Struct LBFGSState 

pub struct LBFGSState {
    pub history_s: Vec<Tensor<1>>,
    pub history_y: Vec<Tensor<1>>,
    pub d: Option<Tensor<1>>,
    pub t: Option<f64>,
    pub prev_flat_grad: Option<Tensor<1>>,
    pub prev_loss: Option<f64>,
    pub g_iter: usize,
}

L-BFGS optimizer state

Fields

history_s: Vec<Tensor<1>>

Historical displacement vectors

history_y: Vec<Tensor<1>>

Historical gradient difference vectors

d: Option<Tensor<1>>

Search direction

t: Option<f64>

Step size from the previous iteration

prev_flat_grad: Option<Tensor<1>>

Flattened gradient from the previous iteration

prev_loss: Option<f64>

Loss value from the previous iteration

g_iter: usize

Global iteration count

Implementations

impl LBFGSState

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

Moves all historical tensors to the target device.

Trait Implementations

impl Clone for LBFGSState

fn clone(&self) -> LBFGSState

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Default for LBFGSState

fn default() -> Self

Returns the “default value” for a type.

impl RecordState for LBFGSState

fn state_flatten(&self, prefix: &str, out: &mut StateSink)

Flatten self into out, naming every leaf under prefix.

fn state_unflatten( prefix: &str, src: &mut StateSource, device: &Device, ) -> Option<Self>

Rebuild a value from src, reading leaves named under prefix.