| Age | Commit message (Collapse) | Author | 
 | 
 | 
 | 
 | 
 | 
Division by zero was the issue.
 | 
 | 
 | 
 | 
The parameters are now all in [0,1].
 | 
 | 
 | 
 | 
This again makes our sample selection less prone to extreme outliers.
Now only closeness has a quadratic penalty.
 | 
 | 
It can still do so, but it is much more unlikely and generally more
stable now.
 | 
 | 
 | 
 | 
 | 
 | 
The slight adaption was the one deva and I talked about on IRC.
 | 
 | 
 | 
 | 
 | 
 | 
 | 
 | 
 | 
 | 
 | 
 | 
 | 
 | 
 | 
 | 
 | 
 | 
 |