Correct handling of distributions

[Nick Draper]

Batch move of tickets to Iteration 17

[Nick Draper]

Moved as part of iteration end

[Nick Draper]

Moved as part of iteration 18 end

[Nick Draper]

We encountered an issue in the GEM analysis where once the data was corrected against vanadium with a divide the data was essentially a distribution without being marked as one. In this state rebin commands led to unexpedted shifting of the data.

Should divide mark the workspace as a distribution if the units match?

[Nick Draper]

Moved as part of the end of Iteration 19

[Russell Taylor]

(In [3573]) The rebinNonDispersive was found to be buggy in the context it was being used in the SANS analysis. In fact, the rebin function does the same job if the data are flagged as a distribution so change to using that instead (and modify the rebin function to work with cumulative rebinning as the rebinHistogram one does). Re #438.

[Russell Taylor]

(In [3580]) Change DiffractionFocussing2 to use rebin instead of rebinNonDispersive in the same way as Q1D. Tests show that it makes no difference to the result (well < 10^{-11 anyway). Re #438.}

[Russell Taylor]

(In [3592]) Now that the last known use of RebinPreserveValue has been removed in revision [3951], delete the algorithm. Re #438.

[Russell Taylor]

That should have been revision [3591].

[Russell Taylor]

I think I understand why what we're doing in the SANS analysis is OK now....

[Russell Taylor]

OK, we still have an issue to sort out here, and I think I know what it is. Basically, there are (at least) two 'kinds' of distribution: one where the data's been divided by the bin width and another where it's dimensionless (e.g. after a divide). Our rebin gets things right, but ConvertUnits doesn't. Genie accounts for this in it's unit conversion routine (it doesn't multiply by the bin width if it's dimensionless).

So there are several things we need to do:

• Fix ConvertUnits
• Perhaps enhance our YUnit to be more than just a string
• Enhance Divide (& multiply?) to adjust the Y unit and distribution flag in the general case.

This all follow from the following from Rob Dalgliesh:

I've been trying replicate data extraction I can do in openGenie using

Mantid and have come across a show stopper....

The script below illustrates the problem I have.

Essentially if I take a data set full of 1s I should be able to convert

to Q in distribution form and still get a straight line.

This isn't the case (plot "test1" to see). It looks like the division by bin size is wrong as you get a logarithmic increase in the data.

######################################################################
#Python Script Generated by Algorithm History Display 
######################################################################
LoadRaw(Filename="W:/RawFiles/OFFSPEC00005588.raw",OutputWorkspace="OFFSPEC00005588",LoadLogFiles="0")
ConvertUnits(InputWorkspace="OFFSPEC00005588",OutputWorkspace="5588lambda",Target="Wavelength")
CropWorkspace(InputWorkspace="5588lambda",OutputWorkspace="5588spec113",StartWorkspaceIndex="113",EndWorkspaceIndex="113")
Divide(LHSWorkspace="5588spec113",RHSWorkspace="5588spec113",OutputWorkspace="unitywksp")
ReplaceSpecialValues(InputWorkspace="unitywksp",OutputWorkspace="unitywksp",NaNValue="1",NaNError="1",InfinityValue="1",InfinityError="1")
ConvertUnits(InputWorkspace="unitywksp",OutputWorkspace="unityQ",Target="MomentumTransfer")
Rebin(InputWorkspace="unityQ",OutputWorkspace="test1",Params="0.007,-0.01,5")
ConvertToDistribution(Workspace="test1")

[Russell Taylor]

(In [3672]) Index error in SimpleIntegration when the input workspace is a distribution. Re #438.

[Russell Taylor]

(In [3730]) Changes so that we handle distributions better. In particular, Divide now changes its output workspace to be a dimensionless distribution if the inputs have the same YUnit. ConvertUnits handles dimensionless distributions correctly. A new field is added to MatrixWorkspace (YUnitLabel) which holds a descriptive string for plotting. m_YUnit should only really be Counts or nothing. Re #438.

[Russell Taylor]

(In [3732]) Fix test. Re #438.

[Russell Taylor]

(In [3739]) Update reference results following changes to handling of distributions. Re #438.

[Russell Taylor]

(In [3805]) Remove ability to flip workspace distribution flag from Python. Shouldn't be necessary any more. Re #438.

[Russell Taylor]

The file showing the problem. Should end up with a flat line - didn't used to.

[Stuart Campbell]

This ticket has been transferred to github issue ​1286