Performance Improvement for REINVENT4 Priors: libinvent and mol2mol

[natarajbalakrishnan]

Hi,

I have a question regarding the performance of the REINVENT4 priors. I've noticed that under the same basic settings, such as the number of epochs and the number of molecule samples per epoch (set to 300 and 128, respectively), reinvent.prior runs faster compared to other priors like libinvent and all mol2mol prior subtypes. Is there a way to improve the performance of the priors that are lagging?

Regards,
Nataraj

[halx]

Hi,

many thanks for your interest in REINVENT and welcome to the community!

The prior networks have different complexities. I believe the network architecture is printed into the log file for the user to see. So it is not surprising that performance varies. Transformers are the most complex.

Many thanks,
Hannes.