model.NeuralNetworkPotential

class model.NeuralNetworkPotential(*args: Any, **kwargs: Any)

A generic neural network potential class that sequentially applies a list of input modules, a representation module and a list of output modules.

This can be flexibly configured for various, e.g. property prediction or potential energy sufaces with response properties.

Parameters:

• representation – The module that builds representation from inputs.

• input_modules – Modules that are applied before representation, e.g. to modify input or add additional tensors for response properties.

• output_modules – Modules that predict output properties from the representation.

• postprocessors – Post-processing transforms that may be initialized using the datamodule, but are not applied during training.

• input_dtype_str – The dtype of real inputs.

• do_postprocessing – If true, post-processing is activated.