Github user mengxr commented on the pull request:
https://github.com/apache/spark/pull/964#issuecomment-47881287
@vrilleup Both approaches compute the truncated SVD. I still prefer putting
both implementation under `computeSVD` for now. I'm going to implement a
generic ParameterSet (similar to Matlab's `options`). Once we have that, we can
add it to this method and users can choose which to use by setting `method` to
`dense`, `arpack`, or `auto`.
It is okay to use a simple rule to decide which approach to take. For
example, if `n < 100` or `k > n / 2` use dense, otherwise, use sparse. For the
dense approach, we can also use ARPACK instead of full SVD to compute the top k
singular values of the gram matrix. Just put a TODO to the condition and say
this is not optimized.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
