You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When Evaluator parses a ThermoML file, it tries to read SMILES. If the SMILES fails (e.g. on a radical error), the exception means the rest of the file is also ignored, instead of that singular entry. Due to differences in the OpenFF toolkit over different versions (e.g. the addition of the radical error), this means some past work is unnecessarily highly unreproducible (e.g. the filtering of the Sage training dataset for 2.0). Beyond reproducibility concerns, this also impacts the reading of any properties later in files that contain invalid molecules, which I ran into while trying to read viscosities.
Desired solution
Wrap the call below, or the individual smiles call within it, in a try/except:
Describe the bug
When Evaluator parses a ThermoML file, it tries to read SMILES. If the SMILES fails (e.g. on a radical error), the exception means the rest of the file is also ignored, instead of that singular entry. Due to differences in the OpenFF toolkit over different versions (e.g. the addition of the radical error), this means some past work is unnecessarily highly unreproducible (e.g. the filtering of the Sage training dataset for 2.0). Beyond reproducibility concerns, this also impacts the reading of any properties later in files that contain invalid molecules, which I ran into while trying to read viscosities.
Desired solution
Wrap the call below, or the individual smiles call within it, in a try/except:
openff-evaluator/openff/evaluator/datasets/thermoml/thermoml.py
Lines 2187 to 2188 in 7d927d9
Output
Computing environment (please complete the following information):
conda list
Additional context
FWIW I don't think (~70% sure) that fixing this issue will make the Sage filtering 100% reproducible, but... less unreproducible.
The text was updated successfully, but these errors were encountered: