tfep.io.dataset.traj

Utility classes to create PyTorch ``Dataset``s from MDAnalysis trajectories.

The module provides a class TrajectoryDataset that wraps an MDAnalysis Universe object (i.e., an object tying a topology and a trajectory) and implements PyTorch’s Dataset interface. This can be used, for example, to specify the training dataset for the neural network implementing the mapping function of targeted free energy perturbation.

The TrajectoryDataset can be subsampled at constant time interval, while arbitrary subsets of the trajectory can be instead be created using the TrajectorySubset class.

For usage examples see the documentation of - TrajectoryDataset - TrajectorySubset

Functions

get_subsampled_indices(dt, stop[, start, ...])

Subsamples the trajectory at a constant time interval after discarding an initial equilibration.

Classes

TrajectoryDataset(universe[, ...])

PyTorch Dataset wrapping an MDAnalysis trajectory.

TrajectorySubset(dataset, indices)

A subset of a TrajectoryDataset.