Fuzzy Observation Key Matching Needs Work

[cortner]

The current implementation via Levenshtein distance is not sufficiently general. For example,

compare(lowercase("dft_energy"), "energy", Levenshtein())  # -> 0.6

but the requirement is > 0.8. As a result, when importing the Si training set, no observations are found at all.

Two possible ways forward;

• return to a list of admissible keys for each observation
• choose the observation type which best matches the default observation key

E.g.,

compare(lowercase("dft_virial"), "energy", Levenshtein())  # -> 0.1
compare(lowercase("dft_forces"), "energy", Levenshtein())  # -> 0.1
compare(lowercase("e"), "energy", Levenshtein())  # -> 0.167
compare(lowercase("forces"), "energy", Levenshtein())  # -> 0.0
compare(lowercase("f"), "energy", Levenshtein())  # -> 0.0

@casv2 Since you wrote the first version, I hope you'll be willing to work on this.

[casv2]

I think this problem is now fixed