Feature construction

When dealing with functional data we might want to construct new features that can be used as additional inputs to the machine learning algorithms. The expectation is that these features make explicit characteristics that facilitate the learning process.

Feature union

This transformer defines a way of extracting a high number of distinct features in parallel.

skfda.preprocessing.feature_construction.FDAFeatureUnion(...)

Concatenates results of multiple functional transformer objects.

Per class transformer

This method deals with the extraction of features using the information of the target classes It applies as many transformations as classes to every observation. Each transformation is fitted using only the training data of a particular class.

skfda.preprocessing.feature_construction.PerClassTransformer(...)

Per class feature transformer for functional data.

Functional features

The following functions can be used to create new features from functional data.

skfda.exploratory.stats.modified_epigraph_index(X)	Calculate the Modified Epigraph Index of a FDataGrid.
skfda.preprocessing.feature_construction.local_averages(...)	Calculate the local averages of given data in the desired domains.
skfda.preprocessing.feature_construction.occupation_measure(...)	Calculate the occupation measure of a grid.
skfda.preprocessing.feature_construction.number_crossings(fd, *)	Calculate the number of crossings to a level of a FDataGrid.

Some of them are also available as transformers that can be directly used in a pipeline:

skfda.preprocessing.feature_construction.LocalAveragesTransformer(*, ...)	Transforms functional data to its local averages.
skfda.preprocessing.feature_construction.OccupationMeasureTransformer(...)	Transformer that works as an adapter for the occupation_measure function.
skfda.preprocessing.feature_construction.NumberCrossingsTransformer(*)	Transformer that works as an adapter for the number_up_crossings function.