Struct TrainingDataset

pub struct TrainingDataset {
    pub data: Vec<TrainingData>,
}

Training dataset manager

Fields

data: Vec<TrainingData>

Implementations

impl TrainingDataset

pub fn new() -> Self

pub fn load_from_pgn<P: AsRef<Path>>( &mut self, path: P, max_games: Option<usize>, max_moves_per_game: usize, ) -> Result<(), Box<dyn Error>>

Load positions from a PGN file with parallel processing

pub fn evaluate_with_external_engine( &mut self, depth: u8, ) -> Result<(), Box<dyn Error>>

Evaluate all positions using external engine

pub fn evaluate_with_external_engine_parallel( &mut self, depth: u8, num_threads: usize, ) -> Result<(), Box<dyn Error>>

Evaluate all positions using external engine in parallel

pub fn train_engine(&self, engine: &mut ChessVectorEngine)

Train the vector engine with this dataset

pub fn split(&self, train_ratio: f32) -> (TrainingDataset, TrainingDataset)

Split dataset into train/test sets by games to prevent data leakage

pub fn save<P: AsRef<Path>>(&self, path: P) -> Result<(), Box<dyn Error>>

Save dataset to file

pub fn load<P: AsRef<Path>>(path: P) -> Result<Self, Box<dyn Error>>

Load dataset from file

pub fn load_and_append<P: AsRef<Path>>( &mut self, path: P, ) -> Result<(), Box<dyn Error>>

Load and append data from file to existing dataset (incremental training)

pub fn merge(&mut self, other: TrainingDataset)

Merge another dataset into this one

pub fn save_incremental<P: AsRef<Path>>( &self, path: P, ) -> Result<(), Box<dyn Error>>

Save incrementally (append to existing file if it exists)

pub fn save_incremental_with_options<P: AsRef<Path>>( &self, path: P, deduplicate: bool, ) -> Result<(), Box<dyn Error>>

Save incrementally with option to skip deduplication

pub fn save_append_only<P: AsRef<Path>>( &self, path: P, ) -> Result<(), Box<dyn Error>>

Fast append-only save (no deduplication, just append new positions)

pub fn add_position( &mut self, board: Board, evaluation: f32, depth: u8, game_id: usize, )

Add a single training data point

pub fn next_game_id(&self) -> usize

Get the next available game ID for incremental training

pub fn deduplicate(&mut self, similarity_threshold: f32)

Remove near-duplicate positions to reduce overfitting

pub fn deduplicate_fast(&mut self)

Fast hash-based deduplication for exact duplicates (O(n))

pub fn merge_and_deduplicate(&mut self, new_data: Vec<TrainingData>)

Streaming deduplication when merging with existing data (faster for large datasets)

pub fn deduplicate_parallel( &mut self, similarity_threshold: f32, chunk_size: usize, )

Remove near-duplicate positions using parallel comparison (faster for large datasets)

Trait Implementations

impl Default for TrainingDataset

fn default() -> Self

Returns the “default value” for a type.