Struct Adam

pub struct Adam<AB: AutodiffBackend, const D: usize> {
    pub config: AdamConfig,
    pub record: AdamRecord<AB::InnerBackend, D>,
}

Adam optimizer as described in the paper: “Adam: A Method for Stochastic Optimization”.

Fields

config: AdamConfig

Configuration.

record: AdamRecord<AB::InnerBackend, D>

Record.

Implementations

impl<AB: AutodiffBackend, const D: usize> Adam<AB, D>

pub fn update( &mut self, learning_rate: f64, value: Tensor<AB, D>, grad: Tensor<AB::InnerBackend, D>, ) -> Tensor<AB, D>

Optimize a value using Adam optimizer.

Arguments

• learning_rate - The number to multiply the gradient by.
• value - The value to optimize.
• grad - The gradient of the value.

Returns

The optimized value.

pub fn to_device(self, device: &AB::Device) -> Self

Transfer the optimizer to a device.

pub fn load_record( &mut self, record: AdamRecord<AB::InnerBackend, D>, ) -> &mut Self

Load a record into the optimizer.

pub fn into_record(self) -> AdamRecord<AB::InnerBackend, D>

Unload the record from the optimizer.

Trait Implementations

impl<AB: Clone + AutodiffBackend, const D: usize> Clone for Adam<AB, D>

where AB::InnerBackend: Clone,

fn clone(&self) -> Adam<AB, D>

Returns a copy of the value.

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

Performs copy-assignment from source.

impl<AB: Debug + AutodiffBackend, const D: usize> Debug for Adam<AB, D>

where AB::InnerBackend: Debug,

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

Formats the value using the given formatter.

impl<AB: AutodiffBackend, const D: usize> Default for Adam<AB, D>

fn default() -> Self

Returns the “default value” for a type.