add_mating_data.py

"""Annotate a .bag file that stores (FEN, move, win-probability) triples with mate-in-N information produced by Stockfish.

Mate labels:
    "#"   – the move itself gives immediate checkmate
    +N    – mover can force mate in N plies
    -N    – mover will be mated in N plies
    "-"   – no forced mate detected within engine depth / time limit
"""

import math
import os
import tempfile
from multiprocessing import Pool, cpu_count
from pathlib import Path

import chess
import chess.engine
import psutil
from tqdm import tqdm

from athena.datasets.chessbenchmate.utils.bagz import BagReader, BagWriter
from athena.datasets.chessbenchmate.utils.constants import CODERS

ENGINE_PATH = "models/stockfish"
ENGINE_LIMIT = chess.engine.Limit(time=0.05)

# Global variable for each worker process
engine = None


def init_worker():
    """Initialize engine for each worker process."""
    global engine
    engine = chess.engine.SimpleEngine.popen_uci(ENGINE_PATH)
    # Configure engine for faster analysis
    engine.configure({"Threads": 1})


def close_worker():
    """Close engine when worker finishes."""
    global engine
    if engine is not None:
        try:
            engine.close()
        except Exception:
            pass

def print_memory_usage(description=""):
    """Print current and total available memory usage."""
    process = psutil.Process(os.getpid())
    memory_mb = process.memory_info().rss / 1024 / 1024
    total_mem_mb = psutil.virtual_memory().total / 1024 / 1024
    print(f"{description}Memory usage: {memory_mb:.2f} MB / {total_mem_mb:.2f} MB total", flush=True)


def annotate_single_record(record: bytes) -> bytes:
    """Annotate a single record with mate-in-N information."""
    global engine
    assert engine is not None, "Engine not initialized in worker"
    try:
        fen, move_str, _, _ = CODERS["action_value_with_mate"].decode(record)
        board = chess.Board(fen)
        win_prob, mate_label = stockfish_evaluate(board, chess.Move.from_uci(move_str), engine)
        assert 0.0 <= win_prob <= 1.0, f"Win probability should be between 0 and 1, got {win_prob}"

        # Validate mate_label format
        valid_mate_label = (
            mate_label == "#"
            or mate_label == "-"
            or (isinstance(mate_label, int) and abs(mate_label) <= 100)  # Reasonable mate distance
        )
        assert valid_mate_label, f"Invalid mate label: {mate_label}"

        return CODERS["action_value_with_mate"].encode((fen, move_str, win_prob, mate_label))
    except Exception as e:
        print(f"Error processing record: {e}")
        raise


def stockfish_evaluate(
    board: chess.Board, move: chess.Move, engine: chess.engine.SimpleEngine
) -> tuple[float, str | int]:
    """Evaluate a position after a given move using Stockfish.

    Returns:
        win_prob: float between 0 and 1
        mate_label: str or int indicating mate status
    """
    mover = board.turn
    board.push(move)

    if board.is_checkmate():
        return 1.0, "#"

    info = engine.analyse(board, ENGINE_LIMIT)
    score = info.get("score")
    assert score is not None, "Engine analysis did not return a score"

    pov_score = score.pov(mover)

    if pov_score.is_mate():
        mate_val = pov_score.mate()
        if mate_val is not None:
            # Positive mate_val means mover can force mate, negative means opponent can
            win_prob = 1.0 if mate_val > 0 else 0.0
            mate_label = mate_val
        else:
            win_prob = 0.5  # Should not happen with is_mate(), but fallback
            mate_label = "-"
    else:
        cp = pov_score.score()
        assert cp is not None, "Centipawn score should not be None for non-mate scores"
        win_prob = 1 / (1 + math.exp(-cp / 173.718))
        win_prob = min(max(win_prob, 1e-6), 1 - 1e-6)  # clamp safely
        mate_label = "-"

    # Additional validation
    assert 0.0 <= win_prob <= 1.0, f"Win probability out of bounds: {win_prob}"

    return win_prob, mate_label


def add_mate_annotations(bag_path: Path) -> None:
    """Annotate a .bag file with mate-in-N information in-place."""
    if not bag_path.exists():
        print(f"File not found: {bag_path}")
        return

    reader = BagReader(str(bag_path))
    records = list(reader)

    if len(records) == 0:
        print(f"No records found in {bag_path}")
        return

    # Create temporary file for writing
    with tempfile.NamedTemporaryFile(delete=False, suffix=".bag", dir=bag_path.parent) as temp_file:
        temp_path = temp_file.name

    try:
        writer = BagWriter(temp_path)

        print(
            f"Processing {len(records)} records from {bag_path.name} using {cpu_count()} processes"
        )

    
        # Single record processing approach
        with Pool(processes=cpu_count(), initializer=init_worker) as pool:
            for i, annotated_record in tqdm(enumerate(
                pool.imap(annotate_single_record, records)),
                total=len(records),
                unit="record",
                desc=f"Annotating {bag_path.name}",
            ):
                writer.write(annotated_record)
                if i % 10000 == 0:
                    print_memory_usage(f"Processed {i} records. ")

        writer.close()

        # Replace original file with the temporary file
        os.replace(temp_path, bag_path)
        print(f"Successfully annotated {bag_path} with {len(records)} records")

    except Exception as e:
        # Clean up temporary file on error
        if os.path.exists(temp_path):
            os.unlink(temp_path)
        print(f"Error processing {bag_path}: {e}")
        raise e


def main():
    """Main entry point to process .bag files with optional SLURM array division."""
    data_dir = Path("src/athena/datasets/chessbenchmate/data/train")
    bag_files = sorted(list(data_dir.glob("*.bag")))
    num_files = len(bag_files)
    print(f"Found {num_files} bag files in {data_dir}")

    # Detect SLURM environment variables
    slurm_task_id = int(os.environ.get("SLURM_ARRAY_TASK_ID", "0"))
    slurm_array_size = int(os.environ.get("SLURM_ARRAY_TASK_COUNT", "1"))

    # Divide files evenly among SLURM array tasks
    chunk_size = (num_files + slurm_array_size - 1) // slurm_array_size
    start = slurm_task_id * chunk_size
    end = min(start + chunk_size, num_files)
    my_files = bag_files[start:end]

    print(
        f"[Job {slurm_task_id}/{slurm_array_size}] Processing {len(my_files)} files "
        f"from index {start} to {end - 1}"
    )

    for bag_file in tqdm(my_files, total=len(my_files), desc=f"Job {slurm_task_id}"):
        add_mate_annotations(bag_file)

    print(f"[Job {slurm_task_id}] Completed {len(my_files)} files successfully.")


if __name__ == "__main__":
    main()