Add a fit algorithm that does not require a good initial guess
Bug #781752 reported by
Tom Dimiduk
This bug affects 1 person
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
HoloPy |
In Progress
|
Wishlist
|
Unassigned |
Bug Description
The levenberg marquardt fitting algorithm we currently use will only converge if given a very good initial guess. These have to be determined manually and are one of the more labor intensive parts of setting up a fit. It would be nice if we could start without this information and have a possibly staged fitter that would converge anyway.
Changed in holopy: | |
milestone: | none → future |
To post a comment you must log in.
I am working on this-- starting with a hologram of a single particle. My approach is to write a simple simulated annealing algorithm. This should be less than 200 lines of code, so I am writing my own rather than using a boxed simulated annealing algorithm. The goal is to be able to throw a hologram at it and have it return x, y, z, radius, and particle index. The user most likely will have to specify bounds of reasonable values (e.g. x is between 0 and 40 microns, y is between 0 and 40 microns, index between 1.3 and 1.7, z is between 10 and 40 microns, and the radius is between 400 and 600 nm). I hope to make the fit robust enough that the user can enter very generous bounds that would be applicable to a whole series of frames rather than frame-specific guesses.