stratifier

stratifier

Stratification of datasets using GAD.

Classes

Name	Description
Clusterer	Protocol for sklearn clusterer objects.
Stratifier	Stratifier class for determining stratifications of data.

Clusterer

stratifier.Clusterer()

Protocol for sklearn clusterer objects.

Methods

Name	Description
fit_predict	Fit the clusterer and predict cluster labels.

fit_predict

stratifier.Clusterer.fit_predict(X)

Fit the clusterer and predict cluster labels.

Stratifier

stratifier.Stratifier(clusterer, max_dim=1, n_jobs=1)

Stratifier class for determining stratifications of data.

This class relies on the GAD algorithm implemented in dimmer to construct stratifications.

Parameters

Name	Type	Description	Default
clusterer	Clusterer	A scikit-learn style clusterer such as DBSCAN.	required
max_dim	int	The maximal dimension to run GAD.	1
n_jobs	int	The number of processors to use.	1

Note

This class is experimental and has not yet been fully tested. It is written in the style of a scikit-learn transformer.

Methods

Name	Description
fit_predict	Compute stratification labels across all dimensions.
fit_predict_dw	Compute stratification labels in a specified dimension.

fit_predict

stratifier.Stratifier.fit_predict(X, ids, max_dim, y=None)

Compute stratification labels across all dimensions.

This method iteratively applies fit_predict_dw from max_dim down to dimension 1, then clusters all remaining points at dimension <= 1.

Parameters

Name	Type	Description	Default
X	np.ndarray[tuple[int, int], np.dtype[np.float64]]	Array containing n_samples points within ambient Euclidean space of dimension n_features.	required
ids	np.ndarray[tuple[int], np.dtype[np.float64]]	Dimension estimates for each point in X.	required
max_dim	int	Maximum dimension up to which stratification is computed.	required
y	np.ndarray[tuple[int], np.dtype[np.float64]] | None	Ignored. (Default value = None)	None

Returns: Cluster labels

fit_predict_dw

stratifier.Stratifier.fit_predict_dw(X, ids, dim, cluster_labels=None, y=None)

Compute stratification labels in a specified dimension.

This model takes dimension estimates as input in ids (e.g., the output of fit_predict and computes a stratification within a single dimension specified by dim using self.clusterer.

Parameters

Name	Type	Description	Default
X	np.ndarray[tuple[int, int], np.dtype[np.float64]]	Array containing n_samples points within ambient Euclidean space of dimension n_features.	required
ids	np.ndarray[tuple[int], np.dtype[np.float64]]	Dimension estimates for each point in X.	required
dim	int	Dimension in which to compute stratification.	required
cluster_labels	np.ndarray[tuple[int], np.dtype[np.float64]] | None	Already computed labels, usually from a previous degree.	None
y	np.ndarray | None	Ignored.	None

Returns

Name	Type	Description
	np.ndarray	Cluster labels

Functions

Name	Description
construct_region_labels	Construct the Hasse diagram for a stratification.

construct_region_labels

stratifier.construct_region_labels(X, ids, grid_sep=1.0)

Construct the Hasse diagram for a stratification.

This method labels each top-dimensional region with a unique integer, the region id. Points which are not classified as top dimensional are then assigned a label by looking at the closest two top-dimensional points. If both of those points have the same region id, that id is assigned to the point in question. If they have a different region id, both region ids are assigned to the point in question.

Parameters

Name	Type	Description	Default
X	np.ndarray[tuple[int, int], np.dtype[np.float64]]	np.ndarray:	required
ids	np.ndarray[tuple[int], np.dtype[np.float64]]	np.ndarray:	required
grid_sep	float	float: (Default value = 1.0)	1.0

Returns

Name	Type	Description
	dict[int, list[int]]	A dictionary of region ids.

Note

This method is specifically for gridded data. It can certainly be modified for other projects, but this algorithm assumes the data lies on a grid.