xavier_uniform

shrew::nn::init

Function xavier_uniform 

Source 
pub fn xavier_uniform<B>(
    shape: impl Into<Shape>,
    gain: f64,
    dtype: DType,
    device: &<B as Backend>::Device,
) -> Result<Tensor<B>, Error>
where
    B: Backend,
Expand description

Xavier (Glorot) uniform initialization.

Draws from U(-a, a) where a = gain * sqrt(6 / (fan_in + fan_out)). Designed to keep variance constant across layers with linear activations.

ยงArguments

  • gain: scaling factor (1.0 for linear/sigmoid, sqrt(2) for ReLU)