perf: hoist _tiecorrect out of per-group loop in vs-rest Wilcoxon (#4090)

Co-authored-by: Philipp A. <flying-sheep@web.de>
Co-authored-by: Copilot <copilot@github.com>
Co-authored-by: Ilay Kavitzky <ilay.kavitzky@gmail.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Cuiwei Gao <48gaocuiwei@gmail.com>
Co-authored-by: Jhonatan Felix <108437587+JhonatanFelix@users.noreply.github.com>
Co-authored-by: Ilan Gold <ilanbassgold@gmail.com>

Author: 8 people

src/scanpy/tools/_rank_genes_groups.py

@@ -366,11 +366,11 @@ def wilcoxon(
                 tc_coef = np.zeros((n_groups, n_genes))
 
             for ranks, left, right in _ranks(self.X):
+                if tie_correct:
+                    tc_coef[:, left:right] = _tiecorrect(ranks)
                 # sum up adjusted_ranks to calculate W_m,n
                 for group_index, mask_obs in enumerate(self.groups_masks_obs):
                     scores[group_index, left:right] = ranks[mask_obs, :].sum(axis=0)
-                    if tie_correct:
-                        tc_coef[group_index, left:right] = _tiecorrect(ranks)
 
             for group_index, mask_obs in enumerate(self.groups_masks_obs):
                 n_active = np.count_nonzero(mask_obs)