View source on GitHub |
Mish: A Self Regularized Non-Monotonic Neural Activation Function.
tfa.activations.mish(
x: tfa.types.TensorLike
) -> tf.Tensor
Computes mish activation:
\[ \mathrm{mish}(x) = x \cdot \tanh(\mathrm{softplus}(x)). \]
See Mish: A Self Regularized Non-Monotonic Neural Activation Function.
Usage:
x = tf.constant([1.0, 0.0, 1.0])
tfa.activations.mish(x)
<tf.Tensor: shape=(3,), dtype=float32, numpy=array([0.865098..., 0. , 0.865098...], dtype=float32)>
Args | |
---|---|
x
|
A Tensor . Must be one of the following types:
bfloat16 , float16 , float32 , float64 .
|
Returns | |
---|---|
A Tensor . Has the same type as x .
|