EFR code issue with NumPy

[lanctot]

CI tests are failing on the EFR code: https://github.com/google-deepmind/open_spiel/actions/runs/10438874702/job/28948184064

Here is the problem:

  EFRTest.test_efr_kuhn_poker4 ('csps')
  test_efr_kuhn_poker('csps')
  ----------------------------------------------------------------------
  Traceback (most recent call last):
    File "/tmp/tmp.qKWdCXq1hQ/venv/lib/python3.10/site-packages/absl/testing/parameterized.py", line 323, in bound_param_test
      return test_method(self, testcase_params)
    File "/project/open_spiel/python/algorithms/efr_test.py", line 69, in test_efr_kuhn_poker
      efr_solver = efr.EFRSolver(
    File "/tmp/tmp.qKWdCXq1hQ/venv/lib/python3.10/site-packages/open_spiel/python/algorithms/efr.py", line 497, in __init__
      super(EFRSolver, self).__init__(game, deviation_sets)
    File "/tmp/tmp.qKWdCXq1hQ/venv/lib/python3.10/site-packages/open_spiel/python/algorithms/efr.py", line 117, in __init__
      self._initialize_info_state_nodes(self._root_node, hist, empty_path_indices)
    File "/tmp/tmp.qKWdCXq1hQ/venv/lib/python3.10/site-packages/open_spiel/python/algorithms/efr.py", line 188, in _initialize_info_state_nodes
      self._initialize_info_state_nodes(
    File "/tmp/tmp.qKWdCXq1hQ/venv/lib/python3.10/site-packages/open_spiel/python/algorithms/efr.py", line 188, in _initialize_info_state_nodes
      self._initialize_info_state_nodes(
    File "/tmp/tmp.qKWdCXq1hQ/venv/lib/python3.10/site-packages/open_spiel/python/algorithms/efr.py", line 212, in _initialize_info_state_nodes
      info_state_node.relizable_deviations = self._deviation_gen(
    File "/tmp/tmp.qKWdCXq1hQ/venv/lib/python3.10/site-packages/open_spiel/python/algorithms/efr.py", line 818, in return_cs_partial_sequence
      external = return_all_external_modified_deviations(
    File "/tmp/tmp.qKWdCXq1hQ/venv/lib/python3.10/site-packages/open_spiel/python/algorithms/efr.py", line 1236, in return_all_external_modified_deviations
      modification_index = np.where(prior_actions_weight == 0)[0][0]
  ValueError: Calling nonzero on 0d arrays is not allowed. Use np.atleast_1d(scalar).nonzero() instead. If the context of this error is of the form `arr[nonzero(cond)]`, just use `arr[cond]`.

[lanctot]

@Jamesflynn1 can you take a look?

[lanctot]

Specifically, lines 1137 and 1236 of efr.py.

NumPy seems to sometimes have an issue with this line modification_index = np.where(prior_actions_weight == 0)[0][0]. Is there a different way to write that might be more compatible across numpy/python versions?

[Jamesflynn1]

Hi @lanctot, I took a look at the offending coding and the aim of it was to find the index of the first zero of the prior_actions_weight array. There definitely should be a cleaner and more compatible way to do that in NumPy and I will take a look at that now.

[lanctot]

Fixed by #1267