musicaiz.models.transformer_composers.build_torch_loaders

musicaiz.models.transformer_composers.build_torch_loaders(dataset_path: Union[str, pathlib.Path], sequence_length: int, batch_size: int, train_split: float = 0.9, is_splitted: bool = False) Tuple[torch.utils.data.dataloader.DataLoader, torch.utils.data.dataloader.DataLoader][source]

Builds the train a validation dataloaders.

Parameters
dataset_path: Path
sequence_length: int
batch_size: int
dest_path: Union[str, Path]

The destination path if save=True.

train_split: float. Between 0 and 1.

The training…

save: bool

If we want to save in disk the splitted tokens seqs.

is_splitted: bool.

Default is False. If the dataset is already splitted in train and validation (and test) sets, and there’s one token-sequences.txt file in each directory, it reads the token sequences and builds the loaders with them and it won’t split the files automatically.

Returns
train_dataloader: torch.utils.data.Dataloader

The train loader.

val_dataloader: torch.utils.data.Dataloader

The validation loader.