Understanding of the staged-learning results

[GeYan489]

As for the results of staged_learning, I would like to inquire if my understanding of the following is correct.
1、The gent/Prior/Target values represent the possibility of obtaining this SMILES. The lower the value, the greater the probability
2、I created new molecules through mol2mol.smi. So the high "score" value represent that, from the perspective of considering the scoring function, the difference between output smiles and input smiles is smaller

[halx]

Hi,

The prior and agent negative log likelihoods (NLL) do indeed quantify the likelihood of producing that specific sequence. The lower the value the more likely this sequence gets sampled. Note that these NLL values are positive. The augmented (target) NLL is he sum of the prior and the scaled score, see paper.

I am not sure what question 2 asks for- With the mol2mol generator you would be generating output SMILES similar to the input SMILES with the aim to generate output SMILES with high score (and thus low NLL).

HTH,
Hannes.