Opened 15 years ago

Closed 14 years ago

#20 closed bug (fixed)

NSORT default to NO autoscale to get user intervention if bad overlap

Reported by: srkline Owned by: kline
Priority: major Milestone: SANSReduction_5.1_Release
Component: SANS Reduction Keywords:
Cc: Blocking:
Task:

Description

(Originally an RT issue, moved to trac)
(Paul's summary of the issue:
if I understand it and remember it, the final issue was really what can
be done to warn users when splicing data that using data with large
error bars to normalize the other data set is a bad plan. Also I guess
we as instrument scientists need to have a consensus. For example some
always recommend normalizing to the low Q data - which would be wr
exactly wrong for this problem. Others suggest the higher q to avoid
this I think, but that has the problem that the low q of the higher q
data is where the biggest resolution effects come from. Others I think
recommend NOT normalizing because it should be pretty good. Problem of
course is that you then don't catch issues that might arise as quickly.

So as a first step at resolution Can Lionel please indicate what he
suggested as the warning (I can't remember) also Others may want to
weigh in on what they think the proper (standard) way to normalize
spliced data should be?

--

My (ajj) opinion is that the software should default to no auto scaling and users should ensure that
their data from different configs have been independently put on absolute scale.

We should have the ability to preview what will happen without having to save to file and a
warning should pop up if the overlap regions differ by more than 5% (say). That then gives the
user the choice to ignore it or do some scaling. This is where they should be asking for advice if
they aren't sure.

Andrew
--
Andrew Jackson

<URL: http://rt.ncnr.nist.gov/Ticket/Display.html?id=729 >

Change History (5)

comment:1 Changed 15 years ago by srkline

  • Milestone set to SANSReduction_5.1_Release

comment:2 Changed 15 years ago by srkline

The panel does now default to no autoscaling (changed the checkbox value in the panel macro)

Still need to decide about the point of intervention.

comment:3 Changed 14 years ago by srkline

the NSORT panel now also calculates the scaling constants and moves the data when the <Write Combined File> button is pressed. Move the dialog out of the way (or just cancel) to see if the shift is acceptable. It should be obvious if the overlap is "good" when the 2 or 3 data sets are plotted, long before the numerical values are calculated.

Set up this way, the program could intervene if the overlap is "too bad", or the user can save the data as is without caring about what the overlap shifts are.

comment:4 Changed 14 years ago by srkline

How about two buttons - "Preview" and "Save"? If you know it'll be alright, just save. If you're curious what may happen, "Preview" - and some diagnostic dialog can be presented if some tolerance criteria is not met.

comment:5 Changed 14 years ago by srkline

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

Changed NSORT to have two new checkboxes:
1) Preview Only - if checked does the indicated scaling, does not write a file
2) Overlap warning - if AUTO overlap is 5% away from unity, a warning is presented. If you manually enter a number that is too big or small, that is your problem.

Defaults are for NO autoscale, NO preview, YES Warning. This has the effect of writing out files with all scalings == 1.

Note: See TracTickets for help on using tickets.