File size: 8,622 Bytes
09865df 868584e 09865df 4016c5c 868584e 09865df d8e6f68 868584e ba1e298 09865df 46f6011 e3868b1 09865df d8e6f68 c97d40f 868584e 09865df 0254047 09865df 868584e 09865df 99333cf 09865df 0254047 868584e 09865df 868584e 09865df 868584e 09865df d8e6f68 ad46e7d 09865df 0f2cd73 09865df 868584e 09865df 868584e 09865df 868584e 09865df 46f6011 868584e 09865df |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 |
import os
import sys
import chess.pgn
import gradio as gr
import torch
from loguru import logger
import numpy as np
from src.data.data_utils import clean_board
from src.engine.agents.policies import beam_search, eval_board, one_depth_eval
from src.engine.agents.viz_utils import plot_save_beam_search, save_svg, board_to_svg
from src.models.model_space import MultiInputConv
# TEMP_DIR = "./demos/temp/"
CHKPT = "checkpoint.pt"
file = sys.argv[0]
DIR_PATH = os.path.dirname(file)
TEMP_DIR = os.path.join(DIR_PATH, "temp")
chkpt = torch.load(os.path.join(DIR_PATH, f"checkpoints/{CHKPT}"))
model = MultiInputConv()
model.load_state_dict(state_dict=chkpt["model_state_dict"])
model.eval()
os.makedirs(name=TEMP_DIR, exist_ok=True)
@logger.catch(level="DEBUG", reraise=True)
def evaluate_board(board: chess.Board):
"""Evaluate the board.
Args:
board (chess.Board): chess.Board object
Returns:
float: score of the board
"""
board = clean_board(board=board)
save_svg(board=board, filename=os.path.join(TEMP_DIR, "board"), to_png=False)
return os.path.join(TEMP_DIR, "board.svg"), eval_board(model=model, board=board)
@logger.catch(level="DEBUG", reraise=True)
def plot_beam_search(board: chess.Board,
depth: int,
beam_width: int,
player_strategy: str,
opponent_strategy: str,
player_top_k: int,
opponent_top_k: int):
"""Plot the beam search tree.
Args:
board (chess.Board): chess.Board object
depth (int): depth of the search
beam_width (int): width of the beam
player_strategy (str): sampling strategy
opponent_strategy (str): sampling strategy
player_top_k (int): top-k value
opponent_top_k (int): top-k value
Returns:
Image: image of the beam search tree
"""
board = clean_board(board=board)
beam = beam_search(model=model,
board=board,
depth=depth,
beam_width=beam_width,
player_strategy=player_strategy,
opponent_strategy=opponent_strategy,
player_top_k=player_top_k,
opponent_top_k=opponent_top_k,
min_score=-100,
max_score=100)
plot_save_beam_search(
beam=beam,
filename=os.path.join(TEMP_DIR, "beam_search"),
temp_dir=TEMP_DIR,
intermediate_png=True,
)
return os.path.join(TEMP_DIR, "beam_search.png")
@logger.catch(level="DEBUG", reraise=True)
def get_one_depth_eval(board: chess.Board):
"""Get the legal boards from one-depth evaluation.
Args:
board (chess.Board): chess.Board object
Returns:
list: list of tuples of SVG images and scores of the legal boards
"""
board = clean_board(board=board)
legal_boards, legal_moves, scores = one_depth_eval(
model=model, boards=[board], min_score=-100, max_score=100
)
# get scores argsort
argsort = np.argsort(scores[0])
if board.turn:
argsort = argsort[::-1]
scores = np.array(scores[0])[argsort]
legal_boards = np.array(legal_boards[0])[argsort]
legal_moves = np.array(legal_moves[0])[argsort]
[save_svg(board=board, filename=os.path.join(TEMP_DIR, f"board_{i}"), to_png=False) for i, board in
enumerate(legal_boards)]
return (gr.update(value=[(os.path.join(TEMP_DIR, f"board_{i}.svg"), str(scores[i])) for i in range(len(legal_boards))]),
gr.update(choices=[str(move) for move in legal_moves]),
[str(move) for move in legal_moves])
@logger.catch(level="DEBUG", reraise=True)
def select_dropdown_item(moves, evt: gr.SelectData):
"""Select the nth item in the dropdown.
Args:
moves (list): list of moves
evt (gr.EventData): event data
"""
selected_index = evt.index
return gr.update(value=moves[selected_index])
@logger.catch(level="DEBUG", reraise=True)
def update_run_fen(fen, dropdown):
"""Update the FEN board with the selected move.
Args:
fen (str): FEN board
dropdown (str): selected move
"""
board = clean_board(board=fen)
board.push_san(dropdown)
boards, update_dropdown, moves = get_one_depth_eval(board.fen())
return board.fen(), boards, update_dropdown, moves
def update_top_k_visibility(strategy):
return gr.update(visible=(strategy == "top-k"))
with gr.Blocks() as demo:
gr.Markdown("Explore the model")
with gr.Tab("Beam search"):
with gr.Row():
with gr.Column():
board = gr.Textbox(
value="rnbqkbnr/ppp1pppp/8/3p4/2PP4/8/PP2PPPP/RNBQKBNR b KQkq - 0 2",
label="Provide FEN or PGN board:",
)
depth_slider = gr.Slider(value=4, minimum=1, maximum=10, step=1, label="Choose beam depth")
width_slider = gr.Slider(value=4, minimum=1, maximum=10, step=1, label="Choose beam width")
player_strategy = gr.Dropdown(
label="Select the player sampling strategy :",
choices=["greedy", "top-k"],
value="greedy",
interactive=True,
allow_custom_value=False,
)
opponent_strategy = gr.Dropdown(
label="Select the opponent sampling strategy :",
choices=["greedy", "top-k"],
value="greedy",
interactive=True,
allow_custom_value=False,
)
player_top_k = gr.Slider(value=5,
minimum=5,
maximum=20,
step=1,
label="Choose player top-k",
interactive=True,
visible=False)
opponent_top_k = gr.Slider(value=2,
minimum=2,
maximum=20,
step=1,
label="Choose opponent top-k",
interactive=True,
visible=False)
btn = gr.Button("Run beam search")
with gr.Column():
beam = gr.Image()
player_strategy.change(fn=lambda x: gr.update(visible=(x == "top-k")), inputs=[player_strategy], outputs=[player_top_k])
opponent_strategy.change(fn=lambda x: gr.update(visible=(x == "top-k")), inputs=[opponent_strategy], outputs=[opponent_top_k])
width_slider.change(fn=lambda x: gr.update(minimum=x + 1, value=x + 1), inputs=[width_slider], outputs=[player_top_k])
btn.click(fn=plot_beam_search,
inputs=[board, depth_slider, width_slider, player_strategy, opponent_strategy, player_top_k, opponent_top_k],
outputs=beam)
with gr.Tab("One-depth eval"):
moves = gr.State(value=[])
board = gr.Textbox(
value="rnbqkbnr/ppp1pppp/8/3p4/2PP4/8/PP2PPPP/RNBQKBNR b KQkq - 0 2",
label="Provide FEN or PGN board:",
)
btn = gr.Button("Get one-depth evaluation")
gallery = gr.Gallery(
label="Legal boards from one-depth eval",
show_label=False,
elem_id="gallery",
columns=6,
interactive=False
)
dropdown = gr.Dropdown(
label="Select the next move :",
interactive=True,
)
btn_replace = gr.Button("Append selected move and run evaluation")
btn.click(fn=get_one_depth_eval, inputs=[board], outputs=[gallery, dropdown, moves])
gallery.select(fn=select_dropdown_item, inputs=[moves], outputs=dropdown)
btn_replace.click(fn=update_run_fen, inputs=[board, dropdown], outputs=[board, gallery, dropdown, moves])
with gr.Tab("Score a board"):
gr.Interface(
fn=evaluate_board,
inputs=[
gr.Textbox(
value="rnbqkbnr/ppp1pppp/8/3p4/2PP4/8/PP2PPPP/RNBQKBNR b KQkq - 0 2",
label="Provide FEN or PGN board:",
),
],
outputs=["image", "text"],
allow_flagging="never",
)
if __name__ == "__main__":
demo.launch()
|