Transformer

ElbowClassSum

Elbow Class Sum (ECS) transformer to select a subset of channels/variables.

Overview: From the input of multivariate time series data, create a distance matrix [1] by calculating the distance between each class centroid. The ECS selects the subset of channels using the elbow method, which maximizes the distance between the class centroids by aggregating the distance for every class pair across each channel.

Note: Channels, variables, dimensions, features are used interchangeably in literature. E.g., channel selection = variable selection.

Schnellstart

python

from sktime.transformations.channel_selection import ElbowClassSum

estimator = ElbowClassSum(distance=None)

Parameter(1)

distance: sktime pairwise panel transform, str, or callable, optional, default=None: if panel transform, will be used directly as the distance in the algorithm default None = euclidean distance on flattened series, FlatDist(ScipyDist()) if str, will behave as FlatDist(ScipyDist(distance)) = scipy dist on flat series if callable, must be univariate nested_univ x nested_univ -> 2D float np.array

Beispiele

>>> from sktime.transformations.channel_selection import ElbowClassSum
>>> from sktime.utils._testing.panel import make_classification_problem
>>> X, y = make_classification_problem (n_columns = 3, n_classes = 3, random_state = 42)
>>> cs = ElbowClassSum ()
>>> cs. fit (X, y) ElbowClassSum(
... )
>>> Xt = cs. transform (X) Any sktime compatible distance can be used, e.g., DTW distance:
>>> from sktime.dists_kernels import DtwDist
>>> 
>>> cs = ElbowClassSum (distance = DtwDist ())
>>> cs. fit (X, y) ElbowClassSum(
... )
>>> Xt = cs. transform (X)

Referenzen

..[1]: Bhaskar Dhariyal et al. “Fast Channel Selection for Scalable Multivariate Time Series Classification.” AALTD, ECML-PKDD, Springer, 2021