Averaging RMSDs/RMSEs in torsion scans

[mattwthompson]

These probably should do means instead of RMS. I'm less sure this applies to RMSE

---

Just noting that RMSD.from_data computes the RMS RMSD as opposed to the mean. Computing the mean is probably a better way to do it, but maybe RMSD.from_data should also be changed to use the mean? Should RMSD.from_data be used (and modified to have an include_hydrogens kwarg) instead of duplicating (a tiny bit)? Thanks.

Originally posted by @fjclark in #143 (comment)

[fjclark]

Thanks for addressing this.

Computing the mean is probably a better way to do it

This was based on a feeling that I'd seen this done more in the literature. However, having now had a scan of the literature, I'm not sure it's clear which is best/ standard -- sorry. I think I may have originally gone with RMS to be a bit closer to the max RMSD value (reported in the BespokeFit paper) than the mean, but I don't have a strong opinion as to whether mean/ rms/ max RMSD is reported. Danny has a mild preference for RMS.

Whichever we go with, it might be nice to name the metric so it's clear how it's calculated, e.g. RMSRMSD or MeanRMSE in the same way there's RMSE and MeanError.

I think RMSE.from_data is fine as-is.

[mattwthompson]

If neither is obviously better than the other, I'm inclined to keep both in. I think the calculation is relatively cheap, even at scale. The downside is giving the user too many metrics, but I think they (mostly you) can handle it

[fjclark]

Sounds good to me. A barrier height error metric might also be nice. I'll try to get round to adding this soon (currently travelling / at a conference).

[jameseastwood]

Hey team! Please add your planning poker estimate with Zenhub @mattwthompson @j-wags