Opened 18 months ago
Closed 18 months ago
#2502 closed defect (done)
Calculation of confidence bounds for Gaussian process models seems incorrect
Reported by: | gkronber | Owned by: | gkronber |
---|---|---|---|
Priority: | medium | Milestone: | HeuristicLab 3.3.13 |
Component: | Algorithms.DataAnalysis | Version: | 3.3.12 |
Keywords: | Cc: |
Description
The confidence bounds shown in the line chart do not correspond with sigma noise in the model (see attachment).
Attachments (1)
Change History (7)
Changed 18 months ago by gkronber
comment:1 Changed 18 months ago by gkronber
r13121: calculate the variance for the noisy test data instead V(y*) instead of the variance for the posterior GP function V(f*)
comment:2 Changed 18 months ago by gkronber
- Owner set to gkronber
- Status changed from new to accepted
comment:3 Changed 18 months ago by gkronber
- Owner changed from gkronber to mkommend
- Status changed from accepted to reviewing
comment:4 Changed 18 months ago by mkommend
- Owner changed from mkommend to gkronber
- Status changed from reviewing to readytorelease
Reviewed r13121. OK.
The changes in the GP solution line chart are pretty clear. One thing that came to my mind while reviewing is that it should be possible to specify the confidence interval (default is set to 0.95).
The change in the GP-Model class is hard to judge without actually reading the paper (which I don't have at hand). I'll trust you that the noise variance has to be added to calculate the variance of the estimated values.
comment:5 Changed 18 months ago by gkronber
Rasmussen et al. "Gaussian Processes for Machine Learning" chapter 2, page 19: "The algorithm returns the predictive mean and variance for noise free test data - to compute the predictive distribution for noisy test data y∗, simply add the noise variance σ²n to the predictive variance of f∗." (http://www.gaussianprocess.org/gpml/chapters/RW2.pdf)
comment:6 Changed 18 months ago by gkronber
- Resolution set to done
- Status changed from readytorelease to closed
constant model (noise sigma = 0.65)