StichingError

[Owen Arnold]

Error reported by Tim Charlton.

For one of the spin states, running Stitch1DMany results in magnification of error bars for one of the stitched workspaces. The steps below are the minimum to reproduce the reported issue. Screen shot attached.

Load(Filename=r'/Users/spu92482/Desktop/Stitching-errors/nIp_DataOut_RcorrectedVsQz_r1q_rr1.nxs',OutputWorkspace='nIp_DataOut_RcorrectedVsQz_r1q_rr1')
Load(Filename=r'/Users/spu92482/Desktop/Stitching-errors/nIp_DataOut_RcorrectedVsQz_r2q_rr2.nxs',OutputWorkspace='nIp_DataOut_RcorrectedVsQz_r2q_rr2')
Load(Filename=r'/Users/spu92482/Desktop/Stitching-errors/nIp_DataOut_RcorrectedVsQz_r3q_rr3.nxs',OutputWorkspace='nIp_DataOut_RcorrectedVsQz_r3q_rr3')

Stitch1DMany(InputWorkspaces='nIp_DataOut_RcorrectedVsQz_r1q_rr1, nIp_DataOut_RcorrectedVsQz_r2q_rr2 ,nIp_DataOut_RcorrectedVsQz_r3q_rr3',OutputWorkspace='stitchednpl',Params='-0.02')

graph_spec = plotSpectrum('stitchednpl',0, error_bars=True)
l = graph_spec.activeLayer()
l.setAxisScale(Layer.Left, 1e-8, 1, Layer.Log10)

[Owen Arnold]

Error reproduced on version 20th May 2014

[Owen Arnold]

I can actually reproduce this issue via stitch1D directly by stitching rr2 and rr3 together. So the real problem is not in Stitch1DMany, it is in Stitch1D or something lower.

[Owen Arnold]

Following some investigation. This was found to be down to some very large error values in the most RHS workspace. In the integration step, errors are squared and sum across all bins. Therefore the ratio multiplication factor ws, had a huge error value that was scaling all the error bars on the corrected, scaled output workspace.

Changing the StartOverlap region as follows fixed the issue as it cuts out the overlap regions where errors are very high.

Stitch1DMany(InputWorkspaces='nIp_DataOut_RcorrectedVsQz_r1q_rr1, nIp_DataOut_RcorrectedVsQz_r2q_rr2 ,nIp_DataOut_RcorrectedVsQz_r3q_rr3',StartOverlaps=[0.0208,0.0208], EndOverlaps=[0.05,0.05],OutputWorkspace='stitchednpl',Params='-0.02')

What Tim has now asked for is to have the error scale factor output from Stitch1D and Stitch1DMany. This would allow the end-user to spot this problem.

[Owen Arnold]

comment 4 outlines the real problem, and this has been accepted by the user. There are no code changes.

[Michael Reuter]

Sounds fine to me.

[Stuart Campbell]

This ticket has been transferred to github issue ​10367