CODEWITHSUNDEEP
python-2.7scikit-learnpca
slaw
slaw

Reputation: 6869

Truncated SVD vs Partial SVD

Can somebody tell me the difference between truncated SVD as implemented in sklearn and partial SVD as implemented in, say, fbpca?

I couldn't find a definitive answer as I haven't seen anybody use truncated SVD for principal component pursuit (PCP).

Upvotes: 1

Views: 1642

Answers (1)

eickenberg
eickenberg

Reputation: 14377

Truncated or partial means that you only calculate a certain number of components/singular vector-value pairs (the strongest ones).

In scikit-learn parlance, "partial" usually refers to the fact that a method is on line, meaning that it can be fed with partial data. The more data you give it the better it will converge to the expected optimum.

Both can be combined, and have been, also in sklearn: sklearn.decomposition.IncrementalPCA does this.

Upvotes: 2

Related Questions