Update README.md
Browse files
README.md
CHANGED
@@ -23,3 +23,102 @@ configs:
|
|
23 |
- split: train
|
24 |
path: data/train-*
|
25 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
- split: train
|
24 |
path: data/train-*
|
25 |
---
|
26 |
+
|
27 |
+
# 100M Random Chess Boards
|
28 |
+
|
29 |
+
100M randomly sampled chess boards. 100M games were played, with each move being randomly determined until completion (termination). Then a single random board state within the game is sampled.
|
30 |
+
|
31 |
+
- **fen**: The FEN representation of the randomly sampled board
|
32 |
+
- **n_moves**: The number of moves into the game the sampled FEN is.
|
33 |
+
- **term_n_moves**: The total number of moves in the random-walk game from which the board was sampled.
|
34 |
+
- **term_reason**: How the random-walk game ended. [Codes](https://python-chess.readthedocs.io/en/latest/core.html#chess.Termination)
|
35 |
+
- **term_white_wins**: Whether white won the game. `null` indicates a tie. (No games are incomplete, but board states are from an intermediate position in the game)
|
36 |
+
|
37 |
+
# Code
|
38 |
+
|
39 |
+
```python3
|
40 |
+
from tqdm import tqdm
|
41 |
+
import collections
|
42 |
+
import chess
|
43 |
+
import random
|
44 |
+
from datasets import Dataset
|
45 |
+
|
46 |
+
import os
|
47 |
+
from multiprocessing import Process, Queue
|
48 |
+
|
49 |
+
|
50 |
+
def random_fen():
|
51 |
+
"""
|
52 |
+
Play random game to completion
|
53 |
+
backtrack random uniform number of moves
|
54 |
+
"""
|
55 |
+
board = chess.Board()
|
56 |
+
num_moves = 0
|
57 |
+
while True:
|
58 |
+
board.push(random.choice(list(board.legal_moves)))
|
59 |
+
num_moves += 1
|
60 |
+
if board.is_game_over():
|
61 |
+
outcome = board.outcome()
|
62 |
+
revert_moves = random.randint(0, num_moves)
|
63 |
+
for _ in range(revert_moves):
|
64 |
+
board.pop()
|
65 |
+
fen_num_moves = num_moves - revert_moves
|
66 |
+
termination_reason = outcome.termination.value
|
67 |
+
white_is_winner = outcome.winner
|
68 |
+
return board.fen(), fen_num_moves, num_moves, termination_reason, white_is_winner
|
69 |
+
|
70 |
+
|
71 |
+
def batched_random_fen(queue, batch_size=100):
|
72 |
+
while True:
|
73 |
+
fens = []
|
74 |
+
while len(fens) < batch_size:
|
75 |
+
fens.append(random_fen())
|
76 |
+
queue.put(fens)
|
77 |
+
|
78 |
+
|
79 |
+
def generate_fens_parallel(N):
|
80 |
+
all_fens = collections.defaultdict(
|
81 |
+
lambda: {"n_moves": [], "term_n_moves": [], "term_reason": [], "term_white_wins": []}
|
82 |
+
)
|
83 |
+
queue = Queue()
|
84 |
+
num_workers = os.cpu_count() // 2
|
85 |
+
|
86 |
+
# Start worker processes in advance
|
87 |
+
workers = []
|
88 |
+
for _ in range(num_workers):
|
89 |
+
worker = Process(target=batched_random_fen, args=(queue,))
|
90 |
+
worker.daemon = True # Ensures the process terminates with the main process
|
91 |
+
workers.append(worker)
|
92 |
+
worker.start()
|
93 |
+
|
94 |
+
with tqdm(total=N, desc="FENs Captured") as pbar:
|
95 |
+
for i in range(1_000_000_000_000):
|
96 |
+
for fen, board_num_moves, final_state_num_moves, final_result, final_is_white_winner in queue.get():
|
97 |
+
all_fens[fen]["n_moves"].append(board_num_moves)
|
98 |
+
all_fens[fen]["term_n_moves"].append(final_state_num_moves)
|
99 |
+
all_fens[fen]["term_reason"].append(final_result)
|
100 |
+
all_fens[fen]["term_white_wins"].append(final_is_white_winner)
|
101 |
+
|
102 |
+
if i % 100 == 0:
|
103 |
+
pbar.n = len(all_fens)
|
104 |
+
pbar.last_print_n = len(all_fens)
|
105 |
+
pbar.refresh()
|
106 |
+
|
107 |
+
if len(all_fens) >= N:
|
108 |
+
break
|
109 |
+
|
110 |
+
# Terminate workers
|
111 |
+
for worker in workers:
|
112 |
+
worker.terminate()
|
113 |
+
|
114 |
+
# flatten
|
115 |
+
return [{"fen": k, **v} for k, v in all_fens.items()]
|
116 |
+
|
117 |
+
|
118 |
+
if __name__ == "__main__":
|
119 |
+
board_dicts = list(generate_fens_parallel(100_000_000))
|
120 |
+
ds = Dataset.from_dict({key: [dic[key] for dic in board_dicts] for key in board_dicts[0]})
|
121 |
+
ds.save_to_disk("random_uniform_chess_boards.hf")
|
122 |
+
# ds = datasets.load_from_disk("random_uniform_chess_boards.hf")
|
123 |
+
# ds.push_to_hub(...)
|
124 |
+
```
|