Source code for metatrain.utils.per_atom

from typing import Dict, List

import torch
from metatensor.torch import TensorBlock, TensorMap
from metatomic.torch import System


[docs] def average_by_num_atoms( tensor_map_dict: Dict[str, TensorMap], systems: List[System], per_structure_keys: List[str], ) -> Dict[str, TensorMap]: """ Averages a dictionary of ``TensorMap`` objects by the number of atoms in each system. This function averages by the number of atoms in each system. Targets that are present in ``per_structure_keys`` will not be averaged. :param tensor_map_dict: A dictionary of ``TensorMap`` objects. :param systems: The systems used to compute the predictions. :param per_structure_keys: A list of keys whose corresponding ``TensorMap`` objects that should not be averaged. :return: The dictionary of averaged ``TensorMap`` objects. """ averaged_tensor_map_dict = {} device = systems[0].device num_atoms = torch.tensor([len(s) for s in systems], device=device) for key in tensor_map_dict.keys(): if key in per_structure_keys: averaged_tensor_map_dict[key] = tensor_map_dict[key] else: averaged_tensor_map_dict[key] = divide_by_num_atoms( tensor_map_dict[key], num_atoms ) return averaged_tensor_map_dict
[docs] def divide_by_num_atoms(tensor_map: TensorMap, num_atoms: torch.Tensor) -> TensorMap: """Takes the average values per atom of a ``TensorMap``. Since some quantities might already be per atom (e.g., atomic energies or position gradients), this function only divides a block (or gradient block) by the number of atoms if the block's samples do not contain the "atom" key. In practice, this guarantees the desired behavior for the majority of the cases, including energies, forces, and virials, where the energies and virials should be divided by the number of atoms, while the forces should not. :param tensor_map: The input tensor map. :param num_atoms: The number of atoms in each system. :return: A new tensor map with the values divided by the number of atoms. """ blocks = [] for block in tensor_map.blocks(): if "atom" in block.samples.names: new_block = block else: values = block.values / num_atoms.view( -1, *[1] * (len(block.values.shape) - 1) ) new_block = TensorBlock( values=values, samples=block.samples, components=block.components, properties=block.properties, ) for gradient_name, gradient in block.gradients(): if "atom" in gradient.samples.names: new_gradient = gradient else: values = gradient.values / num_atoms.view( -1, *[1] * (len(gradient.values.shape) - 1) ) new_gradient = TensorBlock( values=values, samples=gradient.samples, components=gradient.components, properties=gradient.properties, ) new_block.add_gradient(gradient_name, new_gradient) blocks.append(new_block) return TensorMap( keys=tensor_map.keys, blocks=blocks, )