Spaces:
Running on Zero
Running on Zero
| """Analyse a position: `python -m laya_chess.cli --fen "<FEN>" --nodes 50`.""" | |
| import argparse | |
| import time | |
| import chess | |
| from .model import DEFAULT_CHECKPOINT, LayaChessModel | |
| from .search import MCTS, winprob_to_cp | |
| def main(): | |
| p = argparse.ArgumentParser(description="Analyse a chess position with LayaChess") | |
| p.add_argument("--fen", default=chess.STARTING_FEN) | |
| p.add_argument("--checkpoint", default=DEFAULT_CHECKPOINT) | |
| p.add_argument("--revision", default=None) | |
| p.add_argument("--device", default="auto") | |
| p.add_argument("--nodes", type=int, default=0, help="network evaluations to search (0 = network only)") | |
| p.add_argument("--seconds", type=float, default=None) | |
| p.add_argument("--top", type=int, default=8) | |
| p.add_argument("--batch-leaves", type=int, default=4) | |
| a = p.parse_args() | |
| model = LayaChessModel(a.checkpoint, revision=a.revision, device=a.device) | |
| board = chess.Board(a.fen) | |
| print(board, "\n") | |
| t = time.time() | |
| scored = model.score_moves(board) | |
| dt = time.time() - t | |
| print(f"network only ({len(scored)} moves in {dt:.2f}s):") | |
| for mv, q in scored[:a.top]: | |
| print(f" {board.san(mv):8s} win chance {q:5.1%}") | |
| if a.nodes or a.seconds: | |
| r = MCTS(model, batch_leaves=a.batch_leaves).search(board, nodes=a.nodes or None, seconds=a.seconds) | |
| print(f"\nsearch: {r.nodes} positions, {r.simulations} simulations, depth {r.depth}, {r.seconds:.1f}s " | |
| f"({r.nodes / max(r.seconds, 1e-9):.2f} positions/s)") | |
| print(f" best {board.san(r.move)} value {r.value:.1%} (cp {winprob_to_cp(r.value)}) pv {board.variation_san(r.pv)}") | |
| for mv, q, n in r.scores[:a.top]: | |
| print(f" {board.san(mv):8s} net {q:5.1%} visits {n}") | |
| if __name__ == "__main__": | |
| main() | |