peal.sparse_dictionaries.singular_value_decomposition

Truncated SVD (PCA) as the simplest PEAL sparse dictionary.

The principal directions of a feature matrix serve as a dense, non-sparse baseline dictionary that adaptors such as DiDAE can search for class-relevant directions in the same way they search an SAE or MSAE dictionary.

Classes

SVDDictionary

Dictionary whose components are the top right-singular vectors of the data.

SVDDictionaryConfig

Config for SVDDictionary.