CellProfiler-Analyst icon indicating copy to clipboard operation
CellProfiler-Analyst copied to clipboard

Add switch to properties file to exclude zero variance features

Open holgerhennig opened this issue 9 years ago • 7 comments

"exclude_zero_variance = yes" (set default value to yes)

holgerhennig avatar Jun 14 '16 18:06 holgerhennig

Does this interfere with any functionality (e.g. classification) or is it just a nuisance? @holgerhennig Closing for now, might add Future consideration label if warranted based on your reply.

AnneCarpenter avatar Sep 05 '16 12:09 AnneCarpenter

@holgerhennig You can analyze all zero variance features using Advanced->Check features. Then either drop those features manually in the properties file or use the normalization tool to delete them.

screen shot 2016-09-06 at 11 57 51

daviddao avatar Sep 06 '16 10:09 daviddao

Thanks @daviddao, that's very helpful! The normalization tool sounds like a good option then to delete zero variance features. Dropping the zero variance features manually in the prop file might be a good option for some users too, but can be more cumbersome. Suggest to add your description above and @AnneCarpenter's decription for the normalization tool in issue #198 to the CPA manual, something like "normalization tool" and "how to delete zero variance features".

holgerhennig avatar Sep 06 '16 10:09 holgerhennig

@holgerhennig Could you please write that documentation? @N3llz could add it to the CPA manual website once ready.

AnneCarpenter avatar Sep 06 '16 20:09 AnneCarpenter

sure, will do. I'll let you and Jeanelle @N3llz know when it's ready

holgerhennig avatar Sep 07 '16 07:09 holgerhennig

@AnneCarpenter and @N3llz: Done :) I've forked the CPA manual, updated the CPA manual and filed a pull request. I added section on data preparation (normalization of features, removing zero variance features). Pull request is here: https://github.com/daviddao/CPA-docs/pull/1 Let me know if there's anything else I can do

holgerhennig avatar Oct 25 '16 14:10 holgerhennig

Awesome thanks so much! The current maintained version is inside CPA itself (https://github.com/CellProfiler/CellProfiler-Analyst/tree/master/docs) , please forget CPA-docs. It's depricated, sry for this! I will just import your change into CPA if you don't mind :)

daviddao avatar Oct 25 '16 17:10 daviddao