Details
-
Type:
Story
-
Status: Done
-
Resolution: Done
-
Fix Version/s: None
-
Component/s: jointcal
-
Labels:
-
Story Points:8
-
Epic Link:
-
Sprint:AP S18-6, AP F18-1
-
Team:Alert Production
Description
Pierre Astier suggests that we should implement a line search in jointcal, to deal with the non-linearities in the photometric model (and because we may eventually have those in astrometry, as the astrometry model gets more complex). He suggested using Brent's method, which is available in GSL. It might be easier to just take the code directly from GSL, instead of writing an interface layer, given the way the nature of the jointcal models.
At the top of the outlier rejection loop in FitterBase.minimize(), this would look something like:
Eigen::VectorXd delta = chol.solve(grad);
|
scale = lineSearch(delta);
|
offsetParams(delta*scale));
|
Attachments
Issue Links
- has to be done before
-
- Done
Activity
Pim Schellart [X] do you mind taking on this review (less than 100 lines changed)?
Sorry for the delay. See suggestions on PR.
Pim Schellart [X] (Inactive)
added a comment - Sorry for the delay. See suggestions on PR. Post-cleanup and rebase jenkins run: https://ci.lsst.codes/blue/organizations/jenkins/stack-os-matrix/detail/stack-os-matrix/28153/pipeline
Thanks for the review. I dealt with, or responded to, all of your comments.
Merged and done.
If you take the code from GSL please ensure that a full license statement is made in the relevant file and README. It's important that we record when we integrate external GPL code. We can change the license of our own code, we can't change the license if we include external GPL code.