Struct ReplayBuffer 

pub struct ReplayBuffer { /* private fields */ }

Experience Replay Buffer for storing and sampling past training examples

Implementations

impl ReplayBuffer

pub fn new(capacity: usize) -> Self

Create a new replay buffer with specified capacity

Arguments

• capacity - Maximum number of entries to store

pub fn add(&mut self, query: &[f32], positive_ids: &[usize])

Add a new entry to the buffer using reservoir sampling

Reservoir sampling ensures uniform distribution over all samples seen, even as old samples are evicted due to capacity constraints.

Arguments

• query - Query vector
• positive_ids - IDs of positive nodes for this query

pub fn sample(&self, batch_size: usize) -> Vec<&ReplayEntry>

Sample a batch of entries uniformly at random

Arguments

• batch_size - Number of entries to sample

Returns

Vector of references to sampled entries (may be smaller than batch_size if buffer is small)

pub fn detect_distribution_shift(&self, recent_window: usize) -> f32

Detect distribution shift between recent samples and overall distribution

Uses Kullback-Leibler divergence approximation based on mean and variance changes.

Arguments

• recent_window - Number of most recent samples to compare

Returns

Shift score (higher values indicate more significant distribution shift) Returns 0.0 if insufficient data

pub fn len(&self) -> usize

Get the number of entries currently in the buffer

pub fn is_empty(&self) -> bool

Check if the buffer is empty

pub fn capacity(&self) -> usize

Get the total capacity of the buffer

pub fn total_seen(&self) -> usize

Get the total number of samples seen (including evicted ones)

pub fn distribution_stats(&self) -> &DistributionStats

Get a reference to the distribution statistics

pub fn clear(&mut self)

Clear all entries from the buffer