Aromatic `[o+]` in chromenylium is not supported

[pechersky]

We were training a model to recognize dyes, and it failed on Q-Switch 5. Is there a way to expand the list of acceptable tokens?

ValueError: Tokens {'[o+]'} in c1c(-c2[o+]c3c(c(-c4ccccc4)c2)CCCC3=CC2=CC(=CC3=C4C(=C(c5ccccc5)C=C(c5ccccc5)O4)CCC3)CCC2)cccc1 are not supported by the model
Allowed tokens are: ({'1', 's', '[n+]', '2', ')', '=', 'C', '7', '[N+]', '[nH]', '[S+]', 'n', '(', '%10', '3', 'O', '^', 'S', 'c', '5', '#', '[N-]', '8', 'Br', 'o', '6', '4', 'Cl', '-', 'N', '[O-]', '$', '9', 'F'}, set())

[halx]

Oxonium and similar ions are not supported in the current priors. You would need to retrain a new model. Question is how useful that would be. Pubchem for instance has only about 6500 examples.