Ticket #7914 (assigned)
Implement Poisson cost function for Mantid Fit
Reported by: | Mike Thomas | Owned by: | Anders Markvardsen |
---|---|---|---|
Priority: | major | Milestone: | Backlog |
Component: | Framework | Keywords: | |
Cc: | Blocked By: | ||
Blocking: | Tester: |
Description
Investigation done for ticket #7383 shows a need for a poisson cost function when dealing with low counts. The attached paper 'Effecieint Levenberg-Marquardt Minimization of the Maxiumum Likelihood Estimator for Poisson Deviates' describes the poisson cost function and it's derivatives.
Attachments
Change History
comment:3 Changed 7 years ago by Anders Markvardsen
- Milestone changed from Release 3.0 to Backlog
Note somehow when this user select this cost function there should be a warning presented to the user if the data can be spotted not to be with the y-axis as counts. Without having looked at this in detail, I believe, this cost function is only valid for this case. For example if you are fitting the difference between two data then using this cost function the user would not want to use this cost function since as far as I can think now the weight on the data point would potentially be complement - hence users should be warned about this