Struct ActivationCache 

pub struct ActivationCache<L: LinAlg = CpuLinAlg> { /* private fields */ }

Cache for hidden layer activations captured during inference.

Used by CCA neuron alignment during crossover to compare functional neuron responses between parent networks. Activations are recorded during normal fitness evaluation at zero additional compute cost.

Generic over a LinAlg backend L. Defaults to CpuLinAlg.

Examples

use pc_rl_core::pc_actor_critic::ActivationCache;
use pc_rl_core::linalg::cpu::CpuLinAlg;

let cache: ActivationCache<CpuLinAlg> = ActivationCache::new(2);
assert_eq!(cache.batch_size(), 0);
assert_eq!(cache.num_layers(), 2);

Implementations

impl<L: LinAlg> ActivationCache<L>

pub fn new(num_layers: usize) -> Self

Creates a new empty cache with the given number of hidden layers.

Arguments

• num_layers - Number of hidden layers to track.

pub fn batch_size(&self) -> usize

Returns the number of samples recorded so far.

pub fn num_layers(&self) -> usize

Returns the number of hidden layers in the cache.

pub fn record(&mut self, hidden_states: &[L::Vector])

Records hidden layer activations from a single inference.

Arguments

• hidden_states - Per-layer activations from InferResult::hidden_states.

pub fn layer(&self, layer_idx: usize) -> &[L::Vector]

Returns the recorded activations for a given layer.

Arguments

• layer_idx - Index of the hidden layer.

Trait Implementations

impl<L: Clone + LinAlg> Clone for ActivationCache<L>

where L::Vector: Clone,

fn clone(&self) -> ActivationCache<L>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<L: Debug + LinAlg> Debug for ActivationCache<L>

where L::Vector: Debug,

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

Formats the value using the given formatter.