Repository navigation
Expand file tree
/
Copy pathmcts_player.cc
More file actions
373 lines (316 loc) · 11.4 KB
/
Copy pathmcts_player.cc
File metadata and controls
373 lines (316 loc) · 11.4 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
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
// 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.
#include "cc/mcts_player.h"
#include <algorithm>
#include <cstdlib>
#include <sstream>
#include <utility>
#include "absl/memory/memory.h"
#include "absl/strings/str_cat.h"
#include "absl/strings/str_format.h"
#include "absl/strings/str_join.h"
#include "absl/time/clock.h"
#include "cc/logging.h"
#include "cc/random.h"
namespace minigo {
std::ostream& operator<<(std::ostream& os, const MctsPlayer::Options& options) {
os << options.tree << " inject_noise:" << options.inject_noise
<< " virtual_losses:" << options.virtual_losses
<< " num_readouts:" << options.num_readouts
<< " seconds_per_move:" << options.seconds_per_move
<< " time_limit:" << options.time_limit
<< " decay_factor:" << options.decay_factor
<< " fastplay_frequency:" << options.fastplay_frequency
<< " fastplay_readouts:" << options.fastplay_readouts
<< " target_pruning:" << options.target_pruning
<< " random_seed:" << options.random_seed << std::flush;
return os;
}
float TimeRecommendation(int move_num, float seconds_per_move, float time_limit,
float decay_factor) {
// Divide by two since you only play half the moves in a game.
int player_move_num = move_num / 2;
// Sum of geometric series maxes out at endgame_time seconds.
float endgame_time = seconds_per_move / (1.0f - decay_factor);
float base_time;
int core_moves;
if (endgame_time > time_limit) {
// There is so little main time that we're already in 'endgame' mode.
base_time = time_limit * (1.0f - decay_factor);
core_moves = 0;
} else {
// Leave at least endgame_time seconds for the end, and play at
// seconds_per_move for as long as possible.
base_time = seconds_per_move;
core_moves = (time_limit - endgame_time) / seconds_per_move;
}
return base_time *
std::pow(decay_factor, std::max(player_move_num - core_moves, 0));
}
MctsPlayer::MctsPlayer(std::unique_ptr<Model> model,
std::shared_ptr<InferenceCache> inference_cache,
Game* game, const Options& options)
: model_(std::move(model)),
game_(game),
rnd_(options.random_seed, Random::kUniqueStream),
options_(options),
inference_cache_(std::move(inference_cache)),
inference_mix_(rnd_.UniformUint64()) {
NewGame();
}
MctsPlayer::~MctsPlayer() = default;
void MctsPlayer::InitializeGame(const Position& position) {
tree_ = absl::make_unique<MctsTree>(position, options_.tree);
game_->NewGame();
}
void MctsPlayer::NewGame() { InitializeGame(Position(Color::kBlack)); }
bool MctsPlayer::UndoMove() {
if (!tree_->UndoMove()) {
return false;
}
game_->UndoMove();
return true;
}
Coord MctsPlayer::SuggestMove(int new_readouts, bool inject_noise) {
auto start = absl::Now();
if (inject_noise) {
InjectNoise(kDirichletAlpha);
}
const auto* root = tree_->root();
int target_readouts = root->N() + new_readouts;
if (options_.seconds_per_move > 0) {
// Use time to limit the number of reads.
float seconds_per_move = options_.seconds_per_move;
if (options_.time_limit > 0) {
seconds_per_move =
TimeRecommendation(root->position.n(), seconds_per_move,
options_.time_limit, options_.decay_factor);
}
while (absl::Now() - start < absl::Seconds(seconds_per_move)) {
TreeSearch(options_.virtual_losses, target_readouts);
}
} else {
// Use a fixed number of reads.
while (root->N() < target_readouts) {
TreeSearch(options_.virtual_losses, target_readouts);
}
}
if (ShouldResign()) {
return Coord::kResign;
}
return tree_->PickMove(&rnd_, true);
}
void MctsPlayer::TreeSearch(int num_leaves, int max_num_reads) {
MaybeExpandRoot();
SelectLeaves(num_leaves, max_num_reads);
ProcessLeaves();
}
void MctsPlayer::InjectNoise(float dirichlet_alpha) {
MaybeExpandRoot();
tree_->InjectNoise(rnd_.Dirichlet<kNumMoves>(kDirichletAlpha),
options_.noise_mix);
}
void MctsPlayer::MaybeExpandRoot() {
if (!tree_->root()->is_expanded) {
SelectLeaves(1, tree_->root()->N() + 1);
ProcessLeaves();
}
}
void MctsPlayer::SelectLeaves(int num_leaves, int max_num_reads) {
tree_search_inferences_.clear();
ModelOutput cached_output;
int max_cache_misses = num_leaves * 2;
int num_selected = 0;
int num_cache_misses = 0;
while (num_cache_misses < max_cache_misses &&
tree_->root()->N() < max_num_reads) {
auto* leaf = tree_->SelectLeaf(true);
if (leaf->game_over()) {
float value =
leaf->position.CalculateScore(game_->options().komi) > 0 ? 1 : -1;
tree_->IncorporateEndGameResult(leaf, value);
++num_cache_misses;
continue;
}
// Calculate the symmetry we want to use for inference.
auto inference_sym = GetInferenceSymmetry(leaf);
auto canonical_sym = GetCanonicalSymmetry(leaf);
InferenceCache::Key cache_key;
if (inference_cache_ != nullptr) {
cache_key =
InferenceCache::Key(leaf->move, canonical_sym, leaf->position);
if (inference_cache_->TryGet(cache_key, canonical_sym, inference_sym,
&cached_output)) {
tree_->IncorporateResults(leaf, cached_output.policy,
cached_output.value);
continue;
}
}
++num_cache_misses;
tree_search_inferences_.emplace_back(cache_key, canonical_sym,
inference_sym, leaf);
auto& input = tree_search_inferences_.back().input;
input.sym = inference_sym;
// TODO(tommadams): add a method to Model that returns the required position
// history size.
auto* node = leaf;
for (int i = 0; i < input.position_history.capacity(); ++i) {
input.position_history.push_back(&node->position);
node = node->parent;
if (node == nullptr) {
break;
}
}
tree_->AddVirtualLoss(leaf);
if (++num_selected == num_leaves) {
// We found enough leaves.
break;
}
if (leaf == tree_->root()) {
// If the root is a leaf, we can't possibly find any other leaves.
break;
}
}
}
void MctsPlayer::ProcessLeaves() {
if (tree_search_inferences_.empty()) {
return;
}
input_ptrs_.clear();
output_ptrs_.clear();
for (auto& x : tree_search_inferences_) {
input_ptrs_.push_back(&x.input);
output_ptrs_.push_back(&x.output);
}
// Run inference.
model_->RunMany(input_ptrs_, &output_ptrs_, &inference_model_);
// Record some information about the inference.
if (!inference_model_.empty()) {
if (inferences_.empty() || inference_model_ != inferences_.back().model) {
inferences_.emplace_back(inference_model_, tree_->root()->position.n());
}
inferences_.back().last_move = tree_->root()->position.n();
inferences_.back().total_count += tree_search_inferences_.size();
}
// Incorporate the inference outputs back into tree search.
for (auto& inference : tree_search_inferences_) {
auto& output = inference.output;
// Merge the inference output with those in the inference cache, possibly
// updating the values in `output`.
if (inference_cache_ != nullptr) {
inference_cache_->Merge(inference.cache_key, inference.canonical_sym,
inference.inference_sym, &output);
}
// Propagate the results back up the tree to the root.
tree_->IncorporateResults(inference.leaf, output.policy, output.value);
tree_->RevertVirtualLoss(inference.leaf);
}
if (tree_search_cb_ != nullptr) {
std::vector<const MctsNode*> leaves;
leaves.reserve(tree_search_inferences_.size());
for (auto& inference : tree_search_inferences_) {
leaves.push_back(inference.leaf);
}
tree_search_cb_(leaves);
}
}
bool MctsPlayer::ShouldResign() const {
return game_->options().resign_enabled &&
tree_->root()->Q_perspective() < game_->options().resign_threshold;
}
void MctsPlayer::SetTreeSearchCallback(TreeSearchCallback cb) {
tree_search_cb_ = std::move(cb);
}
std::string MctsPlayer::GetModelsUsedForInference() const {
std::vector<std::string> parts;
parts.reserve(inferences_.size());
for (const auto& info : inferences_) {
parts.push_back(absl::StrCat(info.model, "(", info.first_move, ",",
info.last_move, ")"));
}
return absl::StrJoin(parts, ", ");
}
bool MctsPlayer::PlayMove(Coord c, bool is_trainable) {
if (tree_->is_game_over()) {
MG_LOG(ERROR) << "Can't play move " << c << ", game is over";
return false;
}
// Handle resignations.
if (c == Coord::kResign) {
game_->SetGameOverBecauseOfResign(OtherColor(tree_->to_play()));
return true;
}
if (!tree_->is_legal_move(c)) {
MG_LOG(ERROR) << "Move " << c << " is illegal";
// We're probably about to crash. Dump the player's options and the moves
// that got us to this point.
MG_LOG(ERROR) << "MctsPlayer options: " << options_;
MG_LOG(ERROR) << "Game options: " << game_->options();
for (int i = 0; i < game_->num_moves(); ++i) {
const auto* move = game_->GetMove(i);
MG_LOG(ERROR) << move->color << " " << move->c;
}
return false;
}
// Adjust the visits before adding the move's search_pi to the Game.
if (is_trainable && options_.target_pruning) {
tree_->ReshapeFinalVisits(true);
}
UpdateGame(c, is_trainable);
tree_->PlayMove(c);
// Handle consecutive passing or termination by move limit.
if (tree_->is_game_over()) {
game_->SetGameOverBecauseOfPasses(
tree_->CalculateScore(game_->options().komi));
}
return true;
}
void MctsPlayer::PlayOpponentsMove(Coord c) {
if (!tree_->is_game_over()) {
MG_CHECK(c != Coord::kResign);
MG_CHECK(tree_->is_legal_move(c));
tree_->PlayMove(c);
}
}
void MctsPlayer::UpdateGame(Coord c, bool is_trainable) {
const auto* root = tree_->root();
// Record which model(s) were used when running tree search for this move.
std::vector<std::string> models;
if (!inferences_.empty()) {
for (auto it = inferences_.rbegin(); it != inferences_.rend(); ++it) {
if (it->last_move < root->position.n()) {
break;
}
models.push_back(it->model);
}
std::reverse(models.begin(), models.end());
}
// Build a comment for the move.
auto comment = tree_->Describe();
if (!models.empty()) {
comment =
absl::StrCat("models:", absl::StrJoin(models, ","), "\n", comment);
}
// Update the game history.
if (is_trainable && c != Coord::kResign) {
auto search_pi = tree_->CalculateSearchPi();
game_->AddTrainableMove(tree_->to_play(), c, root->position,
std::move(comment), root->Q(), root->N(),
search_pi);
} else {
game_->AddNonTrainableMove(tree_->to_play(), c, root->position,
std::move(comment), root->Q(), root->N());
}
}
} // namespace minigo