API Reference

`pydpc.Cluster(points, fraction=0.02, autoplot=True, **kwargs)`

Bases: Graph

Density-peak clustering of a set of points.

Constructing a Cluster computes the pairwise distances, local density, and decision-graph quantities (density, delta) for points. Cluster centers are then chosen interactively by inspecting the decision graph and calling :meth:assign with density/delta thresholds that isolate the outlying points (the candidate centers).

Examples:

>>> clu = Cluster(points)
>>> clu.assign(min_density=20, min_delta=1.5)
>>> clu.membership  # cluster index for each point

Parameters:

Name	Type	Description	Default
`points`	`ndarray of shape (n_points, n_dim)`	Coordinates of the data points to cluster.	required
`fraction`	`float`	Average fraction of all other points to treat as neighbours when estimating the density kernel size.	`0.02`
`autoplot`	`bool`	If True, automatically draw the decision graph on construction and again (with the chosen thresholds indicated) whenever :meth:`assign` is called.	`True`
`**kwargs`		Additional keyword arguments forwarded to :class:`Graph`, e.g. a precomputed `distances` matrix or an explicit `kernel_size`.	`{}`

Attributes:

Name	Type	Description
`density`	`ndarray of shape (n_points,)`	Local density of each data point.
`delta`	`ndarray of shape (n_points,)`	Minimal distance of each point to a point of higher density.
`clusters`	`ndarray of int`	Indices of the chosen cluster centers. Set by :meth:`assign`.
`membership`	`ndarray of shape (n_points,)`	Cluster index assigned to each point. Set by :meth:`assign`.
`halo_idx`	`ndarray of int`	Indices of points classified as halo (low-confidence) members. Set by :meth:`assign`.
`core_idx`	`ndarray of int`	Indices of points classified as core (high-confidence) members. Set by :meth:`assign`.

`assign(min_density, min_delta, border_only=False)`

Choose cluster centers via decision-graph thresholds and assign points.

Points with density above min_density and delta above min_delta are taken as cluster centers; every other point is then assigned to the cluster of its nearest denser neighbour.

Parameters:

Name	Type	Description	Default
`min_density`	`float`	Minimum density for a point to be considered a cluster center.	required
`min_delta`	`float`	Minimum delta for a point to be considered a cluster center.	required
`border_only`	`bool`	If False (default), classify as halo any point whose density is below its cluster's border density (the original Rodriguez/Laio criterion). If True, only classify points as halo where they border a different cluster, which is less strict and keeps more points in the core.	`False`

Returns:

Type	Description
`None`	Results are stored on the instance; see the `clusters`, `membership`, `border_density`, `border_member`, `halo_idx`, and `core_idx` attributes.

Raises:

Type	Description
`ValueError`	If no point satisfies both thresholds, i.e. no cluster center is found.

`draw_decision_graph(min_density=None, min_delta=None)`

Plot the decision graph (density vs. delta) for all points.

Parameters:

Name	Type	Description	Default
`min_density`	`float`	If given together with `min_delta`, draw red threshold lines marking the region density > `min_density` and delta > `min_delta`.	`None`
`min_delta`	`float`	See `min_density`.	`None`

Returns:

Name	Type	Description
`fig`	`Figure`
`ax`	`Axes`