@janosh This seem to cause the behaviour change in #215, is this an accidental change? I think the original default value of 0 is correct here? i.e. using 0 for element this is missing?

thanks for looking into this! 🙏 i think returning None for missing elements in count_elements is more correct/semantic than 0. so ideally, ptable_heatmap_ratio should be updated to work with the new behavior

i think returning None for missing elements in count_elements is more correct/semantic than 0

I guess both choices hold merit from certain standpoint, i.e. None is more semantic to suggest that particular element is missing, 0 is also valid as its count is indeed zero.

For the particular function, however, I slight prefer using zero because it seems to be designed to "count element occurrence":

So its occurrence would be zero if that particular element doesn't appear in the data.

ideally, ptable_heatmap_ratio should be updated to work with the new behavior

I'm afraid this would have broader impact as there're quite a few usage that need to be checked:

For example:

from pymatviz import count_elements
from pymatviz.enums import ElemCountMode, Key
from matminer.datasets import load_dataset


dataset = load_dataset("matbench_expt_gap")[Key.composition]
counted_dataset = count_elements(dataset, ElemCountMode.composition, fill_value=0)
print(counted_dataset)

With fill value zero:

symbol
H      235.00
He       0.00
Li     967.66
Be     121.00
B     1168.00
       ...   
Fl       0.00
Mc       0.00
Lv       0.00
Ts       0.00
Og       0.00

With None:

symbol
H      235.00
He        NaN
Li     967.66
Be     121.00
B     1168.00
       ...   
Fl        NaN
Mc        NaN
Lv        NaN
Ts        NaN
Og        NaN

Using None seem to require an additional process step .

So do we want to change this completely or revert this?