Cover corner cases in tests

Author: Baschdl

tests/integration_test.py

@@ -27,16 +27,18 @@ def setUpClass(cls):
         scvelo.tl.velocity_embedding(adata)
         cls.__adata = adata
 
-    @parameterized.expand([['neighbors', 2], ['velocities', None], ['velocities', 2],
-                           ['velocities-explicit', None]])
-    def test_run_hypothesis_test_on_pancreas(self, null_distribution, cosine_empty_neighborhood):
+    @parameterized.expand([['neighbors', 2, None], ['neighbors', 2, 10], ['velocities', None, None],
+                           ['velocities', None, 10], ['velocities', 2, None], ['velocities', 2, 10],
+                           ['velocities-explicit', None, None], ['velocities-explicit', None, 10]])
+    def test_run_hypothesis_test_on_pancreas(self, null_distribution, cosine_empty_neighborhood, exclusion_degree):
         adata = self.__adata.copy()
 
         # Run test
         uncorrected_p_values, h0_rejected, _ = run_hypothesis_test_on(adata, number_neighborhoods=100,
                                                                       number_neighbors_to_sample_from=20,
                                                                       null_distribution=null_distribution,
-                                                                      cosine_empty_neighborhood=cosine_empty_neighborhood)
+                                                                      cosine_empty_neighborhood=cosine_empty_neighborhood,
+                                                                      exclusion_degree=exclusion_degree)
         assert uncorrected_p_values.shape[0] == adata.n_obs
         assert h0_rejected.shape[0] == adata.n_obs
         if null_distribution == 'neighbors':

tests/test_test_statistic_function.py

@@ -72,7 +72,8 @@ def test_matching_p_values_old_implementation(self, exclusion_deg):
                                                             number_neighbors_to_sample_from=15,
                                                             null_distribution="velocities",
                                                             cosine_empty_neighborhood=None,
-                                                            exclusion_degree=exclusion_deg)
+                                                            exclusion_degree=exclusion_deg,
+                                                            restrict_to_velocity_genes=False)
         p_values_explicit, _ = run_explicit_test_from(adata, number_neighbors_to_sample_from=15,
                                                       exclusion_gamma_deg=exclusion_deg)
 

velotest/hypothesis_testing.py

@@ -262,13 +262,13 @@ def run_hypothesis_test(
 
     if null_distribution == 'neighbors' or null_distribution == 'velocities':
         debug_dict['neighborhoods_all'] = neighborhoods
-        if cosine_empty_neighborhood is not None:
+        if isinstance(test_statistics, torch.Tensor):
             test_statistics_velocity = test_statistics[:, 0]
             test_statistics_random = test_statistics[:, 1:]
         else:
             test_statistics_velocity = [test_statistic[0] for test_statistic in test_statistics]
             test_statistics_random = [test_statistic[1:] for test_statistic in test_statistics]
-        if cosine_empty_neighborhood is not None:
+        if isinstance(test_statistics, torch.Tensor):
             p_values_ = p_values(test_statistics_velocity, test_statistics_random).numpy()
         else:
             p_values_ = p_values_list(test_statistics_velocity, test_statistics_random).numpy()

velotest/test_statistic.py

@@ -96,7 +96,7 @@ def mean_cos_directionality_varying_neighbors(expression: torch.Tensor,
     number_neighborhoods = len(neighborhoods[0])
     if cosine_empty_neighborhood is not None:
         mean_cos_neighborhoods = torch.zeros((number_cells, number_neighborhoods))
-        used_neighborhoods = None
+        used_neighborhoods = torch.ones((number_cells, number_neighborhoods), dtype=torch.bool)
     else:
         mean_cos_neighborhoods = []
         used_neighborhoods = torch.ones((number_cells, number_neighborhoods), dtype=torch.bool)