Struct NtmConfig 

pub struct NtmConfig<A: Float + ScalarOperand + Debug> {
    pub memory_slots: usize,
    pub memory_width: usize,
    pub learning_rate: A,
    pub read_sharpness: A,
    pub erase_gate: A,
    pub addressing_mode: AddressingMode,
    pub memory_weight: A,
    pub gradient_weight: A,
    pub seed: u64,
}

Configuration for an NtmOptimizer.

The defaults mirror the values recommended in the project specification: 32 slots × 16-wide memory, learning rate 0.01, hybrid addressing, read_sharpness = 1, erase_gate = 0.5, and 0.7 / 0.3 weighting between the raw gradient and the memory read vector.

Fields

memory_slots: usize

Number of memory slots N (rows of the memory matrix).

memory_width: usize

Width of each memory slot W (columns of the memory matrix).

learning_rate: A

Base learning rate applied to the final update.

read_sharpness: A

Read-attention sharpness factor β (the NTM paper’s β).

erase_gate: A

Erase gate (a scalar in [0, 1]) controlling how aggressively the write head zeroes a memory cell before adding the new content.

addressing_mode: AddressingMode

Addressing mode for the read head.

memory_weight: A

Weight assigned to the (tiled) read vector when forming the update.

gradient_weight: A

Weight assigned to the raw gradient when forming the update.

seed: u64

RNG seed used for any stochastic initialisation. Memory itself is initialised to zero; the seed is kept so subclasses/tests can rely on deterministic behaviour.

Trait Implementations

impl<A: Clone + Float + ScalarOperand + Debug> Clone for NtmConfig<A>

fn clone(&self) -> NtmConfig<A>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<A: Debug + Float + ScalarOperand + Debug> Debug for NtmConfig<A>

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

Formats the value using the given formatter.

impl<A: Float + ScalarOperand + Debug> Default for NtmConfig<A>

fn default() -> Self

Returns the “default value” for a type.