Github user sethah commented on the pull request:
https://github.com/apache/spark/pull/12419#issuecomment-210570393
@psuszyns I have some high level comments. To me, it does not make sense to
train a PCA model, keeping k components, and then trim by variance explained.
If I have a model with 10 columns, and I train a PCA model with k = 6
components, I retain some fraction of the variance. Then I request to trim the
model by some fraction that might be _greater_ than the variance I originally
retained, so it will be impossible.
I think this should be implemented by having two parameters `k` and
`retainedVariance` where the full PCA is trained, and then the model is trimmed
by one of the two possible methods. When you set one of the params, you can
automatically unset the other since it doesn't make sense to use them both
(this is done, for example, in Logistic Regression with `threshold` and
`thresholds`. This would require changing ML _and_ MLlib, which is ok. Perhaps
@srowen could provide some thoughts.
---
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.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
