hard_sigmoid

hard_sigmoid(x: &Tensor<T>) -> Result<Tensor<C>, TensorError>

Compute $\large \text{max}(0, \text{min}(1, \frac{x}{6} + 0.5))$ for all elements. A piece-wise linear approximation of the sigmoid function.

Parameters:

x: Input values

Returns:

Tensor with type C

Examples:

use hpt::{ops::FloatUnaryOps, Tensor, error::TensorError};

fn main() -> Result<(), TensorError> {
    let a = Tensor::<f32>::new([2.0]);
    let b = a.hard_sigmoid()?;
    println!("{}", b);  // prints: 0.8333333
    Ok(())
}

Backend Support

Backend	Supported
CPU	✅
Cuda	✅