Merge pull request #76 from pbarbarant/fix/fix-barycenter-tests

[BUGFIX] Fixes CI + other tests

src/fugw/scripts/coarse_to_fine.py

@@ -316,10 +316,13 @@ def compute_sparsity_mask(
             device = torch.device("cuda", 0)
         else:
             device = torch.device("cpu")
+
+    # Convert coarse plan to numpy
+    coarse_plan = coarse_mapping.pi.to("cpu").numpy()
     if method == "quantile":
         # Method 1: keep first percentile
-        threshold = np.percentile(coarse_mapping.pi, 99.95)
-        rows, cols = np.nonzero(coarse_mapping.pi > threshold)
+        threshold = np.percentile(coarse_plan, 99.95)
+        rows, cols = np.nonzero(coarse_plan > threshold)
 
     elif method == "topk":
         # Method 2: keep topk indices per line and per column
@@ -328,12 +331,12 @@ def compute_sparsity_mask(
         rows = np.concatenate(
             [
                 np.arange(source_sample.shape[0]),
-                np.argmax(coarse_mapping.pi, axis=0),
+                np.argmax(coarse_plan, axis=0),
             ]
         )
         cols = np.concatenate(
             [
-                np.argmax(coarse_mapping.pi, axis=1),
+                np.argmax(coarse_plan, axis=1),
                 np.arange(target_sample.shape[0]),
             ]
         )

tests/mappings/test_barycenter.py

@@ -19,10 +19,12 @@
     product(devices, callbacks),
 )
 def test_fugw_barycenter(device, callback):
+    """Tests the FUGW barycenter fitting on toy data."""
     np.random.seed(0)
     n_subjects = 4
     n_voxels = 100
     n_features = 10
+    nits_barycenter = 3
 
     # Generate random training data for n subjects
     features_list = []
@@ -38,10 +40,30 @@ def test_fugw_barycenter(device, callback):
         geometry_list.append(geometry)
 
     fugw_barycenter = FUGWBarycenter()
-    fugw_barycenter.fit(
+
+    # Fit the barycenter
+    (
+        barycenter_weights,
+        barycenter_features,
+        barycenter_geometry,
+        plans,
+        _,
+        losses_each_bar_step,
+    ) = fugw_barycenter.fit(
         weights_list,
         features_list,
         geometry_list,
+        solver_params={"nits_bcd": 2, "nits_uot": 5},
+        nits_barycenter=nits_barycenter,
         device=device,
         callback_barycenter=callback,
     )
+
+    assert isinstance(barycenter_weights, torch.Tensor)
+    assert barycenter_weights.shape == (n_voxels,)
+    assert isinstance(barycenter_features, torch.Tensor)
+    assert barycenter_features.shape == (n_features, n_voxels)
+    assert isinstance(barycenter_geometry, torch.Tensor)
+    assert barycenter_geometry.shape == (n_voxels, n_voxels)
+    assert len(plans) == n_subjects
+    assert len(losses_each_bar_step) == nits_barycenter

tests/mappings/test_sparse_barycenter.py

@@ -19,7 +19,8 @@
     "device, callback",
     product(devices, callbacks),
 )
-def test_fugw_barycenter(device, callback):
+def test_fugw_sparse_barycenter(device, callback):
+    """Tests the FUGW sparse barycenter fitting on toy data."""
     np.random.seed(0)
     n_subjects = 4
     n_voxels = 100
@@ -39,23 +40,27 @@ def test_fugw_barycenter(device, callback):
         weights_list.append(weights)
         features_list.append(features)
 
-    fugw_barycenter = FUGWSparseBarycenter()
+    geometry_embedding_normalized = (
+        geometry_embedding / geometry_embedding.norm()
+    )
+    fugw_sparse_barycenter = FUGWSparseBarycenter()
 
     # Fit the barycenter
     (
         barycenter_weights,
         barycenter_features,
         plans,
         losses_each_bar_step,
-    ) = fugw_barycenter.fit(
+    ) = fugw_sparse_barycenter.fit(
         weights_list,
         features_list,
-        geometry_embedding,
+        geometry_embedding_normalized,
         mesh_sample=mesh_sample,
         coarse_mapping_solver_params={"nits_bcd": 2, "nits_uot": 5},
         fine_mapping_solver_params={"nits_bcd": 2, "nits_uot": 5},
         nits_barycenter=nits_barycenter,
         device=device,
+        callback_barycenter=callback,
     )
 
     assert isinstance(barycenter_weights, torch.Tensor)