fixed a sklearn 1.6 issue but disabled it due to skorch incompatibility

Author: dholzmueller

README.md

@@ -136,7 +136,8 @@ and https://docs.ray.io/en/latest/cluster/vms/user-guides/community/slurm.html
 ## Releases (see git tags)
 
 - v1.1.2: 
-  - Compatibility fixes for scikit-learn 1.6.
+  - Some compatibility improvements for scikit-learn 1.6 
+    (but disabled 1.6 since skorch is not compatible with it).
   - Improved documentation for Pytorch-Lightning interface.
   - Other small bugfixes and improvements.
 - v1.1.1:

docs/requirements.txt

@@ -21,7 +21,7 @@ pytorch_lightning>=2.0
 pyyaml>=5.0
 ray>=2.8
 requests>=2.0
-scikit-learn>=1.3
+scikit-learn>=1.3,<1.6
 seaborn>=0.0.13
 skorch>=0.15
 sphinx>=7.0

pyproject.toml

@@ -29,7 +29,7 @@ dependencies = [
     "torch>=2.0",
     "numpy>=1.25,<2.0",
     "pandas>=2.0",
-    "scikit-learn>=1.3",
+    "scikit-learn>=1.3,<1.6",
     "xgboost>=2.0",
     "catboost>=1.2",
     "lightgbm>=4.1",

pytabkit/models/data/conversion.py

@@ -87,7 +87,10 @@ def transform(self, x: Union[np.ndarray, pd.DataFrame, pd.Series, DictDataset])
 
         if set(x.columns) != self.fitted_columns:
             print('Raising column error')
-            raise ValueError(f'Different columns during fit() and predict(): {self.fitted_columns} and {set(x.columns)}')
+            # second line is to satisfy the sklearn test
+            # check_n_features_in_after_fitting in scikit-learn >= 1.6
+            raise ValueError(f'Different columns during fit() and predict(): {self.fitted_columns} and {set(x.columns)}\n'
+                             f'X has {len(x.columns)} features, but estimator is expecting {len(self.fitted_columns)} features as input')
 
         x_cont = torch.as_tensor(self.num_tf.transform(x), dtype=torch.float32)
         x_cat = torch.as_tensor(self.cat_tf.transform(x) + 1, dtype=torch.long)