Closed amineebenamor closed 5 years ago
The more robust function is edit_mol
in rexgen_direct/scripts/eval_by_smiles.py
. This version applies some additional pre- and post-fixes so that the graph edit representation of reactions is higher fidelity (i.e., is able to describe a larger proportion of training examples). The function that is used in mol_graph_direct_useScores.py
does not contain all of these fixes and serves only as a quick check to see if the true product SMILES is already contained in the list of candidates to avoid having excessive numbers of duplicate candidates during training.
Hello, You said in mol_graph_direct_useScores.py that: "note: get_product_smiles is HIGHLY imperfect, but that's not a huge deal. training tries to pick the right bonds. The evaluation script has a more robust function to get product_smiles." Could you explain why get_product_smiles is imperfect? I tried to find the more robust function to get_product_smiles but I didn't find it in your code. Could you tell me where it is? Thank you for your help!