vyntri.data

`split(dataset, *, train=0.7, val=None, test=0.2, seed=42)`

Create an explicit train/val/test split from a folder-per-class dataset. Returns a SplitResult.

Parameter	Type	Description
dataset	str	Path to a folder-per-class dataset
train	float	Training fraction (default 0.7)
val	float \| None	Validation fraction. When None, computed as 1 - train - test.
test	float	Test fraction (default 0.2)
seed	int	Deterministic seed (default 42)

Returns: SplitResult

Raises: SplitError if the dataset already has an explicit layout.

`SplitResult`

Reusable result of split(). Contains logical FolderDataset subsets (no file copy).

Field	Type	Description
train	FolderDataset	Training subset
val	FolderDataset \| None	Validation subset (None for two-way split)
test	FolderDataset \| None	Test subset (None if test=0)
seed	int	Seed used for the split
fractions	dict	Fractions per split
counts	dict	Sample counts per split
manifest	dict	Full provenance record
class_names	list[str]	Sorted class names

Example

from vyntri.data import split
from vyntri import Vyntri

s = split("./my_dataset", train=0.7, test=0.2, seed=42)
model = Vyntri()
model.fit(train=s.train, val=s.val)
result = model.evaluate(s.test)
print(f"Accuracy: {result.accuracy:.1%}")

`FolderDataset.from_ids(root, sample_ids)`

Create a logical subset by sample IDs. No physical file copy.