Opened 3 weeks ago

Last modified 4 days ago

#1204 reopened defect

5.0 not giving same fit results as 4.2

Reported by: richardh Owned by: richardh
Priority: critical Milestone: SasView 5.0.0
Component: SasView Keywords:
Cc: Work Package: SasView QA and testing

Description

Load e.g. AOT_microemulsion-shell_contrast.xml
Fit e.g. just scale & length with cylinder model, taking care that all params are at their default values (this may involve switching model category and re-selecting cylinder to get them back if you have done other things first).

4.2 and 5.0 do not give the same fit results for a "poor fit" such as this. They should do if using the same fit engine, and same fit options.
I suspect an initialisation or default for weighting, resolution, fit engine or something else is not working properly but I have not yet managed to find the culprit.

For "good" fits the parameter values and their errors are very close but still not identical as they perhaps ought to be.

Attachments (5)

coil1.png (222.5 KB) - added by richardh 2 weeks ago.
starting parameters
coil2.png (219.8 KB) - added by richardh 2 weeks ago.
turn on 2 params
coil3.png (224.3 KB) - added by richardh 2 weeks ago.
more params fitting
coil4_iQwt.png (226.4 KB) - added by richardh 2 weeks ago.
change from di weights to I(Q) weights
coil5.png (229.1 KB) - added by richardh 2 weeks ago.
same results with default dI weighting IF start at same place

Download all attachments as: .zip

Change History (10)

comment:1 Changed 3 weeks ago by richardh

PS there are times when 4.2 and 5.0 do give the same results, so there is something intermittent here, I am beginning to suspect the "set weighting" flag in 4.2.

comment:2 Changed 3 weeks ago by richardh

  • Resolution set to invalid
  • Status changed from new to closed

The set weighting choice in 4.2 seems to have no effect, whereas it does in 5.0, I am going to close this ticket for now write a new one for 4.2

Changed 2 weeks ago by richardh

starting parameters

Changed 2 weeks ago by richardh

turn on 2 params

Changed 2 weeks ago by richardh

more params fitting

Changed 2 weeks ago by richardh

change from di weights to I(Q) weights

comment:3 Changed 2 weeks ago by richardh

  • Resolution invalid deleted
  • Status changed from closed to reopened

Some more direct comparisons between 4.2 and 5.0, see screen shots of same fits in each giving different answers, 5.0 on left, 4.2 on right.
These for AOT_microemulsion shell contrast, fit to mono_gauss_coil

The initial start up model looks the same, but the residuals plot is rather different!
None of the fit results are identical, and are very different when change to I(Q) weights.

Suspect something is perhaps not being initialised properly, or passed properly???

This time I was in the beta_approx branch, not master, of sasmodels, but I don't think that mattters.

(Apologies I should not have turned on both scale and I_zero here as they are redundant in this model, however the conclusions are the same).

Last edited 2 weeks ago by richardh (previous) (diff)

Changed 2 weeks ago by richardh

same results with default dI weighting IF start at same place

comment:4 Changed 2 weeks ago by richardh

In default dI weighting can get 5.0 and 4.2 to give same answer, as per latest attached screen shot, IF start at the same place, but this is not the case in the other weighting schemes.

comment:5 Changed 4 days ago by richardh

Looks like there was a long standing issue in 4.x (and possibly even 3.x), see #1205
Now need to check weighting scheme effect in 5.0 and 4.2 side by side again

Note: See TracTickets for help on using tickets.