Struct NtmOptimizerConfig 

pub struct NtmOptimizerConfig {
    pub num_slots: usize,
    pub num_locations: usize,
    pub mem_width: usize,
    pub hidden_dim: usize,
    pub shift_range: usize,
    pub mem_init_scale: f64,
    pub base_lr: f64,
    pub momentum_decay: f64,
    pub rms_decay: f64,
    pub max_grad_norm: f64,
    pub seed: u64,
}

Configuration for NtmOptimizer.

Numeric hyper-parameters are stored as f64 and converted to the optimizer’s scalar type T when the optimizer is constructed.

Fields

num_slots: usize

Requested number of parameter chunks (clamped to the parameter count).

num_locations: usize

Number of memory locations (rows of the memory matrix M).

mem_width: usize

Width of each memory location (columns of M); also the read/key width.

hidden_dim: usize

Controller hidden-layer dimension.

shift_range: usize

Maximum convolutional shift magnitude; allowed shifts are -shift_range ..= shift_range, giving 2·shift_range + 1 entries.

mem_init_scale: f64

Half-width of the uniform interval used to initialise the memory matrix.

base_lr: f64

Base learning rate applied before the learned per-chunk step scale.

momentum_decay: f64

EMA decay for the gradient (momentum), often called beta1.

rms_decay: f64

EMA decay for the squared gradient (RMS), often called beta2.

max_grad_norm: f64

Maximum gradient L2 norm; larger gradients are rescaled before use.

seed: u64

Seed for deterministic weight and memory initialisation.

Trait Implementations

impl Clone for NtmOptimizerConfig

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for NtmOptimizerConfig

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

Formats the value using the given formatter.

impl Default for NtmOptimizerConfig

fn default() -> Self

Returns the “default value” for a type.