, width] 一致， valid 为[height - kernel_size +1, width-kernel_size +1]
• data_format: 默认是channels_last ，即(batch, depth, height, width, channels)，也可以是channels_first，即(batch, channels, depth, height, width)

pool

主要包含average_pooling 和 max_pooling两种方式, 处理的维度有1，2，3维。

• tf.layers.average_pooling1d
• tf.layers.average_pooling2d
• tf.layers.average_pooling3d
• tf.layers.max_pooling1d
• tf.layers.max_pooling2d
• tf.layers.max_pooling3d

以max_pooling3d函数为例

max_pooling3d(
inputs,
pool_size,
strides,
padding=’valid’,
data_format=’channels_last’,
name=None
)

需要注意的参数有：

• inputs: 输入的张量
• pool_size: 3个整数的列表或者元组对应于 (pool_depth, pool_height, pool_width)上的大小，或者一个整数指定这三个。
• strides: 在3个维度上的移动的步长
• padding：同上

全连接

tf.layers.dense

dense(
inputs,
units,
activation=None,
use_bias=True,
kernel_initializer=None,
bias_initializer=tf.zeros_initializer(),
kernel_regularizer=None,
bias_regularizer=None,
activity_regularizer=None,
trainable=True,
name=None,
reuse=None
)

• inputs： 输入的张量
• units: 输出channel数
• activation: 激活函数

输出为：activation(inputs.kernel + bias)

dropout

tf.layers.dropout

dropout(
inputs,
rate=0.5,
noise_shape=None,
seed=None,
training=False,
name=None
)

• inputs：输入的张量
• training: True时会加入dropout,false即测试时不会加入dropout
• rate: 丢失率

BN

tf.layers.batch_normalization

batch_normalization(
inputs,
axis=-1,
momentum=0.99,
epsilon=0.001,
center=True,
scale=True,
beta_initializer=tf.zeros_initializer(),
gamma_initializer=tf.ones_initializer(),
moving_mean_initializer=tf.zeros_initializer(),
moving_variance_initializer=tf.ones_initializer(),
beta_regularizer=None,
gamma_regularizer=None,
training=False,
trainable=True,
name=None,
reuse=None,
renorm=False,
renorm_clipping=None,
renorm_momentum=0.99
)

• inputs: 输入的张量

Reference:
tf.layers