[WIP] Harmonize fixtures (#248)

* fixtures return dataset

* fix test_mapping

* fix test_selector

* fix test label_prop

* fix test reweight

* fix test ot

* fix test pipeline

* getting rid of wrong fixture

* fix test scorer

* bring back da_binary_dataset

conftest.py

@@ -23,16 +23,16 @@ def set_seed():
 
 @pytest.fixture(scope='session')
 def da_reg_dataset():
-    X, y, sample_domain = make_shifted_datasets(
+    return make_shifted_datasets(
         n_samples_source=20,
         n_samples_target=21,
         shift="concept_drift",
         mean=0.5,
         noise=0.3,
         label="regression",
         random_state=43,
+        return_dataset=True,
     )
-    return X, y, sample_domain
 
 
 @pytest.fixture(scope='session')
@@ -45,6 +45,7 @@ def da_reg_datasets():
         noise=0.3,
         label="regression",
         random_state=42,
+        return_dataset=True,
     )
 
     da_reg_dataset_2 = make_shifted_datasets(
@@ -55,33 +56,34 @@ def da_reg_datasets():
         noise=0.3,
         label="regression",
         random_state=42,
+        return_dataset=True,
     )
     return da_reg_dataset_1, da_reg_dataset_2
 
 @pytest.fixture(scope='session')
 def da_multiclass_dataset():
-    X, y, sample_domain = make_shifted_datasets(
+    return make_shifted_datasets(
         n_samples_source=20,
         n_samples_target=21,
         shift="concept_drift",
         noise=0.1,
         label="multiclass",
         random_state=42,
+        return_dataset=True,
     )
-    return X, y, sample_domain
 
 
 @pytest.fixture(scope='session')
 def da_binary_dataset():
-    X, y, sample_domain = make_shifted_datasets(
+    return make_shifted_datasets(
         n_samples_source=20,
         n_samples_target=21,
         shift="concept_drift",
         noise=0.1,
         label="binary",
         random_state=42,
+        return_dataset=True,
     )
-    return X, y, sample_domain
 
 
 @pytest.fixture(scope='session')
@@ -95,7 +97,7 @@ def da_blobs_dataset():
         shift=0.13,
         random_state=42,
         cluster_std=0.05,
-        return_X_y=True,
+        return_dataset=True,
     )
 
 

skada/tests/test_label_prop.py

@@ -32,7 +32,7 @@
     ],
 )
 def test_label_prop_estimator(estimator, da_blobs_dataset):
-    X, y, sample_domain = da_blobs_dataset
+    X, y, sample_domain = da_blobs_dataset.pack(as_sources=["s"], as_targets=["t"])
     X_source, X_target, y_source, y_target = source_target_split(
         X, y, sample_domain=sample_domain
     )
@@ -69,7 +69,7 @@ def test_label_prop_estimator(estimator, da_blobs_dataset):
     ],
 )
 def test_label_prop_estimator_reg(estimator, da_reg_dataset):
-    X, y, sample_domain = da_reg_dataset
+    X, y, sample_domain = da_reg_dataset.pack(as_sources=["s"], as_targets=["t"])
     X_source, X_target, y_source, y_target = source_target_split(
         X, y, sample_domain=sample_domain
     )

skada/tests/test_mapping.py

@@ -75,7 +75,7 @@
     ],
 )
 def test_mapping_estimator(estimator, da_blobs_dataset):
-    X, y, sample_domain = da_blobs_dataset
+    X, y, sample_domain = da_blobs_dataset.pack(as_sources=["s"], as_targets=["t"])
     X_source, X_target, y_source, y_target = source_target_split(
         X, y, sample_domain=sample_domain
     )
@@ -257,7 +257,7 @@ def test_reg_new_X_adapt(estimator):
     ],
 )
 def test_mapping_source_samples(estimator, da_blobs_dataset):
-    X, y, sample_domain = da_blobs_dataset
+    X, y, sample_domain = da_blobs_dataset.pack(as_sources=["s"], as_targets=["t"])
     X_source, X_target, y_source, y_target = source_target_split(
         X, y, sample_domain=sample_domain
     )

skada/tests/test_ot.py

@@ -15,7 +15,7 @@
 
 
 def test_JDOTRegressor(da_reg_dataset):
-    X, y, sample_domain = da_reg_dataset
+    X, y, sample_domain = da_reg_dataset.pack(as_sources=["s"], as_targets=["t"])
     rng = np.random.default_rng(42)
     w = rng.uniform(size=(X.shape[0],))
 
@@ -42,7 +42,7 @@ def test_JDOTRegressor(da_reg_dataset):
 
 
 def test_JDOTRegressor_pipeline(da_reg_dataset):
-    X, y, sample_domain = da_reg_dataset
+    X, y, sample_domain = da_reg_dataset.pack(as_sources=["s"], as_targets=["t"])
     Xs, Xt, ys, yt = source_target_split(X, y, sample_domain=sample_domain)
 
     jdot = make_da_pipeline(
@@ -60,7 +60,7 @@ def test_JDOTRegressor_pipeline(da_reg_dataset):
 def test_JDOTClassifier(da_multiclass_dataset, da_binary_dataset):
     rng = np.random.default_rng(43)
     for dataset in [da_multiclass_dataset, da_binary_dataset]:
-        X, y, sample_domain = dataset
+        X, y, sample_domain = dataset.pack(as_sources=["s"], as_targets=["t"])
         w = rng.uniform(size=(X.shape[0],))
         Xs, Xt, ys, yt = source_target_split(X, y, sample_domain=sample_domain)
 
@@ -123,7 +123,7 @@ def test_JDOTClassifier(da_multiclass_dataset, da_binary_dataset):
 
 
 def test_jdot_class_cost_matrix(da_multiclass_dataset):
-    X, y, sample_domain = da_multiclass_dataset
+    X, y, sample_domain = da_multiclass_dataset.pack(as_sources=["s"], as_targets=["t"])
 
     Xs, Xt, ys, yt = source_target_split(X, y, sample_domain=sample_domain)
 
@@ -157,7 +157,7 @@ def test_jdot_class_cost_matrix(da_multiclass_dataset):
 
 
 def test_jdot_class_tgt_loss(da_multiclass_dataset):
-    X, y, sample_domain = da_multiclass_dataset
+    X, y, sample_domain = da_multiclass_dataset.pack(as_sources=["s"], as_targets=["t"])
 
     Xs, Xt, ys, yt = source_target_split(X, y, sample_domain=sample_domain)
 

skada/tests/test_pipeline.py

@@ -212,7 +212,7 @@ def fit_transform(self, X, y=None, **params):
 
 
 def test_adaptation_output_propagate_labels(da_reg_dataset):
-    X, y, sample_domain = da_reg_dataset
+    X, y, sample_domain = da_reg_dataset.pack(as_sources=["s"], as_targets=["t"])
     _, X_target, _, target_domain = source_target_split(
         X, sample_domain, sample_domain=sample_domain
     )

skada/tests/test_reweight.py

@@ -161,7 +161,9 @@ def test_reg_reweight_estimator(estimator):
 
 
 def _base_test_new_X_adapt(estimator, da_dataset):
-    X_train, y_train, sample_domain = da_dataset
+    X_train, y_train, sample_domain = da_dataset.pack_train(
+        as_sources=["s"], as_targets=["t"]
+    )
 
     # fit works with no errors
     estimator.fit(X_train, y_train, sample_domain=sample_domain)
@@ -281,7 +283,7 @@ def test_KMMReweight_new_X_adapt(da_dataset):
     ],
 )
 def test_adaptation_output_propagation_multiple_steps(da_reg_dataset, mediator):
-    X, y, sample_domain = da_reg_dataset
+    X, y, sample_domain = da_reg_dataset.pack(as_sources=["s"], as_targets=["t"])
     _, X_target, _, target_domain = source_target_split(
         X, sample_domain, sample_domain=sample_domain
     )
@@ -307,7 +309,7 @@ def predict(self, X, sample_weight=None):
 
 
 def test_select_source_target_output_merge(da_reg_dataset):
-    X, y, sample_domain = da_reg_dataset
+    X, y, sample_domain = da_reg_dataset.pack(as_sources=["s"], as_targets=["t"])
     _, X_target, _, target_domain = source_target_split(
         X, sample_domain, sample_domain=sample_domain
     )

skada/tests/test_scorer.py

@@ -300,7 +300,7 @@ def test_mixval_scorer(da_dataset):
 
 
 def test_mixval_scorer_regression(da_reg_dataset):
-    X, y, sample_domain = da_reg_dataset
+    X, y, sample_domain = da_reg_dataset.pack(as_sources=["s"], as_targets=["t"])
 
     estimator = make_da_pipeline(DensityReweightAdapter(), LinearRegression())
 

skada/tests/test_selector.py

@@ -196,7 +196,7 @@ def transform(
     ],
 )
 def test_source_selector_with_estimator(da_multiclass_dataset, selector_cls, side):
-    X, y, sample_domain = da_multiclass_dataset
+    X, y, sample_domain = da_multiclass_dataset.pack(as_sources=["s"], as_targets=["t"])
     X_source, X_target = source_target_split(X, sample_domain=sample_domain)
     output = {}
 
@@ -240,7 +240,7 @@ def predict(self, X):
 def test_source_selector_with_transformer(
     da_multiclass_dataset, selector_cls, side, _fit_transform
 ):
-    X, y, sample_domain = da_multiclass_dataset
+    X, y, sample_domain = da_multiclass_dataset.pack(as_sources=["s"], as_targets=["t"])
     X_source, X_target = source_target_split(X, sample_domain=sample_domain)
     output = {}
 
@@ -282,7 +282,7 @@ def fit_transform(self, X, y=None):
     ],
 )
 def test_source_selector_with_weights(da_multiclass_dataset, selector_cls, side):
-    X, y, sample_domain = da_multiclass_dataset
+    X, y, sample_domain = da_multiclass_dataset.pack(as_sources=["s"], as_targets=["t"])
     sample_weight = np.ones(X.shape[0])
     X_source, X_target = source_target_split(X, sample_domain=sample_domain)
     output = {}
@@ -318,7 +318,7 @@ def predict(self, X, sample_weight=None):
 def test_source_target_selector(
     da_multiclass_dataset, source_estimator, target_estimator
 ):
-    X, y, sample_domain = da_multiclass_dataset
+    X, y, sample_domain = da_multiclass_dataset.pack(as_sources=["s"], as_targets=["t"])
     source_masks = extract_source_indices(sample_domain)
     # make sure sources and targets have significantly different mean
     X[source_masks] += 100 * np.ones((source_masks.sum(), X.shape[1]))
@@ -354,7 +354,7 @@ def test_source_target_selector(
 
 
 def test_source_target_selector_fails_on_missing_domain(da_multiclass_dataset):
-    X, y, sample_domain = da_multiclass_dataset
+    X, y, sample_domain = da_multiclass_dataset.pack(as_sources=["s"], as_targets=["t"])
     source_masks = extract_source_indices(sample_domain)
     pipe = make_da_pipeline(SelectSourceTarget(StandardScaler()), SVC())