Add files via upload

Author: Videodr0me

src/analyzer/analyzer.cc

@@ -0,0 +1,264 @@
+/*
+  This file is part of Leela Chess Zero.
+  Copyright (C) 2018 The LCZero Authors
+
+  Leela Chess is free software: you can redistribute it and/or modify
+  it under the terms of the GNU General Public License as published by
+  the Free Software Foundation, either version 3 of the License, or
+  (at your option) any later version.
+
+  Leela Chess is distributed in the hope that it will be useful,
+  but WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+  GNU General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with Leela Chess.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#include "analyzer/analyzer.h"
+#include "analyzer/table.h"
+#include "mcts/search.h"
+#include "neural/factory.h"
+#include "neural/loader.h"
+#include "utils/optionsparser.h"
+#include "utils/string.h"
+
+#include <algorithm>
+#include <cmath>
+#include <iostream>
+#include <sstream>
+#include <string>
+#include <vector>
+
+namespace lczero {
+namespace {
+const char* kTsvReportStr = "Filename of the tab-separated report file";
+const char* kTxtReportStr = "Filename of the text report file";
+const char* kMovesStr = "Moves in UCI format, space separated";
+const char* kMovesToAnalyzeStr = "Number of (last) moves to analyze";
+const char* kNodesStr = "(comma separated) How many nodes to calculate";
+const char* kTrainExamplesStr =
+    "How many examples of training data to generate";
+const char* kWeightsStr = "Network weights file path";
+const char* kNnBackendStr = "NN backend to use";
+const char* kNnBackendOptionsStr = "NN backend parameters";
+
+const char* kAutoDiscover = "<autodiscover>";
+}  // namespace
+
+Analyzer::Analyzer() {
+  options_parser_.AddContext("play");
+  options_parser_.AddContext("training");
+  play_options_ = &options_parser_.GetOptionsDict("play");
+  play_options_ = &options_parser_.GetOptionsDict("training");
+
+  options_parser_.Add<StringOption>(kTsvReportStr, "tsv-report");
+  options_parser_.Add<StringOption>(kTxtReportStr, "txt-report");
+  options_parser_.Add<StringOption>(kMovesStr, "moves");
+  options_parser_.Add<IntOption>(kMovesToAnalyzeStr, 1, 999, "num-moves") = 4;
+  options_parser_.Add<StringOption>(kNodesStr, "nodes-list") =
+      "10,50,100,200,400,600,800,1200,1600,5000,10000";
+  options_parser_.Add<IntOption>(kTrainExamplesStr, 1, 999,
+                                 "training-examples") = 10;
+  options_parser_.Add<StringOption>(kWeightsStr, "weights", 'w') =
+      kAutoDiscover;
+
+  // TODO(mooskagh) Move all that to network factory.
+  const auto backends = NetworkFactory::Get()->GetBackendsList();
+  options_parser_.Add<ChoiceOption>(kNnBackendStr, backends, "backend") =
+      backends.empty() ? "<none>" : backends[0];
+  options_parser_.Add<StringOption>(kNnBackendOptionsStr, "backend-opts");
+
+  Search::PopulateUciParams(&options_parser_);
+
+  // Overriding default options of search.
+  auto defaults = options_parser_.GetMutableDefaultsOptions();
+  defaults->Set<int>(Search::kMiniBatchSizeStr, 1);      // Minibatch = 1
+  defaults->Set<bool>(Search::kSmartPruningStr, false);  // No smart pruning
+  defaults->Set<bool>(Search::kVerboseStatsStr, true);   // Verbose stats
+}
+
+void Analyzer::GatherStats(Table3d* table, const Node* root_node,
+                           std::string& col, bool flip) {
+  uint64_t total_n = 0;
+  float factor = play_options_->Get<float>(Search::kCpuctStr) *
+                 std::sqrt(std::max(root_node->GetN(), 1u));
+  const float parent_q =
+      -root_node->GetQ(0) - play_options_->Get<float>(Search::kFpuReductionStr);
+
+  for (Node* node : root_node->Children()) {
+    const auto n = node->GetNStarted();
+    total_n += n;
+    const auto u = factor * node->GetU();
+    const auto q = node->GetQ(parent_q);
+    const auto move = node->GetMove(flip).as_string();
+    table->Add3dVal(col, move, "N", std::to_string(n));
+    table->Add3dVal(col, move, "U", std::to_string(u));
+    table->Add3dVal(col, move, "Q", std::to_string(q));
+    table->Add3dVal(col, move, "U+Q", std::to_string(u + q));
+  }
+
+  for (Node* node : root_node->Children()) {
+    auto n = node->GetNStarted();
+    table->Add3dVal(col, node->GetMove(flip).as_string(), "N%",
+                    std::to_string(static_cast<double>(n) / total_n));
+  }
+}
+
+void Analyzer::RunOnePosition(const std::vector<Move>& moves) {
+  NNCache cache(200000);
+  NodeTree tree;
+  tree.ResetToPosition(ChessBoard::kStartingFen, moves);
+
+  auto nodeses = ParseIntList(play_options_->Get<std::string>(kNodesStr));
+  std::sort(nodeses.begin(), nodeses.end());
+
+  Table3d table;
+  std::vector<std::string> cols;
+
+  // Run search in increasing number of nodes.
+  for (int nodes : nodeses) {
+    WriteToLog("Nodes: " + std::to_string(nodes));
+    SearchLimits limits;
+    limits.visits = nodes;
+
+    Search search(tree, network_.get(),
+                  std::bind(&Analyzer::OnBestMove, this, std::placeholders::_1),
+                  std::bind(&Analyzer::OnInfo, this, std::placeholders::_1),
+                  limits, *play_options_, &cache);
+
+    search.RunBlocking(1);
+
+    auto col = std::to_string(nodes);
+    cols.push_back(col);
+    GatherStats(&table, tree.GetCurrentHead(), col, tree.IsBlackToMove());
+    table.AddColVal(col, "bestmove", search.GetBestMove().first.as_string());
+  }
+
+  // Fetch MCTS-agnostic per-move stats P and V.
+  std::vector<const Node*> nodes;
+  for (Node* iter : tree.GetCurrentHead()->Children()) {
+    nodes.emplace_back(iter);
+  }
+  std::sort(nodes.begin(), nodes.end(), [](const Node* a, const Node* b) {
+    return a->GetNStarted() > b->GetNStarted();
+  });
+  std::vector<std::string> rows;
+  for (const Node* node : nodes) {
+    auto move = node->GetMove(tree.IsBlackToMove()).as_string();
+    rows.push_back(move);
+    table.AddRowVal(move, "P", std::to_string(node->GetP()));
+    table.AddRowVal(move, "V", std::to_string(node->GetV()));
+  }
+
+  // Dump table to log.
+  auto lines = table.RenderTable(cols, rows, {"N", "N%", "U", "Q", "U+Q"},
+                                 {"P", "V"}, {"bestmove"});
+  for (const auto& line : lines) WriteToTsvLog(line);
+}
+
+void Analyzer::Run() {
+  if (!options_parser_.ProcessAllFlags()) return;
+
+  // Open log files.
+  if (!play_options_->Get<std::string>(kTxtReportStr).empty()) {
+    log_.open(play_options_->Get<std::string>(kTxtReportStr).c_str());
+  }
+  if (!play_options_->Get<std::string>(kTsvReportStr).empty()) {
+    tsvlog_.open(play_options_->Get<std::string>(kTsvReportStr).c_str());
+  }
+
+  // Load network
+  InitializeNetwork();
+
+  // Parse moves.
+  std::vector<std::string> moves_str =
+      StrSplitAtWhitespace(play_options_->Get<std::string>(kMovesStr));
+  std::vector<Move> moves;
+  for (const auto& move : moves_str) moves.emplace_back(move);
+
+  for (int i = 0; i < play_options_->Get<int>(kMovesToAnalyzeStr); ++i) {
+    // Write position to logs.
+    std::string initial_pos;
+    if (moves_str.empty()) {
+      WriteToLog("Position: startpos");
+      WriteToTsvLog({"Startpos."});
+    } else {
+      WriteToLog("Position: moves " + StrJoin(moves_str));
+      WriteToTsvLog({"Moves " + StrJoin(moves_str)});
+    }
+
+    // Run Mcts at different depths.
+    RunOnePosition(moves);
+
+    // Run training several times.
+
+    if (moves.empty()) break;
+    moves.pop_back();
+    moves_str.pop_back();
+    WriteToTsvLog({});
+  }
+
+  // DumpFlags();
+}
+
+void Analyzer::InitializeNetwork() {
+  std::string network_path = play_options_->Get<std::string>(kWeightsStr);
+  std::string backend = play_options_->Get<std::string>(kNnBackendStr);
+  std::string backend_options =
+      play_options_->Get<std::string>(kNnBackendOptionsStr);
+
+  std::string net_path = network_path;
+  if (net_path == kAutoDiscover) {
+    net_path = DiscoveryWeightsFile();
+  }
+  Weights weights = LoadWeightsFromFile(net_path);
+
+  OptionsDict network_options = OptionsDict::FromString(
+      backend_options, &options_parser_.GetOptionsDict());
+
+  network_ = NetworkFactory::Get()->Create(backend, weights, network_options);
+}
+
+void Analyzer::WriteToLog(const std::string& line) const {
+  std::cout << line << std::endl;
+  if (log_) log_ << line << std::endl;
+}
+
+void Analyzer::WriteToTsvLog(const std::vector<std::string>& line) const {
+  if (!tsvlog_) return;
+  bool first = true;
+  for (const auto& chunk : line) {
+    if (first) {
+      first = false;
+    } else {
+      tsvlog_ << "\t";
+    }
+    tsvlog_ << chunk;
+  }
+  tsvlog_ << std::endl;
+}
+
+void Analyzer::OnBestMove(const BestMoveInfo& move) const {
+  WriteToLog("BestMove: " + move.bestmove.as_string());
+}
+
+void Analyzer::OnInfo(const ThinkingInfo& info) const {
+  std::string res = "Info";
+  if (info.depth >= 0) res += " depth " + std::to_string(info.depth);
+  if (info.seldepth >= 0) res += " seldepth " + std::to_string(info.seldepth);
+  if (info.time >= 0) res += " time " + std::to_string(info.time);
+  if (info.nodes >= 0) res += " nodes " + std::to_string(info.nodes);
+  if (info.score) res += " score cp " + std::to_string(*info.score);
+  if (info.nps >= 0) res += " nps " + std::to_string(info.nps);
+
+  if (!info.pv.empty()) {
+    res += " pv";
+    for (const auto& move : info.pv) res += " " + move.as_string();
+  }
+  if (!info.comment.empty()) res += " string " + info.comment;
+  WriteToLog(res);
+}
+
+}  // namespace lczero

src/analyzer/analyzer.h

@@ -0,0 +1,56 @@
+/*
+  This file is part of Leela Chess Zero.
+  Copyright (C) 2018 The LCZero Authors
+
+  Leela Chess is free software: you can redistribute it and/or modify
+  it under the terms of the GNU General Public License as published by
+  the Free Software Foundation, either version 3 of the License, or
+  (at your option) any later version.
+
+  Leela Chess is distributed in the hope that it will be useful,
+  but WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+  GNU General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with Leela Chess.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#pragma once
+
+#include <fstream>
+#include "analyzer/table.h"
+#include "chess/board.h"
+#include "chess/callbacks.h"
+#include "mcts/node.h"
+#include "neural/network.h"
+#include "utils/optionsparser.h"
+
+namespace lczero {
+
+class Analyzer {
+ public:
+  Analyzer();
+  void Run();
+
+ private:
+  void RunOnePosition(const std::vector<Move>& position);
+
+  void WriteToLog(const std::string& line) const;
+  void WriteToTsvLog(const std::vector<std::string>& line) const;
+
+  void InitializeNetwork();
+  void OnBestMove(const BestMoveInfo& move) const;
+  void OnInfo(const ThinkingInfo& info) const;
+  void GatherStats(Table3d* table, const Node* root_node, std::string& col,
+                   bool flip);
+
+  std::unique_ptr<Network> network_;
+  OptionsParser options_parser_;
+  const OptionsDict* play_options_;
+  // const OptionsDict* training_options_;
+  mutable std::ofstream log_;
+  mutable std::ofstream tsvlog_;
+};
+
+}  // namespace lczero