Look at adding RMSD as a metric.

[CompRhys]

For the forcefield methods we can also report the RMSD of the relaxed structure to the final structure as a metric. This would require us to store the final structures somewhere which may necessitate LFS.

[DanielYang59]

Hi @CompRhys sorry for the late response, just remove fmt (now value_fmt) and cbar_fmt (now cbar_value_fmt) and the default settings would work.

 ax_mp_cnt = pmv.ptable_heatmap(  # matplotlib version looks better for SI 
     elem_counts, 
-    fmt=lambda x, _: si_fmt(x, ".0f"),  
-    cbar_fmt=lambda x, _: si_fmt(x, ".0f"),  
     zero_color="#efefef", 
     cbar_title=f"{dataset.upper()} Element Count", 
     log=(log := SymLogNorm(linthresh=100)), 
 ) 

I got the following (updated):

mp-element-counts-by-composition-symlog.pdf
mp-element-counts-by-occurrence-symlog.pdf
wbm-element-counts-by-composition-symlog.pdf
wbm-element-counts-by-occurrence-symlog.pdf

We removed support of value_fmt as a Callable during a previous refactor, but I would be happy to discuss this if there're use cases where a Callable is necessary (si_fmt for example).

Let me know if there are more issues.

[DanielYang59]

Quote from Janosh:

if those custom kwargs fix the issue, we should make those values the new defaults.

Turns out the default for ptable_heatmap is good, i.e. simply removing both fmt (now value_fmt) and cbar_fmt(now cbar_label_fmt) args would generate proper quality plots (PDFs in above comment updated).

[janosh]

Turns out the default for ptable_heatmap is good

thanks for pointing that out @DanielYang59! fixed in c1f34da

[DanielYang59]

No problem!

[CompRhys]

Closing in favour of #152