Question about tensorboard scores

[Horizon371]

Hello,

I am running reinvent with only one scoring component and Scaffold DF. I have a couple of questions regarding the tensorboard output.

1. Why would it be that the average score increases while the score of the component decreases?
2. Why do the raw and normal scores of the component differ since I am not using any transformations on it?

Best regards,
Cristian

[halx]

Hi,

the transformed score takes the duplicates into account and so will be smaller than the raw scores. You are producing quite a bit of duplicatres, in fact, the number is increasing over your run. You also have a scaffold filter which does 2 things: 1) score compounds as zero as soon as its scaffold has been found more than N times,, 2) score every SMILES that is found at least a second time as zero. It follows that the total score can behave quite differently from the single component score.

Hope that helps,
Hannes.

[halx]

Corrections: the transformed score is a product (logical and) of the raw score, the invalid and the duplicate mask.