How to know the variability explained by individual PCA (widely_svd)

[aalsharef]

Hello,

Thanks for the great package! It is not clear to me how to select the number of PCA inside the function "widely_svd". Can I know the variability explained by individual PCA (i.e., selecting the optimal nv) ? This would justify selecting the number of PCAs. For now, I'm setting it to 100 (nv = 100) following the suggestion in "Supervised Machine Learning for Text Analysis in R" by Emil Hvitfeldt and Julia Silge.

Thank you very much!

[juliasilge]

Thanks for your patience with this issue! 🙌

In its current implementation, we only return the u matrix from the SVD:

widyr/R/widely_svd.R

Lines 77 to 89 in a6696d6

	perform_svd <- function(m) {
	s <- irlba::irlba(m, nv = nv, ...)
	if (weight_d) {
	ret <- t(s$d * t(s$u))
	} else {
	ret <- s$u
	}
	rownames(ret) <- rownames(m)
	ret
	}
	sparse <- TRUE
	}

Let's think through how we might return other, more complete info from the SVD in the tidy format we use in this package. In the meantime, I would recommend that you use a lower-level interface to SVD like irlba so you can get out all the information you want.