Consider different handling of negative input

[GoogleCodeExporter]

The original ALS formulation does not allow negative values. The current 
implementation does, but assigns them very low weight. This is better than 
negative weight, which is ill-formed, but not as principled as it could be.

While it's a corner case, and not intended to be used with negative input, it 
should be possible to modify the formulation to use *increasing* weight for 
more negative values, but penalize difference from 0 instead of 1. This would 
be more principled, and likely to give more intuitive results in the case that 
someone does want to use negative input.

Original issue reported on code.google.com by srowen@myrrix.com on 31 Oct 2012 at 12:48

[GoogleCodeExporter]

This is an easy change to make. A patch is attached. However, it seems to not 
improve, or slightly hurt, AUC / precision metrics on several test sets. I'm 
closing this idea for now as something that does not apparently help, for data 
sets with negative input, compared to the current formulation.

Original comment by srowen@myrrix.com on 22 Jan 2013 at 11:47

• Changed state: WontFix

Attachments:

• Issue_29.patch