I like the proposal! However, my numpy.ndarrays do not feature a .count() method. Which version is yours?

I got this one to work:

import cclib
import numpy
def stoichoimetry(atomnos):
    """Return the stoichemistry of the object."""
    pt = cclib.parser.utils.PeriodicTable()
    formula = []
    unique, counts = numpy.unique(atomnos, return_counts=True)
    counter = {u:c for (u,c) in zip(unique, counts)}
    for ano in sorted(unique, reverse=True):
        count = counter[ano] if counter[ano] > 1 else ''
        formula.append('{}{}'.format(pt.element[ano], count))
    return ''.join(formula)

However, this is not the same as Gaussian's!

For example, I have a Gold complex with this stoichiometry: C17H26AuN2(1+), but this approach outputs AuN2C17H26. While we could add charges, the sorting is non-trivial I think...

I like the proposal! However, my numpy.ndarrays do not feature a .count() method. Which version is yours?

You are right, I forgot atomnos is an ndarray. I just commited a fix, but you can also just incorporate your fix.

However, this is not the same as Gaussian's!

If you are concerned about the order of elements, we can add some heuristics to the code. Do you have a good reference for how the formula should be formed?

For example, I have a Gold complex with this stoichiometry: C17H26AuN2(1+), but this approach outputs AuN2C17H26. While we could add charges, the sorting is non-trivial I think...

We can add the charge, just use the charge attribute.