data.random_split¶

lucid.data.random_split(dataset: Dataset, lengths: Sequence[int | float], seed: int | None = None) → tuple[Subset, ...]¶

The random_split function divides a dataset into non-overlapping subsets of specified lengths or fractional proportions. This function is often used to create train/validation/test splits in data pipelines.

Function Signature¶

def random_split(
    dataset: Dataset,
    lengths: Sequence[int | float],
    seed: int | None = None
) -> tuple[Subset, ...]

Parameters¶

dataset (lucid.data.Dataset): The dataset to split. Must implement __len__ and __getitem__.
lengths (Sequence[int | float]): The lengths (or fractions) of each split.
- If integers, they must sum to len(dataset).
- If floats, they are treated as fractions that must sum to 1.0.
seed (int, optional): A random seed for reproducibility. If None, it uses global random seed from lucid.random package.

Returns¶

tuple[Subset, …]: A tuple of lucid.data.Subset instances, each wrapping the original dataset and exposing only its respective samples.

Mathematical Explanation¶

\[ \begin{align}\begin{aligned}\text{indices} = \text{Shuffle}(\{0, 1, ..., N-1\})\\\text{subset}_k = \{ \text{dataset}[i] \mid i \in \text{indices}_{k} \}\end{aligned}\end{align} \]

where:

\(N\) is the total number of samples in the dataset.
\(\text{indices}_k\) represents the segment of indices assigned to the k-th subset.

Examples¶

Splitting by integer lengths:

>>> from lucid.data import random_split
>>> data = list(range(10))
>>> train_ds, val_ds = random_split(data, [8, 2], seed=42)
>>> len(train_ds), len(val_ds)
(8, 2)

Splitting by fractional proportions:

>>> from lucid.data import random_split
>>> data = list(range(10))
>>> train_ds, val_ds = random_split(data, [0.8, 0.2], seed=1)
>>> len(train_ds), len(val_ds)
(8, 2)

Tip

You can safely use each split with lucid.data.DataLoader without additional wrapping or conversions.