[FIX] new requirements to be in line with new rdkit versions

Author: jcapels

extra-requirements.txt

@@ -1,4 +1,4 @@
-rdkit==2023.9.6: preprocessing, deep-learning, machine-learning, test
+rdkit==2025.9.6: preprocessing, deep-learning, machine-learning, test
 smilespe==0.0.3: preprocessing, deep-learning, machine-learning, test
 seaborn: preprocessing, deep-learning, machine-learning, test
 joblib: preprocessing, deep-learning, machine-learning, test

requirements.txt

@@ -1,4 +1,4 @@
-rdkit==2023.9.6
+rdkit==2025.9.6
 smilespe==0.0.3
 seaborn
 torchvision

setup.cfg

@@ -24,7 +24,7 @@ python_requires = >=3.9
 zip_safe = False
 include_package_data = True
 install_requires =
-    rdkit==2023.9.6
+    rdkit==2025.9.6
     cached_property==1.5.2
     seaborn
     joblib

src/deepmol/models/__init__.py

@@ -2,9 +2,9 @@
     from .deepchem_models import DeepChemModel
     from .transformer_models import TransformerModelForMaskedLM, DeBERTa, ModernBERT, BERT, RoBERTa
     from .atmol.atmol import AtMolLightning as ATMOL
-    from .keras_models import KerasModel
 except NameError:
     pass
 from .sklearn_models import SklearnModel
 from .ensembles import VotingClassifier
 from .models import Model
+from .keras_models import KerasModel

tests/integration_tests/dataset/test_dataset_features.py

@@ -26,7 +26,7 @@ def test_dataset_with_similarity_matrix(self):
     def test_dataset_with_nc_fp(self):
         from deepmol.compound_featurization import NcMfp
         NcMfp().featurize(self.small_dataset_to_test, inplace=True)
-        self.assertEqual(self.small_dataset_to_test.X.shape[0], 13)
+        self.assertEqual(self.small_dataset_to_test.X.shape[0], 11)
         self.assertEqual(self.small_dataset_to_test.X.shape[1], 254399)
 
     def test_dataset_with_neural_npfp(self):

tests/integration_tests/dataset/test_dataset_models.py

@@ -26,7 +26,7 @@
 import numpy as np
 
 def model_build(): # num = number of categories
-    input_f = layers.Input(shape=(210,))
+    input_f = layers.Input(shape=(217,))
 
     X = layers.Dense(2048, activation = 'relu')(input_f)
     X = layers.BatchNormalization()(X)
@@ -42,7 +42,7 @@ def model_build(): # num = number of categories
     return model
 
 def model_build_regression(): # num = number of categories
-    input_f = layers.Input(shape=(210,))
+    input_f = layers.Input(shape=(217,))
 
     X = layers.Dense(2048, activation = 'relu')(input_f)
     X = layers.BatchNormalization()(X)

tests/unit_tests/splitters/test_butina_splitter.py

@@ -131,7 +131,7 @@ def test_scaffold_spliter_non_homogenous_dataset(self):
                                                                        seed=123,
                                                                        homogenous_datasets=False)
 
-        self.assertEqual(len(train_dataset.smiles), len(test_dataset.smiles))
+        self.assertAlmostEqual(len(train_dataset.smiles), len(test_dataset.smiles), delta=1)
         self.assertAlmostEqual(
             len(train_dataset.y[train_dataset.y == 1]) / len(train_dataset.y),
             len(self.binary_dataset.y[self.binary_dataset.y == 1]) / len(self.binary_dataset.y),

tests/unit_tests/splitters/test_scaffold_splitter.py

@@ -57,7 +57,7 @@ def test_scaffold_splitter_larger_dataset_binary_classification(self):
                                                                          frac_train=0.5,
                                                                          seed=123)
 
-        self.assertEqual(len(train_dataset.smiles), len(test_dataset.smiles))
+        self.assertAlmostEqual(len(train_dataset.smiles), len(test_dataset.smiles), delta=1)
         self.assertAlmostEqual(
             len(train_dataset.y[train_dataset.y == 1]) / len(train_dataset.y),
             len(self.binary_dataset.y[self.binary_dataset.y == 1]) / len(self.binary_dataset.y),
@@ -107,7 +107,7 @@ def test_scaffold_spliter_non_homogenous_dataset(self):
                                                                          seed=123,
                                                                          homogenous_datasets=False)
 
-        self.assertEqual(len(train_dataset.smiles), len(test_dataset.smiles))
+        self.assertAlmostEqual(len(train_dataset.smiles), len(test_dataset.smiles), delta=1)
         self.assertAlmostEqual(
             len(train_dataset.y[train_dataset.y == 1]) / len(train_dataset.y),
             len(self.binary_dataset.y[self.binary_dataset.y == 1]) / len(self.binary_dataset.y),