Make ValueHead a separate search algorithm.

Author: mooskagh

src/chess/callbacks.h

@@ -68,34 +68,34 @@ struct ThinkingInfo {
   // Hash fullness * 1000
   int hashfull = -1;
   // Moves to mate.
-  std::optional<int> mate;
+  std::optional<int> mate = std::nullopt;
   // Win in centipawns.
-  std::optional<int> score;
+  std::optional<int> score = std::nullopt;
   // Win/Draw/Lose probability * 1000.
   struct WDL {
     int w;
     int d;
     int l;
   };
-  std::optional<WDL> wdl;
+  std::optional<WDL> wdl = std::nullopt;
   // Number of successful TB probes (not the same as playouts ending in TB hit).
   int tb_hits = -1;
   // Best line found. Moves are from perspective of white player.
-  std::vector<Move> pv;
+  std::vector<Move> pv = {};
   // Multipv index.
   int multipv = -1;
   // Freeform comment.
-  std::string comment;
+  std::string comment = "";
 
   // Those are extensions and not really UCI protocol.
   // 1 if it's "player1", 2 if it's "player2"
   int player = -1;
   // Index of the game in the tournament (0-based).
   int game_id = -1;
   // The color of the player, if known.
-  std::optional<bool> is_black;
+  std::optional<bool> is_black = std::nullopt;
   // Moves left
-  std::optional<int> moves_left;
+  std::optional<int> moves_left = std::nullopt;
 };
 
 // Is sent when a single game is finished.

src/chess/position.h

@@ -98,6 +98,8 @@ class PositionHistory {
   PositionHistory() = default;
   PositionHistory(const PositionHistory& other) = default;
   PositionHistory(PositionHistory&& other) = default;
+  PositionHistory(std::span<const Position> positions)
+      : positions_(positions.begin(), positions.end()) {}
 
   PositionHistory& operator=(const PositionHistory& other) = default;
   PositionHistory& operator=(PositionHistory&& other) = default;

src/engine_classic.cc

@@ -61,11 +61,6 @@ const OptionId kStrictUciTiming{"strict-uci-timing", "StrictTiming",
                                 "The UCI host compensates for lag, waits for "
                                 "the 'readyok' reply before sending 'go' and "
                                 "only then starts timing."};
-const OptionId kValueOnly{
-    "value-only", "ValueOnly",
-    "In value only mode all search parameters are ignored and the position is "
-    "evaluated by getting the valuation of every child position and choosing "
-    "the worst for the opponent."};
 const OptionId kClearTree{"", "ClearTree",
                           "Clear the tree before the next search."};
 
@@ -123,7 +118,6 @@ void EngineClassic::PopulateOptions(OptionsParser* options) {
   options->Add<BoolOption>(kStrictUciTiming) = false;
   options->HideOption(kStrictUciTiming);
 
-  options->Add<BoolOption>(kValueOnly) = false;
   options->Add<ButtonOption>(kClearTree);
   options->HideOption(kClearTree);
 }
@@ -265,74 +259,6 @@ class PonderResponseTransformer : public TransformingUciResponder {
   std::string ponder_move_;
 };
 
-void ValueOnlyGo(classic::NodeTree* tree, Network* network,
-                 const OptionsDict& options,
-                 std::unique_ptr<UciResponder> responder) {
-  auto input_format = network->GetCapabilities().input_format;
-
-  const auto& board = tree->GetPositionHistory().Last().GetBoard();
-  auto legal_moves = board.GenerateLegalMoves();
-  tree->GetCurrentHead()->CreateEdges(legal_moves);
-  PositionHistory history = tree->GetPositionHistory();
-  std::vector<InputPlanes> planes;
-  for (auto edge : tree->GetCurrentHead()->Edges()) {
-    history.Append(edge.GetMove());
-    if (history.ComputeGameResult() == GameResult::UNDECIDED) {
-      planes.emplace_back(EncodePositionForNN(
-          input_format, history, 8, FillEmptyHistory::FEN_ONLY, nullptr));
-    }
-    history.Pop();
-  }
-
-  std::vector<float> comp_q;
-  int batch_size = options.Get<int>(classic::SearchParams::kMiniBatchSizeId);
-  if (batch_size == 0) batch_size = network->GetMiniBatchSize();
-
-  for (size_t i = 0; i < planes.size(); i += batch_size) {
-    auto comp = network->NewComputation();
-    for (int j = 0; j < batch_size; j++) {
-      comp->AddInput(std::move(planes[i + j]));
-      if (i + j + 1 == planes.size()) break;
-    }
-    comp->ComputeBlocking();
-
-    for (int j = 0; j < batch_size; j++) comp_q.push_back(comp->GetQVal(j));
-  }
-
-  Move best;
-  int comp_idx = 0;
-  float max_q = std::numeric_limits<float>::lowest();
-  for (auto edge : tree->GetCurrentHead()->Edges()) {
-    history.Append(edge.GetMove());
-    auto result = history.ComputeGameResult();
-    float q = -1;
-    if (result == GameResult::UNDECIDED) {
-      // NN eval is for side to move perspective - so if its good, its bad for
-      // us.
-      q = -comp_q[comp_idx];
-      comp_idx++;
-    } else if (result == GameResult::DRAW) {
-      q = 0;
-    } else {
-      // A legal move to a non-drawn terminal without tablebases must be a
-      // win.
-      q = 1;
-    }
-    if (q >= max_q) {
-      max_q = q;
-      best = edge.GetMove(tree->GetPositionHistory().IsBlackToMove());
-    }
-    history.Pop();
-  }
-  std::vector<ThinkingInfo> infos;
-  ThinkingInfo thinking;
-  thinking.depth = 1;
-  infos.push_back(thinking);
-  responder->OutputThinkingInfo(&infos);
-  BestMoveInfo info(best);
-  responder->OutputBestMove(&info);
-}
-
 }  // namespace
 
 void EngineClassic::Go(const GoParams& params) {
@@ -374,10 +300,6 @@ void EngineClassic::Go(const GoParams& params) {
     // Strip movesleft information from the response.
     responder = std::make_unique<MovesLeftResponseFilter>(std::move(responder));
   }
-  if (options_.Get<bool>(kValueOnly)) {
-    ValueOnlyGo(tree_.get(), network_.get(), options_, std::move(responder));
-    return;
-  }
 
   if (options_.Get<Button>(kClearTree).TestAndReset()) {
     tree_->TrimTreeAtHead();

src/search/instamove/instamove.cc

@@ -28,6 +28,7 @@
 #include "search/instamove/instamove.h"
 
 #include <algorithm>
+#include <cmath>
 #include <vector>
 
 #include "chess/gamestate.h"
@@ -68,7 +69,83 @@ class PolicyHeadFactory : public SearchFactory {
   }
 };
 
+class ValueHeadSearch : public InstamoveSearch {
+ public:
+  ValueHeadSearch(const SearchContext& context, const GameState& game_state)
+      : InstamoveSearch(context), game_state_(game_state) {}
+
+  Move GetBestMove() final {
+    std::unique_ptr<BackendComputation> computation =
+        backend()->CreateComputation();
+
+    PositionHistory history(game_state_.GetPositions());
+    const ChessBoard& board = history.Last().GetBoard();
+    const std::vector<Move> legal_moves = board.GenerateLegalMoves();
+    std::vector<EvalResult> results(legal_moves.size());
+
+    for (size_t i = 0; i < legal_moves.size(); i++) {
+      Move move = legal_moves[i];
+      history.Append(move);
+      switch (history.ComputeGameResult()) {
+        case GameResult::UNDECIDED:
+          computation->AddInput(
+              EvalPosition{history.GetPositions(), {}},
+              EvalResultPtr{.q = &results[i].q, .d = &results[i].d});
+          break;
+        case GameResult::DRAW:
+          results[i].q = 0;
+          results[i].d = 1;
+          break;
+        default:
+          // A legal move to a non-drawn terminal without tablebases must be a
+          // win.
+          results[i].q = -1;
+          results[i].d = 0;
+      }
+      history.Pop();
+    }
+
+    computation->ComputeBlocking();
+
+    const size_t best_idx =
+        std::min_element(results.begin(), results.end(),
+                         [](const EvalResult& a, const EvalResult& b) {
+                           return a.q < b.q;
+                         }) -
+        results.begin();
+
+    std::vector<ThinkingInfo> infos = {{
+        .depth = 1,
+        .seldepth = 1,
+        .nodes = static_cast<int64_t>(legal_moves.size()),
+        .score = 90 * std::tan(1.5637541897 * results[best_idx].q),
+        .wdl =
+            ThinkingInfo::WDL{
+                static_cast<int>(std::round(
+                    500 * (1 + results[best_idx].q - results[best_idx].d))),
+                static_cast<int>(std::round(1000 * results[best_idx].d)),
+                static_cast<int>(std::round(
+                    500 * (1 - results[best_idx].q - results[best_idx].d)))},
+    }};
+    uci_responder()->OutputThinkingInfo(&infos);
+    return legal_moves[best_idx];
+  }
+
+ private:
+  const GameState game_state_;
+};
+
+class ValueHeadFactory : public SearchFactory {
+  std::string_view GetName() const override { return "valuehead"; }
+  std::unique_ptr<SearchEnvironment> CreateEnvironment(
+      UciResponder* responder, const OptionsDict* options) const override {
+    return std::make_unique<InstamoveEnvironment<ValueHeadSearch>>(responder,
+                                                                   options);
+  }
+};
+
 REGISTER_SEARCH(PolicyHeadFactory);
+REGISTER_SEARCH(ValueHeadFactory);
 
 }  // namespace instamove
 }  // namespace lczero