core.datasets.lmdb_dataset

Copyright (c) Meta, Inc. and its affiliates. This source code is licensed under the MIT license found in the LICENSE file in the root directory of this source tree.

Attributes

T_co

Classes

LmdbDataset

Base Dataset class for all OCP datasets.

Functions

data_list_collater(...)

Module Contents

core.datasets.lmdb_dataset.T_co

class core.datasets.lmdb_dataset.LmdbDataset(config)

Bases: fairchem.core.datasets.base_dataset.BaseDataset

Base Dataset class for all OCP datasets.

sharded: bool

Dataset class to load from LMDB files containing relaxation trajectories or single point computations. Useful for Structure to Energy & Force (S2EF), Initial State to Relaxed State (IS2RS), and Initial State to Relaxed Energy (IS2RE) tasks. The keys in the LMDB must be integers (stored as ascii objects) starting from 0 through the length of the LMDB. For historical reasons any key named “length” is ignored since that was used to infer length of many lmdbs in the same folder, but lmdb lengths are now calculated directly from the number of keys. :param config: Dataset configuration :type config: dict

path

key_mapping

transforms

__getitem__(idx: int) → T_co

connect_db(lmdb_path: pathlib.Path | None = None) → lmdb.Environment

__del__()

sample_property_metadata(num_samples: int = 100)

core.datasets.lmdb_dataset.data_list_collater(data_list: list[torch_geometric.data.data.BaseData], otf_graph: bool = False, to_dict: bool = False) → torch_geometric.data.data.BaseData | dict[str, torch.Tensor]