Title

start out with Data Module.

Fande Data Module

Model 1

class fande.data.FandeDataModuleASE(*args: Any, **kwargs: Any)

calculate_invariants_librascal(soap_params=None, atomic_groups=None, centers_positions=None, derivatives_positions=None, same_centers_derivatives=False, frames_per_batch=10, calculation_context=None, trajectory=None)

Calculate SOAP invariants using librascal. If trajectory is not specified, the invariants are calculated for the training or test set of frames (self.traj_train or self.traj_test).

Parameters:

soap_params: dict

Dictionary containing the parameters for the SOAP descriptors

atomic_groups: list

List of atomic groups to train separate GP models for

centers_positions: list

List of positions of the atomic centers

derivatives_positions: list

List of indices of the atoms to calculate the derivatives with respect to. On these atoms the forces are evaluated.

same_centers_derivatives: bool

If True, the same atomic centers are used for the derivatives as for the descriptors.

frames_per_batch: int

Number of frames per batch when calculating invariants. Extremely useful when memory is limited.

calculation_context: str, [‘train’, ‘test’, ‘production’]

Whether to calculate the invariants for the training/test set of frames (stored within`self.traj_train` or self.traj_test) or for on-the-fly md runs.

Returns:

X: torch.tensor

Tensor containing the SOAP descriptors for the set of frames split by atomic groups specified in atomic_groups.

DX: list(torch.tensor)

list of torch tensors containing the derivatives of the SOAP descriptors for each atomic group specified in atomic_groups.

only for calculation_context=’production’: F: torch.tensor

Tensor containing the forces for the set of frames split by atomic groups specified in atomic_groups.

calculate_test_invariants_librascal(test_centers_positions=None, test_derivatives_positions=None, same_centers_derivatives=False)

Deprecated! Use calculate_invariants_librascal instead.

prepare_train_data_loaders(total_samples_per_group, high_force_samples_per_group)

Prepare data loaders for training. Currently data loaders are created within the FandeDataModule class and then passed to the group model classes.

Parameters

samples_per_grouplist of int

Number of training random samples to use for each group. If None, all samples are used.

high_force_samples_per_grouplist of int

Number of high force samples to be included for each group. high_force_samples_per_group of the tighest (positive) and lowest (negative) forces are selected.

Returns

train_data_loaderslist of torch.utils.data.DataLoader

List of data loaders for training for each atomic group model.