Repository navigation
Expand file tree
/
Copy pathevaluate.py
More file actions
297 lines (237 loc) · 11.7 KB
/
Copy pathevaluate.py
File metadata and controls
297 lines (237 loc) · 11.7 KB
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
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
# Copyright 2018 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Evaluation plays games between two neural nets."""
import itertools
import os
import random
import time
from collections import defaultdict
import pandas as pd
from absl import app, flags, logging
from tensorflow.python import gfile
import coords
import k2net as dual_net
import go
import preprocessing
from model_config import ModelConfig
from run_selfplay import InitPositions
from strategies import MCTSPlayer
import sgf_wrapper
import utils
import myconf
# None = unlimited, 0 = auto-detect
pd.set_option('display.max_columns', None)
pd.set_option('display.width', 120)
flags.DEFINE_string('eval_data_dir', f'{myconf.EXP_HOME}/eval_bots/data', 'Where to write game data.')
flags.DEFINE_string('eval_sgf_dir', f'{myconf.EXP_HOME}/eval_bots/sgfs', 'Where to write evaluation results.')
# From strategies.py
flags.declare_key_flag('verbose')
FLAGS = flags.FLAGS
NUM_OPEN_MOVES = 6
class Ledger:
""" keep tab on win/loss records of two players where they take equal black/white role
No longer check for redundancy, as 9x9 games with soft-pick rarely repeats
"""
def __init__(self):
self._wins_by_black = defaultdict(int)
self._games_by_black = defaultdict(int)
self._model_ids = set()
def _parse_winner_side(self, result_str: str) -> int:
winner = result_str.upper()[0]
return 1 if winner == 'B' else -1
def record_game(self, black_model_id: str, white_model_id: str, result_str):
winner = self._parse_winner_side(result_str)
if winner == 1:
self._wins_by_black[black_model_id] += 1
self._games_by_black[black_model_id] += 1
self._model_ids.add(black_model_id)
self._model_ids.add(white_model_id)
# assert len(self._model_ids) == 2
def report(self):
""" report lists #wins by a model playing as black/white.
model0.lr=0.01_vw=0.5 model0.lr=0.02_vw=0.5
Black 6 8
White 2 4
Total 8 12
This says overall, the 2nd model wins. It wins 8 out 10 games as black, 4 out of 10 games as white.
"""
results_dict = {}
for model in self._model_ids:
opponent = (self._model_ids - {model}).pop()
wins_as_white = self._games_by_black[opponent] - self._wins_by_black[opponent]
d = {'Black': self._wins_by_black[model],
'White': wins_as_white,
'Total': (self._wins_by_black[model] + wins_as_white)}
results_dict[model] = pd.Series(d)
df = pd.DataFrame(results_dict).sort_index(axis=1)
# df.index.name = '#wins'
# print(df)
return df
def test_ledger(argv):
ledger = Ledger()
models = list('AB')
results = ['B+R', 'W+R']
for i in range(5):
result = random.choice(results)
print(f'A as black: {result}')
ledger.record_game('A', 'B', result)
if i < 3:
result = random.choice(results)
print(f'B as black: {result}')
ledger.record_game('B', 'A', result)
ledger.report()
class RunOneSided:
def __init__(self, black_config: ModelConfig, white_config: ModelConfig, sgf_dir: str):
# self.black_model = black_model
# self.white_model = white_model
self.sgf_dir = sgf_dir
utils.ensure_dir_exists(sgf_dir)
utils.ensure_dir_exists(FLAGS.eval_data_dir)
self.black_model_id = black_config.model_id()
self.white_model_id = white_config.model_id()
with utils.logged_timer("Loading weights"):
self.black_net = dual_net.load_net(black_config.model_path())
self.white_net = dual_net.load_net(white_config.model_path())
self.black_player = MCTSPlayer(self.black_net, num_readouts=black_config.num_readouts)
self.white_player = MCTSPlayer(self.white_net, num_readouts=white_config.num_readouts)
self.init_positions = InitPositions(None, None) #['C2'], [1.0])
self._num_games_so_far = 0
def _play(self, init_position: go.Position) -> MCTSPlayer:
""" return None if game is not played """
black, white = self.black_player, self.white_player
for player in [black, white]:
player.initialize_game(init_position)
first_node = player.root.select_leaf()
prob, val = player.network.run(first_node.position)
first_node.incorporate_results(prob, val, first_node)
for num_move in range(init_position.n, 500):
start = time.time()
if num_move % 2:
active, inactive = white, black
else:
active, inactive = black, white
current_readouts = active.root.N
while active.root.N < current_readouts + active.num_readouts:
active.tree_search()
# print some stats on the search
if FLAGS.verbose >= 3:
print(active.root.position)
# First, check the roots for hopeless games.
if active.should_resign(): # Force resign
active.set_result(-1 * active.root.position.to_play, was_resign=True)
inactive.set_result(-1 * active.root.position.to_play, was_resign=True)
break
if active.is_done():
break
move, best_move = active.pick_move(active.root.position.n < FLAGS.softpick_move_cutoff)
active.play_move(move, record_pi=True)
inactive.play_move(move, record_pi=False)
# todo check for benson score final
# if num_move < NUM_OPEN_MOVES:
# history = active.root.position.recent
# assert history[-1].move == move
dur = time.time() - start
if (FLAGS.verbose > 1): # or (FLAGS.verbose == 1 and num_move % 10 == 9):
timeper = (dur / active.num_readouts) * 100.0
print(active.root.position)
print("%d: %d readouts, %.3f s/100. (%.2f sec)" % (num_move,
active.num_readouts,
timeper,
dur))
if active.result == 0:
active.set_result(active.root.position.result(), was_resign=False)
inactive.set_result(active.root.position.result(), was_resign=False)
return active
def _create_sgf(self, ith_game: int, init_position_n: int):
""" merge comments from black and white """
fname = "{:s}-vs-{:s}-{:d}-{:d}.sgf".format(self.black_model_id, self.white_model_id, ith_game,
utils.microseconds_since_midnight())
game_history = self.black_player.position.recent
if len(game_history) < len(self.white_player.position.recent):
game_history = self.white_player.position.recent
black_comments = self.black_player.comments
white_comments = self.white_player.comments
assert len(black_comments) == len(white_comments) and len(black_comments) + init_position_n == len(game_history)
comments = ['initial' for i in range(init_position_n)]
comments.extend([black_comments[i] if (init_position_n + i) % 2 == 0 else white_comments[i] for i in range(len(black_comments))])
with gfile.GFile(os.path.join(self.sgf_dir, fname), 'w') as _file:
sgfstr = sgf_wrapper.make_sgf(game_history,
self.black_player.result_string, komi=self.black_player.position.komi,
comments=comments,
black_name=self.black_model_id, white_name=self.white_model_id)
_file.write(sgfstr)
def _make_tfrecords(self, ith_game: int):
output_name = f'{ith_game}-%d' % utils.microseconds_since_midnight()
game_data_black = self.black_player.extract_data()
game_data_white = self.white_player.extract_data()
tf_examples = preprocessing.make_dataset_from_selfplay(itertools.chain(game_data_black, game_data_white))
# no Hold out 5% of games for validation.
fname = os.path.join(FLAGS.eval_data_dir, f'{output_name}.tfrecord.zz')
preprocessing.write_tf_examples(fname, tf_examples)
def play_a_game(self):
game_idx = self._num_games_so_far
self._num_games_so_far += 1
init_position = self.init_positions.sample()
active = self._play(init_position)
assert active is not None
result_str = active.result_string
self._create_sgf(game_idx, init_position.n)
# self._make_tfrecords(game_idx)
game_history = active.position.recent
move_history_head = ' '.join([coords.to_gtp(game_history[i].move) for i in range(12)])
logging.info(f'Finished game %3d: %3d moves, %-7s %s', game_idx, len(game_history), result_str, move_history_head)
return result_str
def setup_common(argv):
_, black_model, white_model, num_games = argv
models_dir = f'{myconf.MODELS_DIR}'
if not black_model.startswith('/'):
black_model = f'{models_dir}/{black_model}'
if not white_model.startswith('/'):
white_model = f'{models_dir}/{white_model}'
utils.ensure_dir_exists(FLAGS.eval_sgf_dir)
return black_model, white_model, int(num_games)
def setup_oneside(argv):
_, black_config, white_config, num_games = argv
utils.ensure_dir_exists(FLAGS.eval_sgf_dir)
return ModelConfig(black_config), ModelConfig(white_config), int(num_games)
def main_twosided(argv):
"""Play matches between two neural nets, alternating sides"""
black_model, white_model, num_games = setup_common(argv)
ledger = Ledger()
runner1 = RunOneSided(black_model, white_model, FLAGS.eval_sgf_dir)
runner2 = RunOneSided(white_model, black_model, FLAGS.eval_sgf_dir)
logging.info('Tournament: %s vs %s, %d games', runner1.black_model_id, runner1.white_model_id, num_games)
for i in range(num_games // 2):
result_str = runner1.play_a_game()
ledger.record_game(runner1.black_model_id, runner1.white_model_id, result_str)
result_str = runner2.play_a_game()
ledger.record_game(runner2.black_model_id, runner2.white_model_id, result_str)
ledger.report()
def main_oneside(argv) -> int:
""" play a number of games between black and white """
black_config, white_config, num_games = setup_oneside(argv)
ledger = Ledger()
runner1 = RunOneSided(black_config, white_config, FLAGS.eval_sgf_dir)
logging.info('Tournament: %s vs %s, %d games', runner1.black_model_id, runner1.white_model_id, num_games)
for i in range(num_games):
result_str = runner1.play_a_game()
ledger.record_game(runner1.black_model_id, runner1.white_model_id, result_str)
df = ledger.report()
total_black_win = df.loc['Total', runner1.black_model_id]
total_white_win = df.loc['Total', runner1.white_model_id]
assert total_black_win + total_white_win == num_games
return total_black_win
if __name__ == '__main__':
# flags.mark_flag_as_required('eval_sgf_dir')
app.run(main_oneside)