kosmos.ml.dataloader¶
Functions¶
- make_train_test_dataloaders(dataset: kosmos.ml.datasets.dataset.SLDataset, train_split: float, batch_size: int, num_train_subsets: int = 1) tuple[list[torch.utils.data.DataLoader], torch.utils.data.DataLoader]¶
Split a dataset into training and test subsets and wrap them in DataLoaders.
The training loaders shuffle their subset each epoch; the test loader does not.
- Parameters:
dataset – The dataset to be split.
train_split – Fraction of dataset for training. Test split is 1 - train_split.
batch_size – Number of samples per batch in both loaders.
num_train_subsets – Number of partitions to split the training subset into. Defaults to 1.
- Returns:
A tuple containing:
list[DataLoader]: DataLoaders for the partitions of the training subset.
DataLoader: DataLoader for the test subset.
- Return type: