fix: include resolution in stitched curve if present #88
Conversation
src/ess/reflectometry/tools.py
Outdated
| q_res = ( | ||
| sc.DataArray( | ||
| data=c.coords.get( | ||
| 'Q_resolution', sc.scalar(float('nan')) * sc.values(c.data.copy()) |
There was a problem hiding this comment.
Odd default. Is that really the unit you want? How about something like sc.full_like(c.coords['Q'], value=nan)?
src/ess/reflectometry/tools.py
Outdated
| qs_avg = np.nansum(qs * inv_v, axis=0) / np.nansum( | ||
| ~np.isnan(qs) * inv_v, axis=0 | ||
| ) |
There was a problem hiding this comment.
Why are we using np.nansum instead of sc.nansum?
There was a problem hiding this comment.
That's the way it was done in ess.reflectometry.tools.scale_curves_to_overlap (for various reasons, we're using scipy there so it's easier to just convert to numpy types once instead of going back and forth in every iteration) and this was just copied from there.
Fixed now 👍
SimonHeybrock
left a comment
There was a problem hiding this comment.
I cannot say anything about the math, should be reviewed by the scientist(s).
| # This might need to be revisited. The question about how to combine curves | ||
| # with different Q-resolution is not completely resolved. | ||
| # However, in practice the difference in Q-resolution between different curves | ||
| # is small so it's not likely to make a big difference. |
There was a problem hiding this comment.
Maybe this should be discussed in the docstring instead?
|
This has been inactive for a while now. We should either revive or close? |
|
I'll bring it up with Nico and ask him to review the procedure 👍 |
e645f2b to
56852e6
Compare
|
@paracini were you notified to review this one? |
|
@paracini ping? |
|
On Estia delta lambda varies between 4% (12Å) and 8% (3.5Å). The approach to combine curves with different resolutions used here, between measurements performed at multiple sample rotations, is similar to the way we are combining different wavelength resolution when integrating along lines of constant Q within one lambda-theta map (measured at a single sample rotation). It's a sensible way to go about it but it's important to remember the following: In the case of a single sample rotation there does not seem to be any major discrepancies in resolution in the overlap regions obtained by stitching neighbouring wavelength bands of a single lambda-theta map (first figure below) so combining resolutions at a single rotation should not cause any significant loss of information in the data. In the case of combining multiple sample rotations (this PR, second figure below) the resolution effects in the overlap regions is a lot more noticeable so care must be taken when stitching together curves with different resolutions. It is inherent to the variable lambda resolution that information is lost when higher resolution is combined with lower resolution curves and this is typically not done, if this affects the information content in the data (sharp features) then multiple rotations should not be stitched together. Stitching different wavelength bands from a measurement at a single sample rotation, good overlap, no major resolution effects in the overlap regions: Combining multiple sample rotations, can lead to loss of information in the overlap regions, overlap regions show significant resolution effects: |
|
Adding to this, here is a comparison of the Q resolution for each separate measurement (per sample rotation value) and the Q resolution of the combined curve (calculated using the procedure in this PR): 
|
|
Here is the same kind of figure displaying the reflectivity of each separate measurement compared to the combined curve from all measurements: 
Note that the combined curve overlaps the "highest angle" curve relatively closely in every part of the Q-range. |
Adds a
Q_resolutioncoordinate to the combined reflectivity curve if the provided curves have aQ_resolutioncoordinate.