def _test_game(self, game, expected_supports, expected_minimax_values):
support_p1, support_p2, \
minimax_val_p1, minimax_val_p2 = solve_zero_sum_game(game)
supports = (support_p1, support_p2)
minimax_vals = (minimax_val_p1, minimax_val_p2)
# Supports should range over the same number of actions
for actual_s, expected_s in zip(supports, expected_supports):
self.assertEqual(len(actual_s), len(expected_s))
# Supports should distribute probability mass as expected
for actual_s, expected_s in zip(supports, expected_supports):
for actual_action_support, expected_action_support in zip(
actual_s, expected_s):
np.testing.assert_array_almost_equal(actual_action_support,
expected_action_support)
# Minimax values should match
for actual_minimax_val, expected_minimax_val in zip(
minimax_vals, expected_minimax_values):
self.assertAlmostEqual(actual_minimax_val, expected_minimax_val)
def default_meta_game_solver(winrate_matrix: np.ndarray):
return solve_zero_sum_game(winrate_matrix)[0].reshape((-1))
def test_for_none_game_raises_valueerror(self):
with pytest.raises(ValueError) as _:
_ = solve_zero_sum_game(None)
def test_for_non_integer_or_float_list_raises_valueerror(self):
with pytest.raises(ValueError) as _:
_ = solve_zero_sum_game([['a', 'b']])
def test_for_empty_numpy_array_game_raises_valueerror(self):
with pytest.raises(ValueError) as _:
_ = solve_zero_sum_game(np.array(None))