pub fn random_split<D>(
dataset: Arc<D>,
lengths: &[SplitLength],
seed: u64,
) -> Result<Vec<Subset<Arc<D>>>>where
D: Dataset,Expand description
Randomly partitions a shared dataset into non-overlapping subsets.
Count lengths must sum exactly to the dataset length. Fractional lengths are floored, then leftover samples are assigned round-robin from the first split. The seeded permutation is reproducible within this implementation.
ยงErrors
Returns RustTorchError::InvalidConfiguration for mixed length kinds,
invalid fractions, incorrect totals, or count overflow.