sc2_datasets.lightning.datamodules.sc2_replaypack_datamodule

Classes

SC2ReplaypackDataModule

Defines a LightningDataModule abstraction for a single StarCraft II replaypack.

Module Contents

class SC2ReplaypackDataModule(replaypack_name: str, unpack_dir: pathlib.Path | str = Path('./data/unpack').resolve(), download_dir: pathlib.Path | str = Path('./data/download').resolve(), url: str = '', download: bool = True, transform: Callable | None = None, batch_size: int = 256, num_workers: int = 0, unpack_n_workers: int = 16, validator: Callable | None = None)

Bases: pytorch_lightning.LightningDataModule

Defines a LightningDataModule abstraction for a single StarCraft II replaypack.

Parameters:

  • replaypack_name (str) – Specifies a replaypack name which will be used as a directory name.

  • unpack_dir (Path | str, optional) – Specifies the path where the replaypack (dataset) will be unpacked into a custom directory structure, by default “./data/unpack”

  • download_dir (Path | str, optional) – Specifies the path where the replaypack (dataset) will be downloaded, by default “./data/download”

  • url (str, optional) – Specifies the url which will be used to download the replaypack (dataset), by default “”

  • download (bool, optional) – Specifies if the dataset should be downloaded. Otherwise the dataset is loaded from the unpack_dir and a custom directory structure is assumed, by default True

  • transform (Callable | None, optional) – Specifies the PyTorch transforms to be used on the replaypack (dataset), Deprecated since version v1.5: Will be removed in v1.7.0, by default None

  • batch_size (int, optional) – Batch size which will be used for learning tasks, by default 256

  • num_workers (int, optional) – Number of workers used for dataloaders, by default 0

  • unpack_n_workers (int, optional) – Specifies the number of workers that will be used for unpacking the archive, by default 16

  • validator (Callable | None, optional) – Specifies the validation option for fetched data, this can also act as a filtering function that will be applied for the entirety of the dataset, by default None

transform = None
batch_size = 256
num_workers = 0
replaypack_name
unpack_dir
download_dir
url = ''
download = True
unpack_n_workers = 16
validator = None
prepare_data() → None
setup(stage: str | None = None) → None
train_dataloader() → torch.utils.data.dataloader.DataLoader
val_dataloader() → torch.utils.data.dataloader.DataLoader
test_dataloader() → torch.utils.data.dataloader.DataLoader
teardown(stage: str | None = None) → None