test demonstrating issue is to do with infoset order

Author: rahulsavani

tests/games.py

@@ -352,7 +352,7 @@ def create_3_player_with_internal_outcomes_efg(nonterm_outcomes: bool = False) -
         o = g.add_outcome([0, 0, 0])
         g.set_outcome(g.root.children[0].children[0].children[1].children[0], o)
         g.set_outcome(g.root.children[0].children[0].children[1].children[1], o)
-    g.to_efg(f"TEST_{nonterm_outcomes}.efg")
+    # g.sort_infosets()
     return g
 
 
@@ -652,7 +652,7 @@ def create_kuhn_poker_efg(nonterm_outcomes: bool = False) -> gbt.Game:
         g = _create_kuhn_poker_efg_only_term_outcomes()
 
     # Ensure infosets are in the same order as if game was written to efg and read back in
-    g.sort_infosets()
+    # g.sort_infosets()
     return g
 
 
@@ -850,6 +850,107 @@ def create_reduction_both_players_payoff_ties_efg() -> gbt.Game:
     return g
 
 
+def create_problem_example_efg() -> gbt.Game:
+    g = gbt.Game.new_tree(players=["1", "2"], title="")
+    g.append_move(g.root, player="1", actions=["L", "R"])
+    # do the second child first on purpose to diverge from sort infosets order
+    g.append_move(g.root.children[1], "2", actions=["l2", "r2"])
+    g.append_move(g.root.children[0], "2", actions=["l1", "r1"])
+    g.set_outcome(g.root.children[0].children[0], outcome=g.add_outcome(payoffs=[5, -5]))
+    g.set_outcome(g.root.children[0].children[1], outcome=g.add_outcome(payoffs=[2, -2]))
+    g.set_outcome(g.root.children[1].children[0], outcome=g.add_outcome(payoffs=[-5, 5]))
+    g.set_outcome(g.root.children[1].children[1], outcome=g.add_outcome(payoffs=[-2, 2]))
+    return g
+
+
+def create_STOC_simplified() -> gbt.Game:
+    """
+    """
+    g = gbt.Game.new_tree(players=["1", "2"], title="")
+    g.append_move(g.root, g.players.chance, actions=["1", "2"])
+    g.set_chance_probs(g.root.infoset, [0.2, 0.8])
+    g.append_move(g.root.children[0], player="1", actions=["l", "r"])
+    g.append_move(g.root.children[1], player="1", actions=["c", "d"])
+    g.append_move(g.root.children[0].children[1], player="2", actions=["p", "q"])
+    g.append_move(
+        g.root.children[0].children[1].children[0], player="1", actions=["L", "R"]
+    )
+    g.append_infoset(
+        g.root.children[0].children[1].children[1],
+        g.root.children[0].children[1].children[0].infoset,
+    )
+    g.set_outcome(
+        g.root.children[0].children[0],
+        outcome=g.add_outcome(payoffs=[5, -5], label="l"),
+    )
+    g.set_outcome(
+        g.root.children[0].children[1].children[0].children[0],
+        outcome=g.add_outcome(payoffs=[10, -10], label="rpL"),
+    )
+    g.set_outcome(
+        g.root.children[0].children[1].children[0].children[1],
+        outcome=g.add_outcome(payoffs=[15, -15], label="rpR"),
+    )
+    g.set_outcome(
+        g.root.children[0].children[1].children[1].children[0],
+        outcome=g.add_outcome(payoffs=[20, -20], label="rqL"),
+    )
+    g.set_outcome(
+        g.root.children[0].children[1].children[1].children[1],
+        outcome=g.add_outcome(payoffs=[-5, 5], label="rqR"),
+    )
+    g.set_outcome(
+        g.root.children[1].children[0],
+        outcome=g.add_outcome(payoffs=[10, -10], label="c"),
+    )
+    g.set_outcome(
+        g.root.children[1].children[1],
+        outcome=g.add_outcome(payoffs=[20, -20], label="d"),
+    )
+    # g.sort_infosets()
+    return g
+
+
+def create_STOC_simplified2() -> gbt.Game:
+    """
+    """
+    g = gbt.Game.new_tree(players=["1", "2"], title="")
+    g.append_move(g.root, g.players.chance, actions=["1", "2"])
+    g.set_chance_probs(g.root.infoset, [0.2, 0.8])
+    g.append_move(g.root.children[0], player="1", actions=["r"])
+    g.append_move(g.root.children[1], player="1", actions=["c"])
+    g.append_move(g.root.children[0].children[0], player="2", actions=["p", "q"])
+    g.append_move(
+        g.root.children[0].children[0].children[0], player="1", actions=["L", "R"]
+    )
+    g.append_infoset(
+        g.root.children[0].children[0].children[1],
+        g.root.children[0].children[0].children[0].infoset,
+    )
+    g.set_outcome(
+        g.root.children[0].children[0].children[0].children[0],
+        outcome=g.add_outcome(payoffs=[10, -10], label="rpL"),
+    )
+    g.set_outcome(
+        g.root.children[0].children[0].children[0].children[1],
+        outcome=g.add_outcome(payoffs=[15, -15], label="rpR"),
+    )
+    g.set_outcome(
+        g.root.children[0].children[0].children[1].children[0],
+        outcome=g.add_outcome(payoffs=[20, -20], label="rqL"),
+    )
+    g.set_outcome(
+        g.root.children[0].children[0].children[1].children[1],
+        outcome=g.add_outcome(payoffs=[-5, 5], label="rqR"),
+    )
+    g.set_outcome(
+        g.root.children[1].children[0],
+        outcome=g.add_outcome(payoffs=[10, -10], label="c"),
+    )
+    # g.sort_infosets()
+    return g
+
+
 def create_seq_form_STOC_paper_zero_sum_2_player_efg() -> gbt.Game:
     """
     Example from
@@ -928,7 +1029,7 @@ def create_seq_form_STOC_paper_zero_sum_2_player_efg() -> gbt.Game:
     g.root.children[0].children[1].infoset.label = "01"
     g.root.children[2].children[0].infoset.label = "20"
     g.root.children[0].children[1].children[0].infoset.label = "010"
-
+    # g.sort_infosets()
     return g
 
 

tests/test_nash.py

@@ -934,3 +934,48 @@ def test_regrets_tmp2():
     print(profile_rat.max_regret())  # 3/2
     profile_rat = g.mixed_behavior_profile(rational=True, data=prof_data_rat)
     print(profile_rat.max_regret())  # now different! 0
+
+
+@pytest.mark.nash
+@pytest.mark.nash_lp_behavior
+@pytest.mark.parametrize(
+    "game,mixed_behav_prof_data",
+    [
+        (
+            games.create_seq_form_STOC_paper_zero_sum_2_player_efg(),
+            [
+                [[0, 1], ["1/3", "2/3"], ["2/3", "1/3"]],
+                [["5/6", "1/6"], ["5/9", "4/9"]],
+            ],
+        ),
+        (
+            games.create_3_player_with_internal_outcomes_efg(),
+            [
+                [[1, 0], [1, 0]], [[1, 0], ["1/2", "1/2"]],
+                [[1, 0], [0, 1]]
+            ],
+        ),
+        (
+            games.create_STOC_simplified(),
+            [
+                [[0, 1], ["1/3", "2/3"], ["2/3", "1/3"]],
+                [["5/6", "1/6"]],
+            ],
+        ),
+        # (
+        # games.create_STOC_simplified2(),
+        # [
+        # [[1], [1], ["1/3", "2/3"]],
+        # [["5/6", "1/6"]],
+        # ],
+        # ),
+    ],
+)
+def test_repeat_max_regret(game: gbt.Game, mixed_behav_prof_data: list):
+    profile1 = game.mixed_behavior_profile(rational=True, data=mixed_behav_prof_data)
+    mr1 = profile1.max_regret()
+    profile2 = game.mixed_behavior_profile(rational=True, data=mixed_behav_prof_data)
+    mr2 = profile2.max_regret()
+    print()
+    print(mr1, mr2)
+    assert mr1 == mr2