Datasets (`mtb.data`)¶

Download the datasets and precomputed outputs the tutorials use into mtb.config.DEFAULT.data_path, or into the data_path= you pass.

Name	Summary
`mtb.data.fetch`	Download reference datasets into the data root, skipping those present.
`mtb.data.fetch_outputs`	Download precomputed `run_all` outputs for a tutorial dataset.
`mtb.data.fetchable`	Dataset ids `mtb.data.fetch` can download.

mtb.data.fetch ¶

fetch(*datasets: str, data_path=None, quiet: bool = False) -> Path

Download reference datasets into the data root, skipping those present.

PARAMETERS	DESCRIPTION
`*datasets`	Dataset ids to fetch; `mtb.data.fetchable()` lists them. type `str` default `()`
`data_path`	Data root that holds the dataset folders; `None` = `mtb.config.DEFAULT.data_path`. Each dataset lands in `<data_path>/<dataset>/`. type `Path \| str \| None` default `None`
`quiet`	Suppress the one "downloading ..." line per dataset. type `bool` default `False`

RETURNS	DESCRIPTION
`Path`	The data root (not the dataset folder) - pass it as `data_path=`.

RAISES	DESCRIPTION
`ValueError`	A dataset id is not a release asset (the message lists the ids).
`RuntimeError`	The downloaded archive lacks the `<dataset>/` folder.
`OSError`	The download failed; the message names the URL.

Examples

>>> import multibench as mtb
>>> # the data root; D11 lands in <data_path>/D11/ (about 11 MB)
>>> root = mtb.data.fetch("D11")
>>> mtb.scan("D11", "vertical", data_path=root)
>>> mtb.data.fetch("D11", "D28", data_path="data", quiet=True)

Notes

What it covers. The tutorial reference sets published as release assets (mtb.data.fetchable()); the progress line states each one's approximate download size. The full collection is linked from the scMultiBench README ("Get the data" in the installation guide).

Idempotence. A non-empty <data_path>/<dataset>/ counts as present and is left alone, whatever it holds (even for an id that is not a release asset); an empty leftover folder is removed and fetched again.

Atomic extraction. Each archive is unpacked into a temporary folder under the data root and then moved into place. An archive entry that would write outside that folder raises RuntimeError.

mtb.data.fetch_outputs ¶

fetch_outputs(
    dataset: str, methods=None, *, data_path=None, quiet: bool = False
) -> Path

Download precomputed run_all outputs for a tutorial dataset.

Use it on a host without the method environments, such as Colab or a laptop. evaluate and plot then run on the stored outputs.

PARAMETERS	DESCRIPTION
`dataset`	Dataset id with shipped outputs: `D11`, `D28`, `D46` or `D52`. type `str`
`methods`	Method ids that must be in the tree; checked only, the whole tree is downloaded either way. type `list of str` default `None`
`data_path`	Data root that holds the dataset folders; `None` = `mtb.config.DEFAULT.data_path`. type `Path \| str \| None` default `None`
`quiet`	Suppress the one "downloading ..." line. type `bool` default `False`

RETURNS	DESCRIPTION
`Path`	`<data_path>/outputs/<dataset>` - pass it to `mtb.load_batch`.

RAISES	DESCRIPTION
`ValueError`	`dataset` has no shipped outputs (the message lists the ids).
`KeyError`	A name in `methods` is not in the tree (the message lists its methods).
`RuntimeError`	The archive lacks `batch_result.json`, or a non-empty `outputs/<dataset>/` lacks one.
`OSError`	The download failed; the message names the URL.

Examples

>>> import multibench as mtb
>>> out = mtb.data.fetch_outputs("D11")
>>> res = mtb.load_batch(out)
>>> res.summary[["method", "status", "ARI", "NMI"]]

Notes

Tree contents. Exactly what mtb.run_all writes and mtb.load_batch reads: batch_result.json, long.csv, summary.csv and one folder per method holding its embedding.h5.

Selecting methods. methods only checks the names; the whole tree is downloaded. To restrict what is loaded, use mtb.load_batch(out, methods=[...]).

Idempotence. When <data_path>/outputs/<dataset>/batch_result.json exists nothing is downloaded. An empty leftover folder is replaced. A non-empty one without batch_result.json raises RuntimeError. Remove it and call again.

Atomic extraction. The archive is unpacked into a temporary folder and then moved into place. The archive may be rooted at the tree itself or at one folder (outputs-D11/, D11/); an entry that would write outside the temporary folder raises RuntimeError.

mtb.data.fetchable ¶

fetchable() -> list[str]

Dataset ids mtb.data.fetch can download.

The companion of mtb.available_datasets, which lists the ids with stored results; the two sets overlap but are not the same.

RETURNS	DESCRIPTION
`list of str`	Dataset ids in natural order (`D11`, `D28`, `D45`, ...).

Examples

>>> import multibench as mtb
>>> mtb.data.fetchable()
['D11', 'D28', 'D45', 'D46', 'D52']

Notes

Overlap with stored results: D24 has published metric tables and no downloadable file, D46 downloads and has no stored table. The list is exactly the set of ids fetch() accepts.

Datasets (mtb.data)¶

mtb.data.fetch ¶

mtb.data.fetch_outputs ¶

mtb.data.fetchable ¶

Datasets (`mtb.data`)¶