Datasets¶

ur.datasets has small example graphs. Four are bundled in the wheel and load offline. One is downloaded on first use.

Loaders¶

Every loader returns an EdgeFrame with the columns src, dst, and (if weighted) weight.

Function	Nodes	Edges	Weighted	Node labels	Description
`ur.datasets.load_karate(*, with_nodes=False)`	34	78	no	`club`	Zachary's karate club.
`ur.datasets.load_lesmis()`	77	254	yes		Les Misérables co-appearance. The weight is the number of shared chapters.
`ur.datasets.load_florentine()`	15	20	no		Florentine families marriage ties.
`ur.datasets.load_kite()`	10	18	no		Krackhardt's kite.
`ur.datasets.load_facebook()`	4039	88234	no		SNAP ego-Facebook. Downloaded on first use.

load_karate(with_nodes=True) returns a tuple (edges, nodes). nodes is a NodeFrame with the columns id and club.

Each bundled graph is undirected in its source. The loader stores each edge once, as one row. Pass direction="both" to degree, hop, and shortest_path for the undirected result.

`ur.datasets.load`¶

ur.datasets.load(name: str, **kwargs)

Loads a dataset by name. kwargs go to the loader, for example load("karate", with_nodes=True).

`ur.datasets.list_datasets`¶

ur.datasets.list_datasets() -> list[DatasetInfo]

The registry. Each DatasetInfo is a named tuple with the fields name, nodes, edges, bundled, weighted, node_labels, description, source, and license.

The download cache¶

load_facebook() downloads one file from SNAP on the first call and verifies its checksum. The file is cached in $URSA_DATA_HOME, or in ~/.cache/ursa/datasets if the variable is not set. Later calls read the cache and need no network. If the download fails, the loader raises an error that names the URL and the cache path.

Licenses¶

The bundled graphs are redistributed from NetworkX (BSD license) as plain CSV files. The downloaded graph is subject to the SNAP terms. The file LICENSES.md in the installed package lists the original sources.