Simple fitting interface for multiple data sets from multiple detectors/instruments

[krzywon]

After discussions with @timsnow and @butlerpd regarding multi-detector and multi-instrument measurements, I propose a new fitting mode. The mode would be a hybrid between batch fitting and simultaneous/constrained fitting. Any number of data sets could be added to a single fit page, as with batch fitting, but each fit parameter could be constrained and fit to a single value across all data sets, similar to the simultaneous/constrained fitting.

Background:

During the canSAS XI meeting, there was a discussion on storing reduced multiple-detector and multiple-instrument 2D data sets in the NXcanSAS format. The flexibility of the NXcanSAS format allows multiple data objects to be stored in a single entry and/or multiple entries within a single data file. The use cases where this could be useful are a single sample run at different sample environment conditions, different times, or on different instruments, or, conversely, a single SAS measurement on an instrument with more than one detector.

At the time of the discussion at the canSAS meeting, two ideas were offered. The first was to combine the data sets into a single 2D object and store that as a single data element. The second was to keep each 2D panel as a separate data element but have multiple data elements within each entry. The discussions with @timsnow and @butlerpd all came to the conclusion that keeping the data sets separate made the most sense from an error analysis and resolution perspective.

In SasView, the NXcanSAS data loader loads each data object as a separate data1d/data2d object. For multi-detector measurements, this will load n data objects, each independent of one another. A suggestion of combining the data in some way within SasView was quickly thrown out, using the same arguments for why the data should remain separate in the data sets, and the focus switched to fit these data. Currently, it is possible to fit any number of data sets using a simultaneous/constrained fit, but that requires each data set to be loaded into a separate fit page, the fit parameters to be selected in all panels, and then create constraints so all n sets are constrained to one another. This quickly becomes cumbersome when dealing with a large number of data sets.

Use cases:

• One instrument, one measurement, multiple detectors
  • All fit parameters should be constant across all detectors.
• One sample, measured n times each on m instruments
  • All fit parameters should be constant across all detectors
• One sample, n measurements as a function of another variable, e.g. at different times for stability measurements, across multiple temperatures for phase information, etc.
  • Some parameters might change from measurement to measurement, but a check to see if they remain constant would be helpful. Solvent and sample slds would remain constant.
• A particle suspended in a number of different solvents.
  • The solvent sld and background would change, but the particle SLD would remain constant. The particle shape/size might vary
• Any combination of the above
• Others?

Proposed fitting mode options:

In this mode, any number of data sets could be sent to a single fitting window, to be fit to a single model, similar to how a batch fit works. However, any number of parameters could be constrained across all data sets and fit simultaneously to find the best value, similar to a constrained fit. The following suggestions have come up:

1. Modify batch mode and add a 'constrain across all data sets' check box for each fit parameter. If no parameters are constrained, a normal batch fit is performed. If one or more parameters are constrained, constraints are automatically generated by the program to tie the parameter across all data sets and a constrained fit is performed.
2. Modify the base fit page, allowing any number of data sets to be added to each page, thereby changing all base fit pages into batch fit pages. Add the suggestion in (1) of a constrained checkbox for each parameter. The 'batch fit' option in the data explorer window would change to 'append selected to active page'.
3. Modify the simultaneous/constrained fit mode allowing the option to fix a specific parameter across all data sets.
4. Create a separate mode that is a hybrid of batch and simultaneous fitting.
5. Others?

The one issue we want to avoid is adding an extra layer of complexitiy to the UI. Tidying up what already exists might be a good move forward before tackling this.

Related tickets:

This overlaps what is requested in #256, but with more background and a firmer suggestion on how to handle the specific use cases.