Keras layers API
- 원본 링크 : https://keras.io/api/layers/
- 최종 확인 : 2024-11-25
Layers are the basic building blocks of neural networks in Keras.
A layer consists of a tensor-in tensor-out computation function (the layer’s call method)
and some state, held in TensorFlow variables (the layer’s weights).
A Layer instance is callable, much like a function:
import keras
from keras import layers
layer = layers.Dense(32, activation='relu')
inputs = keras.random.uniform(shape=(10, 20))
outputs = layer(inputs)Unlike a function, though, layers maintain a state, updated when the layer receives data
during training, and stored in layer.weights:
>>> layer.weights
[<KerasVariable shape=(20, 32), dtype=float32, path=dense/kernel>,
<KerasVariable shape=(32,), dtype=float32, path=dense/bias>]
Creating custom layers
While Keras offers a wide range of built-in layers, they don’t cover ever possible use case. Creating custom layers is very common, and very easy.
See the guide
Making new layers and models via subclassing
for an extensive overview, and refer to the documentation for the base Layer class.
Layers API overview
The base Layer class
- Layer class
- weights property
- trainable_weights property
- non_trainable_weights property
- add_weight method
- trainable property
- get_weights method
- set_weights method
- get_config method
- add_loss method
- losses property
Layer activations
- relu function
- sigmoid function
- softmax function
- softplus function
- softsign function
- tanh function
- selu function
- elu function
- exponential function
- leaky_relu function
- relu6 function
- silu function
- hard_silu function
- gelu function
- hard_sigmoid function
- linear function
- mish function
- log_softmax function
Layer weight initializers
- RandomNormal class
- RandomUniform class
- TruncatedNormal class
- Zeros class
- Ones class
- GlorotNormal class
- GlorotUniform class
- HeNormal class
- HeUniform class
- Orthogonal class
- Constant class
- VarianceScaling class
- LecunNormal class
- LecunUniform class
- IdentityInitializer class
Layer weight regularizers
Layer weight constraints
Core layers
- Input object
- InputSpec object
- Dense layer
- EinsumDense layer
- Activation layer
- Embedding layer
- Masking layer
- Lambda layer
- Identity layer
Convolution layers
- Conv1D layer
- Conv2D layer
- Conv3D layer
- SeparableConv1D layer
- SeparableConv2D layer
- DepthwiseConv1D layer
- DepthwiseConv2D layer
- Conv1DTranspose layer
- Conv2DTranspose layer
- Conv3DTranspose layer
Pooling layers
- MaxPooling1D layer
- MaxPooling2D layer
- MaxPooling3D layer
- AveragePooling1D layer
- AveragePooling2D layer
- AveragePooling3D layer
- GlobalMaxPooling1D layer
- GlobalMaxPooling2D layer
- GlobalMaxPooling3D layer
- GlobalAveragePooling1D layer
- GlobalAveragePooling2D layer
- GlobalAveragePooling3D layer
Recurrent layers
- LSTM layer
- LSTM cell layer
- GRU layer
- GRU Cell layer
- SimpleRNN layer
- TimeDistributed layer
- Bidirectional layer
- ConvLSTM1D layer
- ConvLSTM2D layer
- ConvLSTM3D layer
- Base RNN layer
- Simple RNN cell layer
- Stacked RNN cell layer
Preprocessing layers
- Text preprocessing
- Numerical features preprocessing layers
- Categorical features preprocessing layers
- Image preprocessing layers
- Image augmentation layers
- Audio preprocessing layers
Normalization layers
Regularization layers
- Dropout layer
- SpatialDropout1D layer
- SpatialDropout2D layer
- SpatialDropout3D layer
- GaussianDropout layer
- AlphaDropout layer
- GaussianNoise layer
- ActivityRegularization layer
Attention layers
Reshaping layers
- Reshape layer
- Flatten layer
- RepeatVector layer
- Permute layer
- Cropping1D layer
- Cropping2D layer
- Cropping3D layer
- UpSampling1D layer
- UpSampling2D layer
- UpSampling3D layer
- ZeroPadding1D layer
- ZeroPadding2D layer
- ZeroPadding3D layer
Merging layers
- Concatenate layer
- Average layer
- Maximum layer
- Minimum layer
- Add layer
- Subtract layer
- Multiply layer
- Dot layer