Migrate to pydantic 2.0+

devtools/conda-envs/mtenn.yaml

@@ -3,14 +3,16 @@ channels:
   - conda-forge
 dependencies:
   - pytorch
-  - pytorch_geometric
+  - pytorch_geometric >=2.5.0
   - pytorch_cluster
   - pytorch_scatter
   - pytorch_sparse
+  - pydantic >=2.0.0a0
   - numpy
   - h5py
   - e3nn
   - dgllife
   - dgl
   - rdkit
   - ase
+  - fsspec

devtools/conda-envs/test_env.yaml

@@ -7,6 +7,7 @@ dependencies:
   - pytorch_cluster
   - pytorch_scatter
   - pytorch_sparse
+  - pydantic >=2.0.0a0
   - numpy
   - h5py
   - e3nn
@@ -19,5 +20,4 @@ dependencies:
   - pytest
   - pytest-cov
   - codecov
-  - pydantic >=1.10.8,<2.0.0a0
   - fsspec

mtenn/config.py

@@ -15,9 +15,9 @@
 
 import abc
 from enum import Enum
-from pydantic import BaseModel, Field, root_validator
+from pydantic import model_validator, ConfigDict, BaseModel, Field
 import random
-from typing import Callable, ClassVar
+from typing import Literal, Callable, ClassVar
 import mtenn.combination
 import mtenn.readout
 import mtenn.model
@@ -140,7 +140,8 @@ class ModelConfigBase(BaseModel):
     to implement the ``_build`` method in order to be used.
     """
 
-    model_type: ModelType = Field(ModelType.INVALID, const=True, allow_mutation=False)
+    model_type: Literal[ModelType.INVALID] = ModelType.INVALID
+
 
     # Random seed optional for reproducibility
     rand_seed: int | None = Field(
@@ -240,9 +241,7 @@ class ModelConfigBase(BaseModel):
             "``comb_substrate``."
         ),
     )
-
-    class Config:
-        validate_assignment = True
+    model_config = ConfigDict(validate_assignment=True)
 
     def build(self) -> mtenn.model.Model:
         """
@@ -394,7 +393,7 @@ def _check_grouped(values):
         Makes sure that a Combination method is passed if using a GroupedModel. Only
         needs to be called for structure-based models.
         """
-        if values["grouped"] and (not values["combination"]):
+        if values.grouped and not values.combination:
             raise ValueError("combination must be specified for a GroupedModel.")
 
 
@@ -436,7 +435,7 @@ class GATModelConfig(ModelConfigBase):
         "biases": bool,
     }  #: :meta private:
 
-    model_type: ModelType = Field(ModelType.GAT, const=True)
+    model_type: Literal[ModelType.GAT] = ModelType.GAT
 
     in_feats: int = Field(
         _CanonicalAtomFeaturizer().feat_size(),
@@ -527,14 +526,16 @@ class GATModelConfig(ModelConfigBase):
     #  num_layers
     _from_num_layers = False
 
-    @root_validator(pre=False)
-    def massage_into_lists(cls, values) -> GATModelConfig:
+    @model_validator(mode="after")
+    def massage_into_lists(self) -> GATModelConfig:
         """
         Validator to handle unifying all the values into the proper list forms based on
         the rules described in the class docstring.
         """
+        values = self.dict()
+
         # First convert string lists to actual lists
-        for param, param_type in cls.LIST_PARAMS.items():
+        for param, param_type in self.LIST_PARAMS.items():
             param_val = values[param]
             if isinstance(param_val, str):
                 try:
@@ -548,7 +549,7 @@ def massage_into_lists(cls, values) -> GATModelConfig:
 
         # Get sizes of all lists
         list_lens = {}
-        for p in cls.LIST_PARAMS:
+        for p in self.LIST_PARAMS:
             param_val = values[p]
             if not isinstance(param_val, list):
                 # Shouldn't be possible at this point but just in case
@@ -577,14 +578,16 @@ def massage_into_lists(cls, values) -> GATModelConfig:
         # If we just want a model with one layer, can return early since we've already
         #  converted everything into lists
         if num_layers == 1:
-            return values
+            # update self with the new values
+            self.__dict__.update(values)
+
 
         # Adjust any length 1 list to be the right length
         for p, list_len in list_lens.items():
             if list_len == 1:
                 values[p] = values[p] * num_layers
 
-        return values
+        return self.__dict__.update(values)
 
     def _build(self, mtenn_params={}):
         """
@@ -681,7 +684,7 @@ class SchNetModelConfig(ModelConfigBase):
     given in PyG.
     """
 
-    model_type: ModelType = Field(ModelType.schnet, const=True)
+    model_type: Literal[ModelType.schnet] = ModelType.schnet
 
     hidden_channels: int = Field(128, description="Hidden embedding size.")
     num_filters: int = Field(
@@ -738,13 +741,14 @@ class SchNetModelConfig(ModelConfigBase):
         ),
     )
 
-    @root_validator(pre=False)
+    @model_validator(mode="after")
+    @classmethod
     def validate(cls, values):
         # Make sure the grouped stuff is properly assigned
         ModelConfigBase._check_grouped(values)
 
         # Make sure atomref length is correct (this is required by PyG)
-        atomref = values["atomref"]
+        atomref = values.atomref
         if (atomref is not None) and (len(atomref) != 100):
             raise ValueError(f"atomref must be length 100 (got {len(atomref)})")
 
@@ -816,7 +820,7 @@ class E3NNModelConfig(ModelConfigBase):
     Class for constructing an e3nn ML model.
     """
 
-    model_type: ModelType = Field(ModelType.e3nn, const=True)
+    model_type: Literal[ModelType.e3nn] = ModelType.e3nn
 
     num_atom_types: int = Field(
         100,
@@ -862,7 +866,8 @@ class E3NNModelConfig(ModelConfigBase):
     num_neighbors: float = Field(25, description="Typical number of neighbor nodes.")
     num_nodes: float = Field(4700, description="Typical number of nodes in a graph.")
 
-    @root_validator(pre=False)
+    @model_validator(mode="after")
+    @classmethod
     def massage_irreps(cls, values):
         """
         Check that the value given for ``irreps_hidden`` can be converted into an Irreps
@@ -874,7 +879,7 @@ def massage_irreps(cls, values):
         ModelConfigBase._check_grouped(values)
 
         # Now deal with irreps
-        irreps = values["irreps_hidden"]
+        irreps = values.irreps_hidden
         # First see if this string should be converted into a dict
         if isinstance(irreps, str):
             if ":" in irreps:
@@ -923,7 +928,7 @@ def massage_irreps(cls, values):
         except ValueError:
             raise ValueError(f"Couldn't parse irreps dict: {orig_irreps}")
 
-        values["irreps_hidden"] = irreps
+        values.irreps_hidden = irreps
         return values
 
     def _build(self, mtenn_params={}):
@@ -994,7 +999,7 @@ class ViSNetModelConfig(ModelConfigBase):
     given in PyG.
     """
 
-    model_type: ModelType = Field(ModelType.visnet, const=True)
+    model_type: Literal[ModelType.visnet] = ModelType.visnet
     lmax: int = Field(1, description="The maximum degree of the spherical harmonics.")
     vecnorm_type: str | None = Field(
         None, description="The type of normalization to apply to the vectors."
@@ -1041,7 +1046,8 @@ class ViSNetModelConfig(ModelConfigBase):
         ),
     )
 
-    @root_validator(pre=False)
+    @model_validator(mode="after")
+    @classmethod
     def validate(cls, values):
         """
         Check that ``atomref`` and ``max_z`` agree.
@@ -1050,10 +1056,10 @@ def validate(cls, values):
         ModelConfigBase._check_grouped(values)
 
         # Make sure atomref length is correct (this is required by PyG)
-        atomref = values["atomref"]
-        if (atomref is not None) and (len(atomref) != values["max_z"]):
+        atomref = values.atomref
+        if (atomref is not None) and (len(atomref) != values.max_z):
             raise ValueError(
-                f"atomref length must match max_z. (Expected {values['max_z']}, got {len(atomref)})"
+                f"atomref length must match max_z. (Expected {values.max_z}, got {len(atomref)})"
             )
 
         return values