Initialization of matrix factorization (stnmf.init)

Collection of functions that offer strategies for initialization of matrix factorization.

Initialization largely contributes to convergence time and the solution that is found [1] [2].

nnsvdlrc(v, r[, lrc, dtype])	Initialize feature matrix with a variation of non-negative singular value decomposition with low-rank correction (NNSVD-LRC)
random(v, r[, rng, dtype])	Initialize feature matrix by random sampling

Notes

The functions provided can be called directly by the matrix factorization classes by passing their name as the argument w0.

Examples

The code below is equivalent.

>>> w = stnmf.init.nnsvdlrc(v, r=20)
>>> stnmf.mf.SparseSemiNMF(v, r=20, w0=w)

>>> stnmf.mf.SparseSemiNMF(v, r=20, w0='nnsvdlrc')

References

[1]

Cichocki, A., Zdunek, R., Phan, A. H., & Amari, S. (2009). Nonnegative Matrix and Tensor Factorizations (1st ed.). Chichester, UK: John Wiley & Sons, Ltd. https://doi.org/10.1002/9780470747278

[2]

Atif, S. M., Qazi, S., & Gillis, N. (2019). Improved SVD-based initialization for nonnegative matrix factorization using low-rank correction. Pattern Recognition Letters, 122, 53–59. https://doi.org/10.1016/J.PATREC.2019.02.018