test: Add predict and predict_proba tests for GWLogisticRegression

gwlearn/tests/test_linear_model.py

@@ -279,3 +279,72 @@ def test_against_mgwr():
     assert_almost_equal(gwlr.aicc_, res.aicc, decimal=0)
     assert_almost_equal(gwlr.effective_df_, res.ENP)
     assert_almost_equal(gwlr.log_likelihood_, res.llf)
+
+
+def test_gwlogistic_predict(sample_data):
+    # Unpack sample dataset (features, target, spatial geometry)
+    X, y, geometry = sample_data
+
+    # Initialize model with keep_models=True (required for prediction)
+    model = GWLogisticRegression(
+        bandwidth=10,
+        fixed=False,
+        keep_models=True,
+        max_iter=1000,
+    )
+
+    # Fit geographically weighted logistic model
+    model.fit(X, y, geometry=geometry)
+
+    # Generate predictions using fitted local models
+    preds = model.predict(X, geometry=geometry)
+
+    # Assert: number of predictions matches number of input samples
+    assert len(preds) == len(X)
+
+    # Assert: predictions are valid binary outputs (0/1 or True/False)
+    assert set(np.unique(preds.dropna())).issubset({0, 1})
+
+    # Ensure predictions are not all NaN
+    assert not preds.isna().all()
+
+
+def test_gwlogistic_predict_proba(sample_data):
+    # Unpack sample dataset
+    X, y, geometry = sample_data
+
+    # Initialize model with keep_models=True for probability prediction
+    model = GWLogisticRegression(
+        bandwidth=10,
+        fixed=False,
+        keep_models=True,
+        max_iter=1000,
+    )
+
+    # Fit model
+    model.fit(X, y, geometry=geometry)
+
+    # Get class probability predictions
+    proba = model.predict_proba(X, geometry=geometry)
+
+    # Assert: number of rows equals number of samples
+    assert proba.shape[0] == len(X)
+
+    # Assert: binary classification → exactly 2 probability columns
+    assert proba.shape[1] == 2
+
+
+def test_predict_requires_keep_models(sample_data):
+    # Unpack sample dataset
+    X, y, geometry = sample_data
+
+    # Initialize model WITHOUT storing local models
+    model = GWLogisticRegression(
+        bandwidth=10, fixed=False, keep_models=False, max_iter=1000
+    )
+    # Fit model
+    model.fit(X, y, geometry=geometry)
+
+    # prediction requires stored local models → should fail
+    with pytest.raises(AttributeError, match="_local_models"):
+        model.predict(X, geometry=geometry)