Saw Tooth Stitching

[Owen Arnold]

The current reflectometry stitching procedure involving combineMulti.py is generating a saw-tooth fit under certain conditions. See screenshot.

Need to investigate and hopefully fix the cause of this.

Script to replicate this is below. Dat files will be attached.

from isis_reflgui import combineMulti

import os

DataDir= os.path.expanduser('~/Desktop/Tims')
print os.path.join(DataDir, 'polref8481.dat')
r8481 = Load(Filename= os.path.join(DataDir, 'polref8481.dat'),Unit='MomentumTransfer') 
r8482 = Load(Filename= os.path.join(DataDir, 'polref8482.dat'),Unit='MomentumTransfer')
r8483 = Load(Filename= os.path.join(DataDir, 'polref8483.dat'),Unit='MomentumTransfer')
r8481 = SortXAxis(r8481)
r8482 = SortXAxis(r8482)
r8483 = SortXAxis(r8483)

Rebin(InputWorkspace='r8481',OutputWorkspace='r8481',Params='0.009,-0.02,0.034',PreserveEvents='0') 
Rebin(InputWorkspace='r8482',OutputWorkspace='r8482',Params='0.017,-0.02,0.06',PreserveEvents='0')
Rebin(InputWorkspace='r8483',OutputWorkspace='r8483',Params='0.027,-0.02,0.100',PreserveEvents='0')

ConvertToHistogram(InputWorkspace='r8481',OutputWorkspace='r8481h')
ConvertToHistogram(InputWorkspace='r8482',OutputWorkspace='r8482h')
ConvertToHistogram(InputWorkspace='r8483',OutputWorkspace='r8483h')
wcomb = combineMulti.combineDataMulti(['r8481h' , 'r8482h', 'r8483h'],'test_IvsQ',[0.009, 0.018, 0.04],[0.03, 0.046, 0.1],0.009,0.1,binning=-0.02,keep=1)
plotSpectrum(['test_IvsQ'], 0)

[Owen Arnold]

It looks like this is an issue resulting from the fact that the input data has errors set to zero. Stitching involves extracting the overlap regions and performing a weighted mean of the contribution from both workspaces being stitched. Because the WeightedMean will multiply each set of bin counts by the harmonic mean of the error values from both workspaces, if the errors in any workspace are zero, then the weighted mean will also be zero.

My way around this is to exit the processing of combineMulti if the input data has completely zero errors.

[Owen Arnold]

It might be more sensible yet, to determine whether there are complete zeros in the errors for all spectra and apply either WeightedMean, or Mean based on the outcome. We have no Mean algorithm, so I'm going to do that under a separate ticket (#7983).

[Nick Draper]

moved to Release 3.0 as these all seem to be active

[Owen Arnold]

This ticket is implicitly fixed by #8034, which is still in a verify state, because Stitch1D now calculates the Mean internally if the WeightedMean cannot be used. Do not verify this until #8034 has been verified. No branches/code changes required here.

[Jay Rainey]

The changes that were made in #8034 address this problem and a saw-tooth fit no longer occurs — ​screenshot.

[Stuart Campbell]

This ticket has been transferred to github issue ​8827