Repository navigation
Expand file tree
/
Copy pathgtp1.py
More file actions
215 lines (171 loc) · 7.58 KB
/
Copy pathgtp1.py
File metadata and controls
215 lines (171 loc) · 7.58 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
# 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.
"""Python implementation of selfplay worker.
This worker is used to set up many parallel selfplay instances."""
import random
import os
import sys
import socket
import time
from absl import app, flags
from tensorflow import gfile
import coords
import dual_net
import preprocessing
from gtp_cmd_handlers import (
BasicCmdHandler, KgsCmdHandler, GoGuiCmdHandler, MiniguiBasicCmdHandler, RegressionsCmdHandler)
import gtp_engine
from strategies import
8000
MCTSPlayer, CGOSPlayer
from utils import dbg
import utils
flags.DEFINE_bool('cgos_mode', False, 'Whether to use CGOS settings.')
flags.DEFINE_bool('kgs_mode', False, 'Whether to use KGS courtesy-pass.')
flags.DEFINE_bool('minigui_mode', False, 'Whether to add minigui logging.')
flags.DEFINE_string('load_file', None, 'Path to model save files.')
flags.DEFINE_string('selfplay_dir', None, 'Where to write game data.')
flags.DEFINE_string('holdout_dir', None, 'Where to write held-out game data.')
flags.DEFINE_string('sgf_dir', None, 'Where to write human-readable SGFs.')
flags.DEFINE_float('holdout_pct', 0.05, 'What percent of games to hold out.')
flags.DEFINE_float('resign_disable_pct', 0.05,
'What percent of games to disable resign for.')
# From strategies.py
flags.declare_key_flag('verbose')
flags.declare_key_flag('num_readouts')
FLAGS = flags.FLAGS
def make_gtp_instance(load_file, cgos_mode=False, kgs_mode=False,
minigui_mode=False):
"""Takes a path to model files and set up a GTP engine instance."""
# Here so we dont try load EdgeTPU python library unless we need to
#if load_file.endswith(".tflite"):
#from dual_net_edge_tpu import DualNetworkEdgeTpu
#n = DualNetworkEdgeTpu(load_file)
#else:
#from dual_net import DualNetwork
#n = DualNetwork(load_file)
n = dual_net.DualNetwork(load_file)
if cgos_mode:
player = CGOSPlayer(network=n, seconds_per_move=5, timed_match=True,
two_player_mode=True)
else:
player = MCTSPlayer(network=n, two_player_mode=True)
name = "Minigo-" + os.path.basename(load_file)
version = "0.2"
engine = gtp_engine.Engine()
engine.add_cmd_handler(
gtp_engine.EngineCmdHandler(engine, name, version))
if kgs_mode:
engine.add_cmd_handler(KgsCmdHandler(player))
engine.add_cmd_handler(RegressionsCmdHandler(player))
engine.add_cmd_handler(GoGuiCmdHandler(player))
if minigui_mode:
engine.add_cmd_handler(MiniguiBasicCmdHandler(player, courtesy_pass=kgs_mode))
else:
engine.add_cmd_handler(BasicCmdHandler(player, courtesy_pass=kgs_mode))
return engine
def play(network):
"""Plays out a self-play match, returning a MCTSPlayer object containing:
- the final position
- the n x 362 tensor of floats representing the mcts search probabilities
- the n-ary tensor of floats representing the original value-net estimate
where n is the number of moves in the game
"""
readouts = FLAGS.num_readouts # defined in strategies.py
# Disable resign in 5% of games
if random.random() < FLAGS.resign_disable_pct:
resign_threshold = -1.0
else:
resign_threshold = None
player = MCTSPlayer(network, resign_threshold=resign_threshold)
player.initialize_game()
# Must run this once at the start to expand the root node.
first_node = player.root.select_leaf()
prob, val = network.run(first_node.position)
first_node.incorporate_results(prob, val, first_node)
while True:
start = time.time()
player.root.inject_noise()
current_readouts = player.root.N
# we want to do "X additional readouts", rather than "up to X readouts".
while player.root.N < current_readouts + readouts:
player.tree_search()
if FLAGS.verbose >= 3:
print(player.root.position)
print(player.root.describe())
if player.should_resign():
player.set_result(-1 * player.root.position.to_play,
was_resign=True)
break
move = player.pick_move()
player.play_move(move)
if player.root.is_done():
player.set_result(player.root.position.result(), was_resign=False)
break
if (FLAGS.verbose >= 2) or (FLAGS.verbose >= 1 and player.root.position.n % 10 == 9):
print("Q: {:.5f}".format(player.root.Q))
dur = time.time() - start
print("%d: %d readouts, %.3f s/100. (%.2f sec)" % (
player.root.position.n, readouts, dur / readouts * 100.0, dur), flush=True)
if FLAGS.verbose >= 3:
print("Played >>",
coords.to_gtp(coords.from_flat(player.root.fmove)))
if FLAGS.verbose >= 2:
utils.dbg("%s: %.3f" % (player.result_string, player.root.Q))
utils.dbg(player.root.position, player.root.position.score())
return player
def run_game(load_file, selfplay_dir=None, holdout_dir=None,
sgf_dir=None, holdout_pct=0.05):
"""Takes a played game and record results and game data."""
if sgf_dir is not None:
minimal_sgf_dir = os.path.join(sgf_dir, 'clean')
full_sgf_dir = os.path.join(sgf_dir, 'full')
utils.ensure_dir_exists(minimal_sgf_dir)
utils.ensure_dir_exists(full_sgf_dir)
if selfplay_dir is not None:
utils.ensure_dir_exists(selfplay_dir)
utils.ensure_dir_exists(holdout_dir)
with utils.logged_timer("Loading weights from %s ... " % load_file):
network = dual_net.DualNetwork(load_file)
with utils.logged_timer("Playing game"):
player = play(network)
output_name = '{}-{}'.format(int(time.time()), socket.gethostname())
game_data = player.extract_data()
if sgf_dir is not None:
with gfile.GFile(os.path.join(minimal_sgf_dir, '{}.sgf'.format(output_name)), 'w') as f:
f.write(player.to_sgf(use_comments=False))
with gfile.GFile(os.path.join(full_sgf_dir, '{}.sgf'.format(output_name)), 'w') as f:
f.write(player.to_sgf())
tf_examples = preprocessing.make_dataset_from_selfplay(game_data)
if selfplay_dir is not None:
# Hold out 5% of games for validation.
if random.random() < holdout_pct:
fname = os.path.join(holdout_dir,
"{}.tfrecord.zz".format(output_name))
else:
fname = os.path.join(selfplay_dir,
"{}.tfrecord.zz".format(output_name))
preprocessing.write_tf_examples(fname, tf_examples)
def main(argv):
"""Entry point for running one selfplay game."""
del argv # Unused
flags.mark_flag_as_required('load_file')
engine = make_gtp_instance(FLAGS.load_file,
cgos_mode=FLAGS.cgos_mode,
kgs_mode=FLAGS.kgs_mode,
minigui_mode=FLAGS.minigui_mode)
dbg("GTP engine ready\n")
for msg in sys.stdin:
if not engine.handle_msg(msg.strip()):
break
if __name__ == '__main__':
app.run(main)