Add ability to read neutron weights from Event Nexus files

[Stuart Campbell]

If we produce an event nexus file from a simulation, then this file will need to contain a list of neutron 'weights'. The obvious name for such a field would be event_weight.

Can we alter LoadEventNexus to load the weights field, if it exists, and return an EventWorkspace with weighted events ?

[Russell Taylor]

A few thoughts on this, partly arrived at via discussion with Michael....

I think it's critical here that we don't slow down loading for people loading actual data event nexus files (as the ratio of those loading real to simulated data will be high number). (Actually, if we implement this very poorly we could slow things down for people loading any kind of file.)

Arguably, it's also undesirable to pollute the main LoadEventNexus algorithm with this if it's going to be very specific kind of file. If we can agree on a standard way of representing simulated data in event nexus files then this is less of a concern. Otherwise, we could have essentially a wrapper algorithm that first calls LoadEventNexus to load the file and then re-parses the events in the file to obtain and add the weights. This will certainly be slower than doing it in one shot, but see my comment about the ratio of users. Having said all this, a first look at the code does suggest the complexity of the two-step approach may not be low.

If we follow the separate algorithms approach, adding access via the generic Load algorithm could be a later step. In that situation it might be desirable for filenames of simulated data to carry a different extension for quick identification. Michael also mentioned that we could make use of the 'definition' field in the nexus file to flag it as containing simulated data, which would be simpler than digging into the actual data groups to check for a field.

[Nick Draper]

I agree with Russell on this one, Our LoadEventNexus is written for RAW data straight for the instruments and is optimised that way.

I'm happy for us to add another algorithm to support this, and it can hapily share code with LoadEventNexus, or event inherit from it, but we should not alter what is arguably our most used algorithm in any way that could slow things down.

[Stuart Campbell]

I think that keeping it separate is a good idea. As long as (for the user) it is easy to load the data I don't think it matters to the user how we do it internally.

I think that we can either modify the definition field, or leave it the same as it would be for a 'real' data file and just add an attribute to it (e.g. simulated = true ?). There is also a field for the program name that generated the data file, this could be used as a secondary check as well - so we don't need to dig into the actual data groups.

[Michael Reuter]

Refs #6090. Add ability to read simulated data.

It turns out that adding the code alongside produced no dramatic changes in the performance of the loader, at least in all the tests that I've run (about 40 or so on two difference machines). Being able to do this simplified the code necessary to achieve this capability.

Changeset: 8ea3af487b2b13de422cbb52d5cddbd785f51623

[Michael Reuter]

Refs #6090. Correcting the value of the weight.

Changeset: 3c4f70463a5cae8a114b7a60a5e2dbd19e334a45

[Michael Reuter]

The last check-in was actually correcting the value for the error squared.

[Michael Reuter]

Refs #6090. Typo in comment.

Changeset: 3c4a7e5014b39672672aa9194e55f7b82e3e7b9d

[Michael Reuter]

Refs #6090. Adding unit test for simulated file.

Changeset: bc756b4a292be47d692ae324b5edd8dc6defad1b

[Michael Reuter]

Reduction script for simulation file

[Michael Reuter]

TESTING: Make sure that the unit test for LoadEventNexus is running. Also, run the attached reduction script to see that data does actually come out. There should be a peak of data at deltaE ~ 770 meV in the 1D plot.

[Michael Reuter]

Refs #6090. Add ability to read simulated data.

It turns out that adding the code alongside produced no dramatic changes in the performance of the loader, at least in all the tests that I've run (about 40 or so on two difference machines). Being able to do this simplified the code necessary to achieve this capability.

Changeset: 8ea3af487b2b13de422cbb52d5cddbd785f51623

[Michael Reuter]

Refs #6090. Correcting the value of the weight.

Changeset: 3c4f70463a5cae8a114b7a60a5e2dbd19e334a45

[Michael Reuter]

Refs #6090. Typo in comment.

Changeset: 3c4a7e5014b39672672aa9194e55f7b82e3e7b9d

[Michael Reuter]

Refs #6090. Adding unit test for simulated file.

Changeset: bc756b4a292be47d692ae324b5edd8dc6defad1b

[Stuart Campbell]

Tested using the ARCS_sim_event.nxs file. Loads with no problems either calling LoadEventNexus() directly or using Load().

[Stuart Campbell]

This ticket has been transferred to github issue ​6936