Closed Luthaf closed 2 years ago
Same thing happens with separate fit/transform calls:
# compute per_atom_soap ...
print(np.linalg.norm(per_atom_soap))
scaler = skcosmo.preprocessing.StandardFlexibleScaler(copy=True)
for _ in range(10):
scaler.fit(per_atom_soap)
X = scaler.transform(per_atom_soap, copy=True)
print(np.linalg.norm(per_atom_soap))
Forget the two previous comments, there is a bug in equistore python bindings & I was using invalid memory. This change is still required but should be fine!
Something strange is hapening here: even with this change, I get the following:
output:
I expect the output to always be the same