Merge pull request google-deepmind#1054 from maichmueller:feat/infostate_binding

PiperOrigin-RevId: 859599054
Change-Id: I75fcf8f11af2d89638a3c130a533398784583fcc

Author: lanctot

.gitignore

@@ -38,6 +38,7 @@ open_spiel/games/universal_poker/double_dummy_solver/
 open_spiel/games/hanabi/hanabi-learning-environment/
 /open_spiel/pybind11_abseil/
 pybind11/
+!open_spiel/python/pybind11
 
 # Install artifacts
 download_cache/

open_spiel/algorithms/infostate_tree.cc

@@ -214,15 +214,16 @@ void InfostateTree::CollectNodesAtDepth(InfostateNode* node, size_t depth) {
     CollectNodesAtDepth(child, depth + 1);
 }
 
-std::ostream& InfostateTree::operator<<(std::ostream& os) const {
-  return os << "Infostate tree for player " << acting_player_ << ".\n"
-            << "Tree height: " << tree_height_ << '\n'
-            << "Root branching: " << root_branching_factor() << '\n'
-            << "Number of decision infostate nodes: " << num_decisions() << '\n'
-            << "Number of sequences: " << num_sequences() << '\n'
-            << "Number of leaves: " << num_leaves() << '\n'
+std::ostream& operator<<(std::ostream& os, const InfostateTree& tree) {
+  return os << "Infostate tree for player " << tree.acting_player_ << ".\n"
+            << "Tree height: " << tree.tree_height_ << '\n'
+            << "Root branching: " << tree.root_branching_factor() << '\n'
+            << "Number of decision infostate nodes: " << tree.num_decisions()
+            << '\n'
+            << "Number of sequences: " << tree.num_sequences() << '\n'
+            << "Number of leaves: " << tree.num_leaves() << '\n'
             << "Tree certificate: " << '\n'
-            << root().MakeCertificate() << '\n';
+            << tree.root().MakeCertificate() << '\n';
 }
 
 std::unique_ptr<InfostateNode> InfostateTree::MakeNode(
@@ -536,7 +537,7 @@ absl::optional<DecisionId> InfostateTree::DecisionIdForSequence(
   }
 }
 absl::optional<InfostateNode*> InfostateTree::DecisionForSequence(
-    const SequenceId& sequence_id) {
+    const SequenceId& sequence_id) const {
   SPIEL_DCHECK_TRUE(sequence_id.BelongsToTree(this));
   InfostateNode* node = sequences_.at(sequence_id.id());
   SPIEL_DCHECK_TRUE(node);
@@ -646,7 +647,7 @@ std::pair<size_t, size_t> InfostateTree::CollectStartEndSequenceIds(
 }
 
 std::pair<double, SfStrategy> InfostateTree::BestResponse(
-    TreeplexVector<double>&& gradient) const {
+    TreeplexVector<double> gradient) const {
   SPIEL_CHECK_EQ(this, gradient.tree());
   SPIEL_CHECK_EQ(num_sequences(), gradient.size());
   SfStrategy response(this);
@@ -698,7 +699,7 @@ std::pair<double, SfStrategy> InfostateTree::BestResponse(
   return {gradient[empty_sequence()], response};
 }
 
-double InfostateTree::BestResponseValue(LeafVector<double>&& gradient) const {
+double InfostateTree::BestResponseValue(LeafVector<double> gradient) const {
   // Loop over all heights.
   for (int d = tree_height_ - 1; d >= 0; d--) {
     int left_offset = 0;

open_spiel/algorithms/infostate_tree.h

@@ -240,6 +240,9 @@ class RangeIterator {
   bool operator!=(const RangeIterator& other) const {
     return id_ != other.id_ || tree_ != other.tree_;
   }
+  bool operator==(const RangeIterator& other) const {
+    return !(this->operator!=(other));
+  }
   Id operator*() { return Id(id_, tree_); }
 };
 
@@ -289,7 +292,7 @@ std::shared_ptr<InfostateTree> MakeInfostateTree(
     const std::vector<InfostateNode*>& start_nodes,
     bool store_world_states = false, int max_move_ahead_limit = 1000);
 
-class InfostateTree final {
+class InfostateTree final : public std::enable_shared_from_this<InfostateTree> {
   // Note that only MakeInfostateTree is allowed to call the constructor
   // to ensure the trees are always allocated on heap. We do this so that all
   // the collected pointers are valid throughout the tree's lifetime even if
@@ -310,6 +313,12 @@ class InfostateTree final {
       const std::vector<const InfostateNode*>&, bool, int);
 
  public:
+  // -- gain shared ownership of the allocated infostate object
+  std::shared_ptr<InfostateTree> shared_ptr() { return shared_from_this(); }
+  std::shared_ptr<const InfostateTree> shared_ptr() const {
+    return shared_from_this();
+  }
+
   // -- Root accessors ---------------------------------------------------------
   const InfostateNode& root() const { return *root_; }
   InfostateNode* mutable_root() { return root_.get(); }
@@ -349,7 +358,8 @@ class InfostateTree final {
   // Returns `None` if the sequence is the empty sequence.
   absl::optional<DecisionId> DecisionIdForSequence(const SequenceId&) const;
   // Returns `None` if the sequence is the empty sequence.
-  absl::optional<InfostateNode*> DecisionForSequence(const SequenceId&);
+  absl::optional<InfostateNode*> DecisionForSequence(
+      const SequenceId& sequence_id) const;
   // Returns whether the sequence ends with the last action the player can make.
   bool IsLeafSequence(const SequenceId&) const;
 
@@ -392,13 +402,13 @@ class InfostateTree final {
   // Compute best response and value based on gradient from opponents.
   // This consumes the gradient vector, as it is used to compute the value.
   std::pair<double, SfStrategy> BestResponse(
-      TreeplexVector<double>&& gradient) const;
+      TreeplexVector<double> gradient) const;
   // Compute best response value based on gradient from opponents over leaves.
   // This consumes the gradient vector, as it is used to compute the value.
-  double BestResponseValue(LeafVector<double>&& gradient) const;
+  double BestResponseValue(LeafVector<double> gradient) const;
 
   // -- For debugging ----------------------------------------------------------
-  std::ostream& operator<<(std::ostream& os) const;
+  friend std::ostream& operator<<(std::ostream& os, const InfostateTree& tree);
 
  private:
   const Player acting_player_;

open_spiel/python/CMakeLists.txt

@@ -26,6 +26,9 @@ endif()
 # List of all Python bindings to add to pyspiel.
 include_directories (../pybind11_abseil ../../pybind11/include)
 set(PYTHON_BINDINGS ${PYTHON_BINDINGS}
+  pybind11/algorithms_infostate_tree.cc
+  pybind11/algorithms_infostate_tree.tcc
+  pybind11/algorithms_infostate_tree.h
   pybind11/algorithms_corr_dist.cc
   pybind11/algorithms_corr_dist.h
   pybind11/algorithms_trajectories.cc
@@ -197,6 +200,7 @@ set(PYTHON_TESTS ${PYTHON_TESTS}
   algorithms/generate_playthrough_test.py
   algorithms/get_all_states_test.py
   algorithms/ismcts_agent_test.py
+  algorithms/infostate_tree_test.py
   algorithms/mcts_agent_test.py
   algorithms/mcts_test.py
   algorithms/minimax_test.py