Using the uncertainties module to propagate errors #1916
Conversation
wpotrzebowski
added
Discuss At The Call
src/sas/sascalc/fit/BumpsFitting.py
Outdated
| reference_params = [v for v in varying if str(v.name) in str(constraints[param_name])] | ||
| err_exp = str(constraints[param_name]) | ||
| # Convert string entries into variable names within the code. | ||
| for i in enumerate(reference_params): |
There was a problem hiding this comment.
this looks a bit awkward, maybe use something like:
for i, p in enumerate(reference_params):
err_exp = err_exp.replace(p.name, f"reference_params[{i}].value")
There was a problem hiding this comment.
Tested functionality on W10/x64 comparing against release 5.0.4.
Tried single fit (using LM) and 3 dataset simultaneous fit (using DREAM).
Behaviour seems 'normal'. Uncertainties on optimised parameters look sensible. Residuals look sensible. DREAM plots looked sensible. DREAM output in Log Explorer looked sensible and parameter values were correctly assigned to the correct parameters (something that didn't happen in PR1682).
Parameter/uncertainty values matched v5.0.4 to many dp when using LM. Correspondence using DREAM was obviously not as good but still acceptable (IMHO); eg:
In 5.0.4:
scale,True,0.068696,0.00058734,0.0,inf,():
background,True,-0.057794,0.061883,-inf,inf,():
radius,True,23.969,0.10756,0.0,inf,():
thickness,True,9.0774,0.06878,0.0,inf,():
In this branch:
scale,True,0.068957,0.00060114,0.0,inf,():
background,True,-0.058371,0.0679,-inf,inf,():
radius,True,23.915,0.11556,0.0,inf,():
thickness,True,9.0682,0.089137,0.0,inf,():
So I approve this PR.
butlerpd
added
the
Discuss At The Call
There was a problem hiding this comment.
It seems that some of @rozyczko requests still need to answered? I've also added few questions
|
@Caddy-Jones as already responded to @rozyczko comments and will look at the latest comments. Iestyn to merge once addressed. |
wpotrzebowski
removed
the
Discuss At The Call
Renamed M as fitting_module and R as fitting_result
As discussed in issue #1914, the pull request from #1682 lead to fitting error in SasView. This pull request attempts to accomplish the same goal of propagating errors on constrained parameters by using code inspired by #1682, while avoiding the fitting issues that lead to #1914.