Opened 4 months ago

Closed 3 months ago

Last modified 3 months ago

#2755 closed enhancement (done)

For gradient boosted trees with LogisticRegressionLoss the threshold for the discriminant function should be set to zero.

Reported by: gkronber Owned by: mkommend
Priority: medium Milestone: HeuristicLab 3.3.15
Component: Algorithms.DataAnalysis Version: 3.3.14
Keywords: Cc:

Description

Currently, the 'accuracy maximizing' threshold calculator is used. However, this is rather slow for many data points and additionally it introduces a potential problem of overfitting. The pragmatic way is to set the threshold to zero.

Change History (10)

comment:1 Changed 4 months ago by gkronber

  • Owner set to gkronber
  • Status changed from new to accepted

comment:2 Changed 4 months ago by gkronber

r14779: set discriminant function threshold to zero for GBT with logistic regression loss

comment:3 Changed 4 months ago by gkronber

  • Owner changed from gkronber to mkommend
  • Status changed from accepted to reviewing

comment:4 Changed 4 months ago by gkronber

r14780: set dataset partitions correctly in GBT with logistic regression loss

comment:5 Changed 3 months ago by abeham

Please also remove unused usings, e.g. in GradientBoostedTreesModelSurrogate.cs.

comment:6 Changed 3 months ago by mkommend

r14955: Removed usings in GBT sources.

comment:7 Changed 3 months ago by mkommend

Reviewed r14779, r14780 and r14955.

comment:8 Changed 3 months ago by mkommend

  • Status changed from reviewing to readytorelease

r14956: Merged r14779, r14780 and r14955 into stable.

comment:9 Changed 3 months ago by mkommend

  • Resolution set to done
  • Status changed from readytorelease to closed

comment:10 Changed 3 months ago by mkommend

r14961: Corrected usings for GBT.

Note: See TracTickets for help on using tickets.