declare 'Shift' parameter for TabulatedFunction

[Jose Borreguero]

Indicated for the scenario when we want to compare two workspaces representing two structure factors that may be shifted with respect to each other.

[Jose Borreguero]

Refs #10279 Added Centre parameter

Changeset: d3891e486f95b1c8390c42eabd79ff484e1df2d1

[Jose Borreguero]

QENS dynamics structure factor

[Jose Borreguero]

fit the data

[Jose Borreguero]

Refs #10279 Bypass the test for the moment

Changeset: 1643dc7f4f5bd68025f70d391eb68b8594c84d7c

[Jose Borreguero]

Refs #10279 Repaired test_derivatives

Changeset: a1bccc219b35f726b8185f8051b5a5d0b39a0a89

[Anders Markvardsen]

Hi Jose,

I have put this ticket as blocked by #10287.

The reason is that your changes have caused a ISIS SANS LOQ systemtest to fail. Please read comment 7 of this ticket, it has been found that your changes to TabulatedFunction is unfortunately now causing TabulatedFunction to fail, see comment 7 in #10287.

Could you either update the code referred to in comment 7 of #10287 or advice me on how I can get TabulatedFunction to work again in the code referred to in this comment 7

[Jose Borreguero]

Refs #10279 Fixed numerical deriv and added usage example

Changeset: 25c74fb856e5d13566e538e63880b92a5d7186f8

[Jose Borreguero]

Refs #10279 static cast required to avoid compiler warning

Changeset: 41ff6e3a27808c9069ad5e745a7e640ff451c8f6

[Martyn Gigg]

There is a Sphinx documentation warning in TabulatedFunction.rst: ​http://builds.mantidproject.org/job/develop_clean/label=win7-build/377/warnings3Result/new/

[Anders Markvardsen]

Hi Jose,

I have removed the block.... After a few more hours I eventually found that one additional reason for the failure was that an additional parameter means that the fitting parameter table output contains one additional row, hence any python code that is using TabulatedFunction fit-function and try to retrive information about the fitted parameter will almost certainly now fails.

More specifically code like this will now fail:

{

param = mtd__fitRescaleAndShift_Parameters?

row1 = param.row(0).items() row2 = param.row(1).items() row3 = param.row(2).items() row4 = param.row(3).items() scale = row1[1][1] shift = row3[1][1] chiSquared = row4[1][1]

}

In addition, for example, in the affected code in #10287, after speaking to SANS instrument scientist, they specifically do not want a Shift parameter to be fitting since they are fitting two spectra against each other that have fixed x-values.

Therefore before putting this ticket to fixed please:

Check to see if TabulatedFunction is used anywhere in the Mantid code. By default assume that the behaviour is to keep Shift fixed. Note in the Code/Mantid/script folder it only used once and this is relate to #10287, and I can update this when relevant.

No c++ algorithms appear to use TabulatedFunction, but I haven’t check python algorithms. However Resolution fit-function uses TabulatedFunction, but speaking to Roman this is OK since here TabulatedFunction parameters are not fitted here. TabulatedFunction is used in ConvolveWorkspaces – check that this is OK.

Optionally email Mantid-develop to inform about this change.

Make sure that the next release note mention that TabulatedFunction have been updated with an additional fitting parameter.

Note if you find that TabulatedFunction is used too heavily, consider leaving TabulatedFunction as is and add a function TabulatedFunctionWithShift.

I have reversed the block so that this ticket now blocks #10287

[Jose Borreguero]

Replying to Martyn Gigg:

There is a Sphinx documentation warning in TabulatedFunction.rst: ​http://builds.mantidproject.org/job/develop_clean/label=win7-build/377/warnings3Result/new/

Thanks, I had missed this. I'll correct it ASAP

[Jose Borreguero]

Hi Anders,

Algorithm ConvolveWorkspaces doesn't fit anything so parameters of Tabulatedfunction are of no relevance. Their default values are used.

Resolution does not have fitting parameters, so again, they are of no relevance. By the way, I don't see the point of this Resolution function. I can't make the difference between this function and a TabulatedFunction with fixed fitting parameters. In fact, a resolution function should incorporate a Shift parameter to accommodate deviation of the experimental data from the origin of energy transfers when fitting against a model, but this is an issue for another ticket.

[Jose Borreguero]

Note to tester:

Build the documentation too, and test for yourself the example given there.

[Anders Markvardsen]

Branched has been merged into master, but this was not recorded on this ticket

[Stuart Campbell]

This ticket has been transferred to github issue ​11121