Enum tch::nn::init::Init

pub enum Init {
    Const(f64),
    Randn {
        mean: f64,
        stdev: f64,
    },
    Uniform {
        lo: f64,
        up: f64,
    },
    Kaiming {
        dist: NormalOrUniform,
        fan: FanInOut,
        non_linearity: NonLinearity,
    },
    Orthogonal {
        gain: f64,
    },
}

Variable initializations.

Variants

Const(f64)

Constant value.

Randn

Fields

mean: f64

stdev: f64

Random normal with some mean and standard deviation.

Uniform

Fields

lo: f64

up: f64

Uniform initialization between some lower and upper bounds.

Kaiming

Fields

dist: NormalOrUniform

fan: FanInOut

non_linearity: NonLinearity

Kaiming uniform initialization. See “Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification” He, K. et al. (2015). This uses a uniform distribution.

Orthogonal

Fields

gain: f64

Orthogonal initialization

Implementations

impl Init

pub fn set(self, tensor: &mut Tensor)

Re-initializes an existing tensor with the specified initialization

Trait Implementations

impl Clone for Init

fn clone(&self) -> Init

Returns a copy of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Init

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Copy for Init