Layers
Common Methods
layer.get_weights()
: 以含有Numpy矩阵的列表形式返回层的权重。layer.set_weights(weights)
: 从含有Numpy矩阵的列表中设置层的权重(与get_weights
的输出形状相同)。layer.get_config()
: 返回包含层配置的字典。
如果一个层具有单节点,得到各项参数:
layer.input
layer.output
layer.input_shape
layer.output_shape
如果为多节点:
layer.get_input_at(node_index)
layer.get_output_at(node_index)
layer.get_input_shape_at(node_index)
layer.get_output_shape_at(node_index)
Core Layers(网络核心层)
Dense—— densely-connect NN layer
Activation
Dropout
Flatten——flattens the input
Input
Reshape
Permute ——permutes the dimensions of the input according to a given pattern
RepeatVector
Lambda—— warps arbitrary expression as a Layer object
ActivityRegularization—— applies an update to the cost function based input activity
Masking——skip the timesteps
SpatialDropout1D
SpatialDropout2D
SpatialDropout3D
Convolutional Layers(卷积层)
Conv1D
Conv2D
SeparableConv1D(可分离)
SeparableConv2D
DepthwiseConv2D
Conv2DTranspose(反卷积)
Conv3D
Conv3DTranspose
Cropping1D(裁剪层)
Cropping2D
Cropping3D
UpSampling1D(上采样层)
UpSampling2D
UpSampling3D
ZeroPadding1D(零填充层)
ZeroPadding2D
ZeroPadding3D
Pooling Layers(池化层)
MaxPooling1D
MaxPooling2D
MaxPooling3D
AveragePooling1D
AveragePooling2D
AveragePooling3D
GlobalMaxPooling1D
GlobalAveragePooling1D
GlobalMaxPooling2D
GlobalAveragePooling2D
GlobalMaxPooling3D
GlobalAveragePooling3D
Locally-connected Layers(局部连接层)
LocallyConnected1D
LocallyConnected2D
Recurrent Layers(循环层)
RNN
SimpleRNN
GRU (Gated Recurrent Unit)
LSTM (Long Short-Term Memory)
ConvLSTM2D
SimpleRNNCell
GRUCell
LSTMCell
CuDNNGRU—— Fast GRU implementation backed by CuDNN(only on GPU)
CuDNNLSTM
Embedding Layers(嵌入层)
Embedding—— Turns positive integers (indexes) into dense vectors of fixed size.
Merge Layers(融合层)
Add
Subtract
Multiply
Average
Maximum
Concatenate
Dot(点积)
add
subtract
multiply
average
maximum
concatenate
dot
Advanced Activations Layers(高级激活层)
LeakyReLU( Leaky version of a Rectified Linear Unit )
PReLU( Parametric Rectified Linear Unit )
ELU( Exponential Linear Unit )
ThresholdedReLU( Thresholded Rectified Linear Unit )
Softmax
ReLU( Rectified Linear Unit )
Normalization Layers(标准化层)
BatchNormalization
Noise Layers
GaussianNoise
GaussianDropout
AlphaDropout
Layer warppers(封装器)
TimeDistributed
Bidirectional