ADR suggestion: Experiment

[henrikjacobsenfys]

General

We need a class to handle everything relating to the experiment: holding data, resolution function, background. It should also be able to plot data. The Experiment class will be attached to job, along with Analysis #5 and SampleModel #4 .

It's unclear to me exactly how multidimensional data should be handled, so for now we'll make a proof of concept with a single one-dimensional data set. In the end, we need multi-dimensional data sets that must be binned in some way before they can be analysed. I think the binning should be handled by Experiment.

Current implementation

There is no Experiment class

Proposed implementation

The class will inherit from the corelib ExperimentBase class. It will have the following classes and methods.

Data

This holds the data as a scipp DataArray

append: appends a data set to the list of data sets
remove: removes a data set

plot: plots data. Probably needs a bunch of settings to make the plot nice.

remove_outliers: I want to be able to detect data points that are clear outliers and remove them. I'm unsure exactly what criteria I will need for this.

We will furthermore need/make use of a bunch of scipp methods such as binning/grouping. We might want to wrap them in a nice way to hide most of the syntax from the users. I'll need to play around with it a bit to see what's needed.

Resolution

This holds the resolution, for now only as a SampleModel. Later, we will allow users to use a measurement as the resolution, but there are some intricacies that must be dealt with first.
set_resolution Sets the resolution. When we have multi-dimensional data, the resolution should be the same dimension as the data.
There will be more settings for the resolution, see #2 . I think that I will expect multiple data sets to have the same resolution (because they would be the same instrument configuration, but e.g. different temperatures or other external parameters), but I'm not sure yet.

Background

The background model will also be a SampleModel, but explicitly without resolution convolution or detailed balancing (#6 )

Other

In addition to the above, we also want to store other experimental information. This could be e.g. temperature, sample name, sample mass, sample formula, general notes. We can consider implementing absolute normalisation, although that depends on what the IDS and instrument responsibles expect. It would require measurements of a standard and some straightforward but tedious math.