SMILES and SELFIES augmentation

[niladell]

It may be useful for people using SMILES or SELFIES as an input to their models to have the option of running data augmentation. Meaning, that instead of always passing the canonical SMILES sequence, having the option to obtain randomly sorted SMILES sequences via datamol.

EDIT: Actually, I just saw that datamol.to_smiles has a randomize feature. Which is literally what I was thinking about. Maybe I would make it more obvious, i.e. create an "augmentation" tag or section. Maybe it could be coupled with other transformations?

[hadim]

Indeed the randomize flag just call dm.randomize_atoms under the hood. We could add some explicit functions such as dm.randomize_smiles and dm.randomize_selfies.

[niladell]

Yeah. I would say that, even if it's just syntactic sugar, explicitly making it easy to do augmentation could be quite useful.