Fix: moltochemprop not compatible with chemprop 2.0.3 (#37)

adapt to chemprop 2.0.3

molpipeline/estimators/chemprop/featurizer_wrapper/__init__.py

@@ -0,0 +1 @@
+"""Wrapper for Chemprop Featurizer."""

molpipeline/estimators/chemprop/featurizer_wrapper/graph_wrapper.py

@@ -0,0 +1,52 @@
+"""Wrapper for Chemprop GraphFeaturizer."""
+
+from dataclasses import InitVar
+from typing import Any
+
+try:
+    from typing import Self  # type: ignore[attr-defined]
+except ImportError:
+    from typing_extensions import Self
+
+from chemprop.featurizers.molgraph import (
+    SimpleMoleculeMolGraphFeaturizer as _SimpleMoleculeMolGraphFeaturizer,
+)
+
+
+class SimpleMoleculeMolGraphFeaturizer(_SimpleMoleculeMolGraphFeaturizer):
+    """Wrapper for Chemprop SimpleMoleculeMolGraphFeaturizer."""
+
+    extra_atom_fdim: InitVar[int]
+    extra_bond_fdim: InitVar[int]
+
+    def get_params(
+        self, deep: bool = True  # pylint: disable=unused-argument
+    ) -> dict[str, InitVar[int]]:
+        """Get parameters for the featurizer.
+
+        Parameters
+        ----------
+        deep: bool, optional (default=True)
+            Used for compatibility with scikit-learn.
+
+        Returns
+        -------
+        dict[str, int]
+            Parameters of the featurizer.
+        """
+        return {}
+
+    def set_params(self, **parameters: Any) -> Self:  # pylint: disable=unused-argument
+        """Set the parameters of the featurizer.
+
+        Parameters
+        ----------
+        parameters: Any
+            Parameters to set. Only used for compatibility with scikit-learn.
+
+        Returns
+        -------
+        Self
+            This featurizer with the parameters set.
+        """
+        return self

molpipeline/mol2any/mol2chemprop.py

@@ -5,17 +5,25 @@
 from __future__ import annotations
 
 import warnings
-from typing import Iterable, Optional
+from typing import Any, Iterable, Optional
+
+import numpy as np
+import numpy.typing as npt
 
 try:
     from chemprop.data import MoleculeDatapoint, MoleculeDataset
-    from chemprop.featurizers.molecule import MoleculeFeaturizer
+    from chemprop.featurizers.base import GraphFeaturizer, VectorFeaturizer
+
+    from molpipeline.estimators.chemprop.featurizer_wrapper.graph_wrapper import (
+        SimpleMoleculeMolGraphFeaturizer,
+    )
 except ImportError:
     warnings.warn(
         "chemprop not installed. MolToChemprop will not work.",
         ImportWarning,
     )
 
+
 from molpipeline.abstract_pipeline_elements.core import MolToAnyPipelineElement
 from molpipeline.utils.molpipeline_types import RDKitMol
 
@@ -31,11 +39,13 @@ class MolToChemprop(MolToAnyPipelineElement):
     [1] https://github.com/chemprop/chemprop/
     """
 
-    featurizer_list: list[MoleculeFeaturizer] | None
+    graph_featurizer: GraphFeaturizer[RDKitMol] | None
+    mol_featurizer: VectorFeaturizer[RDKitMol] | None
 
     def __init__(
         self,
-        featurizer_list: list[MoleculeFeaturizer] | None = None,
+        graph_featurizer: GraphFeaturizer[RDKitMol] | None = None,
+        mol_featurizer: VectorFeaturizer[RDKitMol] | None = None,
         name: str = "Mol2Chemprop",
         n_jobs: int = 1,
         uuid: Optional[str] = None,
@@ -44,16 +54,21 @@ def __init__(
 
         Parameters
         ----------
-        featurizer_list: list[MoleculeFeaturizer] | None, optional (default=None)
-            List of featurizers to use.
+        graph_featurizer: GraphFeaturizer[RDKitMol] | None, optional (default=None)
+            Defines how the graph is featurized. Defaults to None.
+        mol_featurizer: MoleculeFeaturizer | None, optional (default=None)
+            In contrast to graph_featurizer, features from the mol_featurizer are not used during the message passing.
+            These features are concatenated to the neural fingerprints before the feedforward layers.
         name: str, optional (default="Mol2Chemprop")
             Name of the pipeline element. Defaults to "Mol2Chemprop".
         n_jobs: int
             Number of parallel jobs to use. Defaults to 1.
         uuid: str | None, optional (default=None)
             UUID of the pipeline element.
         """
-        self.featurizer_list = featurizer_list
+        self.graph_featurizer = graph_featurizer or SimpleMoleculeMolGraphFeaturizer()
+        self.mol_featurizer = mol_featurizer
+
         super().__init__(
             name=name,
             n_jobs=n_jobs,
@@ -73,7 +88,10 @@ def pretransform_single(self, value: RDKitMol) -> MoleculeDatapoint:
         MoleculeDatapoint
             Molecular representation used as input for ChemProp. None if transformation failed.
         """
-        return MoleculeDatapoint(mol=value, mfs=self.featurizer_list)
+        mol_features: npt.NDArray[np.float64] | None = None
+        if self.mol_featurizer is not None:
+            mol_features = np.array(self.mol_featurizer(value), dtype=np.float64)
+        return MoleculeDatapoint(mol=value, x_d=mol_features)
 
     def assemble_output(
         self, value_list: Iterable[MoleculeDatapoint]
@@ -90,4 +108,70 @@ def assemble_output(
         Any
             Assembled output.
         """
-        return MoleculeDataset(data=list(value_list))
+        return MoleculeDataset(data=list(value_list), featurizer=self.graph_featurizer)
+
+    def get_params(self, deep: bool = True) -> dict[str, Any]:
+        """Get parameters for this pipeline element.
+
+        Parameters
+        ----------
+        deep: bool, optional (default=True)
+            If True, will return the parameters for this pipeline element and its subobjects.
+
+        Returns
+        -------
+        dict
+            Parameters of this pipeline element.
+        """
+        params = super().get_params(deep=deep)
+        params["graph_featurizer"] = self.graph_featurizer
+        params["mol_featurizer"] = self.mol_featurizer
+
+        if deep:
+            if hasattr(self.graph_featurizer, "get_params"):
+                graph_featurizer_params = self.graph_featurizer.get_params(deep=deep)  # type: ignore
+                for key, value in graph_featurizer_params.items():
+                    params[f"graph_featurizer__{key}"] = value
+            if hasattr(self.mol_featurizer, "get_params"):
+                mol_featurizer_params = self.mol_featurizer.get_params(deep=deep)  # type: ignore
+                for key, value in mol_featurizer_params.items():
+                    params[f"mol_featurizer__{key}"] = value
+        return params
+
+    def set_params(self, **parameters: Any) -> MolToChemprop:
+        """Set the parameters of this pipeline element.
+
+        Parameters
+        ----------
+        **parameters: Any
+            Parameters to set.
+
+        Returns
+        -------
+        MolToChemprop
+            This pipeline element with the parameters set.
+        """
+        param_copy = dict(parameters)
+        graph_featurizer = param_copy.pop("graph_featurizer", None)
+        mol_featurizer = param_copy.pop("mol_featurizer", None)
+        if graph_featurizer is not None:
+            self.graph_featurizer = graph_featurizer
+        if mol_featurizer is not None:
+            self.mol_featurizer = mol_featurizer
+        graph_featurizer_params = {}
+        mol_featurizer_params = {}
+        for key in list(param_copy.keys()):
+            if "__" not in key:
+                continue
+            component_name, _, param_name = key.partition("__")
+            if component_name == "graph_featurizer":
+                graph_featurizer_params[param_name] = param_copy.pop(key)
+            elif component_name == "mol_featurizer":
+                mol_featurizer_params[param_name] = param_copy.pop(key)
+        if hasattr(self.graph_featurizer, "set_params"):
+            self.graph_featurizer.set_params(**graph_featurizer_params)  # type: ignore
+        if hasattr(self.mol_featurizer, "set_params"):
+            self.mol_featurizer.set_params(**mol_featurizer_params)  # type: ignore
+
+        super().set_params(**param_copy)
+        return self

requirements_chemprop.txt

@@ -1,2 +1,2 @@
-chemprop >= 2.0.0, < 2.0.3
+chemprop>=2.0.3
 lightning

test_extras/test_chemprop/test_chemprop_pipeline.py

@@ -7,15 +7,12 @@
 import joblib
 import numpy as np
 import pandas as pd
-from chemprop.nn.loss import LossFunction
 from lightning import pytorch as pl
-from lightning.pytorch.accelerators import Accelerator
-from lightning.pytorch.profilers.base import PassThroughProfiler
 from sklearn.base import clone
-from torch import nn
 
 from molpipeline.any2mol import SmilesToMol
 from molpipeline.error_handling import ErrorFilter, FilterReinserter
+from molpipeline.estimators.chemprop.abstract import ABCChemprop
 from molpipeline.estimators.chemprop.component_wrapper import (
     MPNN,
     BinaryClassificationFFN,
@@ -30,6 +27,7 @@
 from molpipeline.mol2any.mol2chemprop import MolToChemprop
 from molpipeline.pipeline import Pipeline
 from molpipeline.post_prediction import PostPredictionWrapper
+from test_extras.test_chemprop.chemprop_test_utils.compare_models import compare_params
 
 
 # pylint: disable=duplicate-code
@@ -199,6 +197,9 @@ def test_clone(self) -> None:
             self.assertEqual(step.__class__, cloned_step.__class__)
             params = step.get_params(deep=True)  # type: ignore
             cloned_params = cloned_step.get_params(deep=True)
+            if isinstance(step, ABCChemprop):
+                compare_params(self, step, cloned_step)
+                continue
             for param_name, param in params.items():
                 # If parm implements get_params, it was cloned as well and we need
                 # to compare the parameters. Since all parameters are listed flat in
@@ -207,27 +208,6 @@ def test_clone(self) -> None:
                     self.assertEqual(
                         param.__class__, cloned_params[param_name].__class__
                     )
-                elif param_name == "lightning_trainer":
-                    # Lightning trainer does not implement get_params so things are a bit tricky
-                    # at the moment. We can only check if the classes are the same.
-                    self.assertEqual(
-                        param.__class__, cloned_params[param_name].__class__
-                    )
-                elif isinstance(param, LossFunction):
-                    self.assertEqual(
-                        param.state_dict()["task_weights"],
-                        cloned_params[param_name].state_dict()["task_weights"],
-                    )
-                    self.assertEqual(type(param), type(cloned_params[param_name]))
-                elif isinstance(param, (nn.Identity, Accelerator, PassThroughProfiler)):
-                    self.assertEqual(type(param), type(cloned_params[param_name]))
-                elif param_name == "lightning_trainer__callbacks":
-                    self.assertIsInstance(cloned_params[param_name], list)
-                    self.assertEqual(len(param), len(cloned_params[param_name]))
-                    for callback, cloned_callback in zip(
-                        param, cloned_params[param_name]
-                    ):
-                        self.assertEqual(type(callback), type(cloned_callback))
                 else:
                     self.assertEqual(
                         param, cloned_params[param_name], f"Failed for {param_name}"

test_extras/test_chemprop/test_models.py

@@ -22,9 +22,12 @@
 from molpipeline.utils.json_operations import recursive_from_json, recursive_to_json
 
 # pylint: disable=relative-beyond-top-level
-from .chemprop_test_utils.compare_models import compare_params
-from .chemprop_test_utils.constant_vars import DEFAULT_PARAMS, NO_IDENTITY_CHECK
-from .chemprop_test_utils.default_models import (
+from test_extras.test_chemprop.chemprop_test_utils.compare_models import compare_params
+from test_extras.test_chemprop.chemprop_test_utils.constant_vars import (
+    DEFAULT_PARAMS,
+    NO_IDENTITY_CHECK,
+)
+from test_extras.test_chemprop.chemprop_test_utils.default_models import (
     get_chemprop_model_binary_classification_mpnn,
 )
 

test_extras/test_chemprop/test_neural_fingerprint.py

@@ -9,8 +9,10 @@
 from molpipeline.utils.json_operations import recursive_from_json, recursive_to_json
 
 # pylint: disable=relative-beyond-top-level
-from .chemprop_test_utils.compare_models import compare_params
-from .chemprop_test_utils.default_models import get_neural_fp_encoder
+from test_extras.test_chemprop.chemprop_test_utils.compare_models import compare_params
+from test_extras.test_chemprop.chemprop_test_utils.default_models import (
+    get_neural_fp_encoder,
+)
 
 logging.getLogger("lightning.pytorch.utilities.rank_zero").setLevel(logging.WARNING)