Enum Initializer

pub enum Initializer {
    Zeros,
    Ones,
    Uniform {
        min: f64,
        max: f64,
    },
    Normal {
        mean: f64,
        std: f64,
    },
    Xavier,
    He,
    LeCun,
}

Initialization strategies for neural network weights

Variants

Zeros

Zero initialization

Ones

One initialization

Uniform

Uniform random initialization

Fields

min: f64

Minimum value

max: f64

Maximum value

Normal

Normal random initialization

Fields

mean: f64

Mean

std: f64

Standard deviation

Xavier

Xavier/Glorot initialization

He

He initialization

LeCun

LeCun initialization

Implementations

impl Initializer

pub fn initialize<F: Float + Debug, R: Rng>( &self, shape: IxDyn, fan_in: usize, fan_out: usize, rng: &mut R, ) -> Result<Array<F, IxDyn>>

Initialize weights according to the strategy

Arguments

• shape - Shape of the weights array
• fan_in - Number of input connections (for Xavier, He, LeCun)
• fan_out - Number of output connections (for Xavier)
• rng - Random number generator

Returns

• Initialized weights array

Trait Implementations

impl Clone for Initializer

fn clone(&self) -> Initializer

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Initializer

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

Formats the value using the given formatter.

impl Copy for Initializer