figure 3 clarification from publication

[jessielyons]

Dear Authors,

Very nice publication.

1. Can you please clarify how you plot the figure 3? Is training loss vs epoch plotted from dreaming or training. Since you mentioned target logp in figure as 3 or -3, i am guessing its from dreaming step.

2. Can you please also clarify what is the meaning of logP vs epochs in figure 3? Is it from dreaming step? Can you share what are plotting from outputs generated? Are you using log file of the run or using "sampled_intermediate_mol" file for a particular target logp?

Any related code on how to generate Figure 3 can help.

Thanks so much,
JL

[MarioKrenn6240]

ad1) yes, loss is from the dreaming process

ad2) logP is computed via RDKit (its for testing the ground truth of our system, and thats why its not a smooth function because we use onehot). Loss is the evolution of the loss during dreaming, which is the MSE between model prediction and target property.

The code for this is basically written in demo.py, let me know if this helps.

[MarioKrenn6240]

Sure, you can just delete these lines, and it will work straightforwardly. The main idea is between line 283 and 300, so just go ahead and remove the rest and adjust the output of the dream function.

[jessielyons]

Awesome, thank you so much for all your quick and helpful replies! I will try accordingly. Best, JL

…

On Sun, Jun 18, 2023 at 7:59 AM Mario Krenn ***@***.***> wrote: Sure, you can just delete these lines, and it will work straightforwardly. The main idea is between line 283 and 300, so just go ahead and remove the rest and adjust the output of the dream function. — Reply to this email directly, view it on GitHub <#6 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AU45PQYH52GOZLELS3BA22LXL4JUBANCNFSM6AAAAAAYCPO3YE> . You are receiving this because you authored the thread.Message ID: ***@***.***>