Explicit activation stage
Add a named activation as an explicit layer after another layer in a model.
Keras · Class
layers.Activation
Keras layers.Activation applies a selected activation function as a layer. Choose a built-in activation by name or provide a callable.
How it works
Activation wraps a chosen activation function as a Keras layer. The function can be selected by its built-in string name or supplied directly as a callable, and the layer applies it to the input while preserving the input shape.
Syntax
keras.layers.Activation(activation, **kwargs)Parameters
| Name | Type | Description |
|---|---|---|
activation | callable or string | Activation function, supplied as a callable or as the name of an activation in the keras.activations namespace. |
**kwargs | keyword arguments | Base layer keyword arguments, such as name and dtype. |
Returns
A tensor produced by applying the selected activation function to the input; the documented example shows the transformed output.
Key points
- Activation is a Keras layer that applies an activation function to an output.
- The activation can be either a callable or a built-in activation name.
- Additional keyword arguments are forwarded as standard Layer options, including name and dtype.
- A built-in activation can be selected with the string name
relu. - The corresponding activation function can instead be passed directly as a callable.
Common mistakes
- A string must identify an activation available in the
keras.activationsnamespace; otherwise pass a callable.
Example: Apply ReLU with an Activation layer
import keras
import numpy as np
layer = keras.layers.Activation("relu")
output = layer(np.array([-3.0, -1.0, 0.0, 2.0]))
print(output)Creates a ReLU activation layer and applies it element-wise to a NumPy array.
When to use it
Built-in activation
Use a built-in activation such as ReLU without writing the function manually.
Custom callable
Apply a user-defined tensor transformation supplied as a callable.
Interview questions
QWhat can the activation argument contain?
The constructor accepts an activation and arbitrary base-layer keyword arguments. The activation may be a callable or the name of an activation in the keras.activations namespace.
Follow-up: How would you express the same operation in a Dense layer? Pass the activation through Dense's activation argument instead of adding a separate Activation layer.
QWhat does keras.layers.Activation do?
The layer applies the selected activation function to its input output. For example, a ReLU layer maps negative values to zero, as shown by the documented example.
Follow-up: Can the activation be custom? Yes, provide a callable that accepts a tensor and returns the transformed tensor.
QHow does Activation compare with using Dense(..., activation=...)?
Activation is useful when the activation should be represented as its own layer. Keras also supports passing activations through the activation argument of forward layers such as Dense.
Follow-up: When is the separate layer especially useful? It makes the transformation explicit in a sequential architecture or lets it be inserted after another layer.
Related
keras.activations.relukeras.layers.Densekeras.layers.ReLU