Merge pull request #1243 from plamentotev:deepcfr

PiperOrigin-RevId: 656457757
Change-Id: If1559e628ba14ed4b0cc3a7807ada6134bd3a82f

open_spiel/python/algorithms/deep_cfr.py

@@ -360,8 +360,9 @@ def _sample_action_from_advantage(self, state, player):
 
     return advantages, matched_regrets
 
-  def action_probabilities(self, state):
+  def action_probabilities(self, state, player_id=None):
     """Returns action probabilities dict for a single batch."""
+    del player_id  # unused
     cur_player = state.current_player()
     legal_actions = state.legal_actions(cur_player)
     info_state_vector = np.array(state.information_state_tensor())

open_spiel/python/algorithms/deep_cfr_tf2.py

@@ -631,8 +631,9 @@ def _sample_action_from_advantage(self, state, player):
         info_state, legal_actions_mask, player)
     return advantages.numpy(), matched_regrets.numpy()
 
-  def action_probabilities(self, state):
+  def action_probabilities(self, state, player_id=None):
     """Returns action probabilities dict for a single batch."""
+    del player_id  # unused
     cur_player = state.current_player()
     legal_actions = state.legal_actions(cur_player)
     legal_actions_mask = tf.constant(

open_spiel/python/jax/deep_cfr.py

@@ -480,8 +480,9 @@ def _sample_action_from_advantage(self, state, player):
         info_state, legal_actions_mask, self._params_adv_network[player])
     return advantages, matched_regrets
 
-  def action_probabilities(self, state):
+  def action_probabilities(self, state, player_id=None):
     """Returns action probabilities dict for a single batch."""
+    del player_id  # unused
     cur_player = state.current_player()
     legal_actions = state.legal_actions(cur_player)
     info_state_vector = jnp.array(

open_spiel/python/pytorch/deep_cfr.py

@@ -416,15 +416,17 @@ def _sample_action_from_advantage(self, state, player):
       matched_regrets[max(legal_actions, key=lambda a: raw_advantages[a])] = 1
     return advantages, matched_regrets
 
-  def action_probabilities(self, state):
+  def action_probabilities(self, state, player_id=None):
     """Computes action probabilities for the current player in state.
 
     Args:
       state: (pyspiel.State) The state to compute probabilities for.
+      player_id: unused, but needed to implement the Policy API.
 
     Returns:
       (dict) action probabilities for a single batch.
     """
+    del player_id
     cur_player = state.current_player()
     legal_actions = state.legal_actions(cur_player)
     info_state_vector = np.array(state.information_state_tensor())