SuperInstance

Each COBRAS variant has its own super-instance type. The specific super-instance decides how the representative of a super-instance is chosen and provides a way to calculate the distance between two super-instances (used for querying the closest super-instances first while merging)

class cobras_ts.superinstance.SuperInstance(data, indices, train_indices, parent=None)

A class representing a super-instance used in the COBRAS algorithm

__init__(data, indices, train_indices, parent=None)

Parameters

• data – the full dataset

• indices – the indices of the instances that are in this super-instance

• train_indices – all training indices (i.e. training indices of the full dataset)

• parent – the parent super-instance (if any)

abstract distance_to(other_superinstance)

Calculates the distance to the given super-instance This is COBRAS variant specific

Parameters

other_superinstance – the super-instance to calculate the distance to

Returns

the distance between this super-instance and the given other_superinstance

SuperInstance_DTW

class cobras_ts.superinstance_dtw.SuperInstance_DTW(data, indices, train_indices, parent=None)

__init__(data, indices, train_indices, parent=None)

Chooses the super-instance representative based on the affinity matrix under the data argument

distance_to(other_superinstance)

Uses the negative entry in the affinity matrix between the two super-instance representatives

note: this is not a correct distance measure but this is not necessary for COBRAS

SuperInstance_kmeans

class cobras_ts.superinstance_kmeans.SuperInstance_kmeans(data, indices, train_indices, parent=None)

__init__(data, indices, train_indices, parent=None)

Chooses the super-instance representative as the instance closest to the super-instance centroid

distance_to(other_superinstance)

The distance between two super-instances is equal to the distance between there centroids

SuperInstance_kShape

class cobras_ts.superinstance_kshape.SuperInstance_kShape(data, indices, train_indices, sbd_centroid=None, parent=None)

__init__(data, indices, train_indices, sbd_centroid=None, parent=None)

If sbd_centroid is given the instance that has the smallest distance to the sbd_centroid is chosen as representative Otherwise the representative index is simply the first instance in this super-instance

distance_to(other_superinstance)

The distance between two super-instances is the distance between their sbd centroids