feat pt : Support property fitting (#3867)

Solve issue #3866

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit


- **New Features**
- Introduced property fitting neural networks with the new
`PropertyFittingNet` class.
- Added the `DeepProperty` class for evaluating properties of structures
using a deep learning model.
- Implemented the `PropertyModel` class to integrate properties specific
to atomic models.

- **Enhancements**
- Added an `intensive` property to several classes to indicate whether a
fitting property is intensive or extensive.
- Enhanced output property definitions and manipulation methods in
various models for improved property evaluation.
- Expanded loss function capabilities to handle the "property" loss
type.
- Improved argument definitions for fitting and loss functionalities,
enhancing configurability.
- Updated the model selection logic to include the new `PropertyModel`.
- Enhanced the `DeepEvalWrapper` class to support additional model
evaluation features, including new methods for retrieving model
characteristics.

- **Documentation**
- Updated class docstrings to reflect new attributes and parameters,
improving clarity and usability.

- **Tests**
- Expanded the set of test examples related to the "property" category
to improve test coverage.
- Introduced new test classes and parameterized tests for improved
validation of property-related functionalities.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Signed-off-by: Chenqqian Zhang <100290172+Chengqian-Zhang@users.noreply.github.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

deepmd/dpmodel/atomic_model/property_atomic_model.py

@@ -0,0 +1,14 @@
+# SPDX-License-Identifier: LGPL-3.0-or-later
+from deepmd.dpmodel.fitting.property_fitting import (
+    PropertyFittingNet,
+)
+
+from .dp_atomic_model import (
+    DPAtomicModel,
+)
+
+
+class DPPropertyAtomicModel(DPAtomicModel):
+    def __init__(self, descriptor, fitting, type_map, **kwargs):
+        assert isinstance(fitting, PropertyFittingNet)
+        super().__init__(descriptor, fitting, type_map, **kwargs)

deepmd/dpmodel/fitting/__init__.py

@@ -17,6 +17,9 @@
 from .polarizability_fitting import (
     PolarFitting,
 )
+from .property_fitting import (
+    PropertyFittingNet,
+)
 
 __all__ = [
     "InvarFitting",
@@ -25,4 +28,5 @@
     "EnergyFittingNet",
     "PolarFitting",
     "DOSFittingNet",
+    "PropertyFittingNet",
 ]

deepmd/dpmodel/fitting/property_fitting.py

@@ -0,0 +1,136 @@
+# SPDX-License-Identifier: LGPL-3.0-or-later
+import copy
+from typing import (
+    List,
+    Optional,
+    Union,
+)
+
+import numpy as np
+
+from deepmd.dpmodel.common import (
+    DEFAULT_PRECISION,
+)
+from deepmd.dpmodel.fitting.invar_fitting import (
+    InvarFitting,
+)
+from deepmd.utils.version import (
+    check_version_compatibility,
+)
+
+
+@InvarFitting.register("property")
+class PropertyFittingNet(InvarFitting):
+    r"""Fitting the rotationally invariant porperties of `task_dim` of the system.
+
+    Parameters
+    ----------
+    ntypes
+            The number of atom types.
+    dim_descrpt
+            The dimension of the input descriptor.
+    task_dim
+            The dimension of outputs of fitting net.
+    neuron
+            Number of neurons :math:`N` in each hidden layer of the fitting net
+    bias_atom_p
+            Average property per atom for each element.
+    rcond
+            The condition number for the regression of atomic energy.
+    trainable
+            If the weights of fitting net are trainable.
+            Suppose that we have :math:`N_l` hidden layers in the fitting net,
+            this list is of length :math:`N_l + 1`, specifying if the hidden layers and the output layer are trainable.
+    intensive
+            Whether the fitting property is intensive.
+    bias_method
+            The method of applying the bias to each atomic output, user can select 'normal' or 'no_bias'.
+            If 'normal' is used, the computed bias will be added to the atomic output.
+            If 'no_bias' is used, no bias will be added to the atomic output.
+    resnet_dt
+            Time-step `dt` in the resnet construction:
+            :math:`y = x + dt * \phi (Wx + b)`
+    numb_fparam
+            Number of frame parameter
+    numb_aparam
+            Number of atomic parameter
+    activation_function
+            The activation function :math:`\boldsymbol{\phi}` in the embedding net. Supported options are |ACTIVATION_FN|
+    precision
+            The precision of the embedding net parameters. Supported options are |PRECISION|
+    mixed_types
+            If false, different atomic types uses different fitting net, otherwise different atom types share the same fitting net.
+    exclude_types: List[int]
+            Atomic contributions of the excluded atom types are set zero.
+    type_map: List[str], Optional
+            A list of strings. Give the name to each type of atoms.
+    """
+
+    def __init__(
+        self,
+        ntypes: int,
+        dim_descrpt: int,
+        task_dim: int = 1,
+        neuron: List[int] = [128, 128, 128],
+        bias_atom_p: Optional[np.ndarray] = None,
+        rcond: Optional[float] = None,
+        trainable: Union[bool, List[bool]] = True,
+        intensive: bool = False,
+        bias_method: str = "normal",
+        resnet_dt: bool = True,
+        numb_fparam: int = 0,
+        numb_aparam: int = 0,
+        activation_function: str = "tanh",
+        precision: str = DEFAULT_PRECISION,
+        mixed_types: bool = True,
+        exclude_types: List[int] = [],
+        type_map: Optional[List[str]] = None,
+        # not used
+        seed: Optional[int] = None,
+    ):
+        self.task_dim = task_dim
+        self.intensive = intensive
+        self.bias_method = bias_method
+        super().__init__(
+            var_name="property",
+            ntypes=ntypes,
+            dim_descrpt=dim_descrpt,
+            dim_out=task_dim,
+            neuron=neuron,
+            bias_atom=bias_atom_p,
+            resnet_dt=resnet_dt,
+            numb_fparam=numb_fparam,
+            numb_aparam=numb_aparam,
+            rcond=rcond,
+            trainable=trainable,
+            activation_function=activation_function,
+            precision=precision,
+            mixed_types=mixed_types,
+            exclude_types=exclude_types,
+            type_map=type_map,
+        )
+
+    @classmethod
+    def deserialize(cls, data: dict) -> "PropertyFittingNet":
+        data = copy.deepcopy(data)
+        check_version_compatibility(data.pop("@version"), 2, 1)
+        data.pop("dim_out")
+        data.pop("var_name")
+        data.pop("tot_ener_zero")
+        data.pop("layer_name")
+        data.pop("use_aparam_as_mask", None)
+        data.pop("spin", None)
+        data.pop("atom_ener", None)
+        obj = super().deserialize(data)
+
+        return obj
+
+    def serialize(self) -> dict:
+        """Serialize the fitting to dict."""
+        dd = {
+            **InvarFitting.serialize(self),
+            "type": "property",
+            "task_dim": self.task_dim,
+        }
+
+        return dd

deepmd/dpmodel/model/__init__.py

@@ -21,12 +21,16 @@
 from .make_model import (
     make_model,
 )
+from .property_model import (
+    PropertyModel,
+)
 from .spin_model import (
     SpinModel,
 )
 
 __all__ = [
     "EnergyModel",
+    "PropertyModel",
     "DPModelCommon",
     "SpinModel",
     "make_model",

deepmd/dpmodel/model/property_model.py

@@ -0,0 +1,27 @@
+# SPDX-License-Identifier: LGPL-3.0-or-later
+from deepmd.dpmodel.atomic_model.dp_atomic_model import (
+    DPAtomicModel,
+)
+from deepmd.dpmodel.model.base_model import (
+    BaseModel,
+)
+
+from .dp_model import (
+    DPModelCommon,
+)
+from .make_model import (
+    make_model,
+)
+
+DPPropertyModel_ = make_model(DPAtomicModel)
+
+
+@BaseModel.register("property")
+class PropertyModel(DPModelCommon, DPPropertyModel_):
+    def __init__(
+        self,
+        *args,
+        **kwargs,
+    ):
+        DPModelCommon.__init__(self)
+        DPPropertyModel_.__init__(self, *args, **kwargs)

deepmd/dpmodel/output_def.py

@@ -186,6 +186,8 @@ class OutputVariableDef:
           If hessian is requred
     magnetic : bool
           If the derivatives of variable have magnetic parts.
+    intensive : bool
+          It indicates whether the fitting property is intensive or extensive.
     """
 
     def __init__(
@@ -199,6 +201,7 @@ def __init__(
         category: int = OutputVariableCategory.OUT.value,
         r_hessian: bool = False,
         magnetic: bool = False,
+        intensive: bool = False,
     ):
         self.name = name
         self.shape = list(shape)
@@ -211,13 +214,17 @@ def __init__(
         self.reducible = reducible
         self.r_differentiable = r_differentiable
         self.c_differentiable = c_differentiable
+        self.intensive = intensive
         if self.c_differentiable and not self.r_differentiable:
             raise ValueError("c differentiable requires r_differentiable")
         if self.reducible and not self.atomic:
             raise ValueError("a reducible variable should be atomic")
+        if self.intensive and not self.reducible:
+            raise ValueError("an intensive variable should be reducible")
         self.category = category
         self.r_hessian = r_hessian
         self.magnetic = magnetic
+        self.intensive = intensive
         if self.r_hessian:
             if not self.reducible:
                 raise ValueError("only reducible variable can calculate hessian")