Unexpected behavior of CrystalNNFingerprint

[utf]

The output of CrystalNNFingerprint does not match up with the output of the CrystalNN class.

For example the following code snippet:

import numpy as np
from matminer.featurizers.site import CrystalNNFingerprint
from pymatgen.ext.matproj import MPRester
from pymatgen.analysis.local_env import CrystalNN

s = MPRester().get_structure_by_material_id("mp-707852")
s.add_oxidation_state_by_guess()
bs = CrystalNN().get_bonded_structure(s)

site_of_interest = 51

cnnf = CrystalNNFingerprint.from_preset("ops")
features = cnnf.featurize(s, 51)

sort_indices = np.argsort(features)
top_id = sort_indices[-1]
second_top_id = sort_indices[-3]

print("coordination of site {}: {}".format(
    site_of_interest, bs.get_coordination_of_site(site_of_interest)))
print("top feature: {} with OP: {}".format(
    cnnf.feature_labels()[top_id], features[top_id]))
print("second top feature: {} with OP: {}".format(
    cnnf.feature_labels()[second_top_id], features[second_top_id]))

Produces the output:

coordination of site 51: 2
top feature: wt CN_1 with OP: 0.6286992249320102
second top feature: wt CN_3 with OP: 0.3295052806667878

The coordination of site 51 obtained by the CrystalNN class does not match up with the coordination indicated by CrystalNNFingerprint. Even the second highest order parameter is for the wrong coordination number.

[ardunn]

@utf any update on why this might be? Fixing this issue might improve automatminer as a secondary effect