Struct Softplus 

pub struct Softplus<D: Float, T, const RANK: usize> {
    pub beta: f64,
    pub threshold: f64,
    /* private fields */
}

Softplus activation layer: a smooth ReLU approximation, (1/beta) * ln(1 + e^(beta*x)), reverting to linear above threshold for numerical stability.

Fields

beta: f64

The beta sharpness parameter.

threshold: f64

Above this value, the layer reverts to linear (x) to avoid overflow in exp.

Implementations

impl<D: Float, T, const RANK: usize> Softplus<D, T, RANK>

pub fn new(beta: f64, threshold: f64) -> Self

Creates a new Softplus layer with the given beta/threshold.

Trait Implementations

impl<D: Float, const RANK: usize> Layer<SymTensor> for Softplus<D, SymTensor, RANK>

type Output = SymTensor

This layer’s output type.

fn call(&self, input: SymTensor) -> SymTensor

Applies this layer to input.

impl<D: Float, T: Tensor<D, RANK> + EagerTensor, const RANK: usize> Layer<T> for Softplus<D, T, RANK>

type Output = T

This layer’s output type.

fn call(&self, _input: T) -> Self::Output

Applies this layer to input.