Version 13 (modified by butler, 21 months ago) (diff)

  • Novice users should not "need to know hidden features", e.g. for saving their fits, changing fit methods etc
  • "fit pages" have compute & fit buttons, could also have "save fit" and "report" buttons, even "save model".
  • "save fit" could put up a list of "fit, P(Q), S(Q), unsmeared, etc" to choose which to save. [FISH does this]
  • Propose that all form factor models have both "rescale" and "volfraction" as parameters, where rescale defaults to 1.0 (correct for neutron SANS data in absolute units).
  • We could have two versions of each S(Q), one as now, with radius_effective in S(Q) constrained to the geometric parameters in the form factor, and a second where a radius_effective parameter appears which can be separately adjusted - as most of us scientists would rather use this.

Even better, with a single version of each S(Q), radius_effective could always appear, with a three state selector - fitting, fixed or constrained.

  • Such a three state selector would be useful to turn constraints on & off without having to edit the constraints tab. Note the current constraints were originally envisaged as operating between two "Fit Page" tabs rather than within a single "Fit Page" tab, which as a general feature would have consequences for polydispersity integrations.
  • Constraints within a single "Fit Page" tab become particularly usful if there is a way to rapdily introduce an extra parameter.

e.g. I want a new parameter R_ratio =(radius_core + thick_shell)/radius_core but don't want to write a re-parametrised model to fit, or more likely fix, the value of R_ratio. We may decide that in sasview the only good way to do this is actually to have to write a new model….. (sigh).

  • Introduce a system in the models to flag an issue if too many scale parameters being asked to adjust simultaneously. e.g. "Fit One Of(rescale, volfraction, sld_core.AND.sld_shell.AND.sld_solvent )" This is not completely foolproof as it fails if say sld_shell = sld_solvent. In theory the least squares ought to be able to spot perfect correlations then complain also.

Plotting

  • When fitting data, the fit should only appear on the plot associated with that fit page ( unlike v4.0 where it also appears on plots where the same data is used in other fit pages.)
  • If say S(Q) or P(Q) has been appended to a fit page plot, then they could auto-update. This may be too complex.
  • Better still add controls to the fit plot to "show rescaled S(Q), show P(Q), show unsmeared" etc as appropriate, with these auto-updating. Would have to remove them if S(Q) is removed from the model.
  • Note a "rescaled S(Q)", multiplied by 10n, can be useful on some types of plot, though frequently on a log(I(Q)) plot it looks rather flat!
  • It would be nice to be able to click on a data point on a graph and get its x,y or x,y,z co-ordinates. May or may not want to keep the current ability to get x,y of any point on the plot by mousing over it.