2021-09-03 BespokeFit meeting notes

Previous notes

• Previous discussion at the bottom of 2021-09-01 Benchmarking followup workshop discussion notes

Workshop content

• JH – I’ve drafted a notebook:

•  

• JW – This looks really good. A few notes:

  • It’d be good to make a big deal about the automatic detection of useful data from QCA.

  • Also automatic generation of data if it’s not in QCArchive

  •  

• JW will try visualizing QCA dataset and showing which fragments are finally accepted for the training data.

• JW will test out installing+running today.

• PB will test env creation and notebook running on linux

• JH – Could show final fit from outputted forcebalance PDF

  • (Jupyter notebook PDF integration code): python - How to embed pdfs files in jupyter notebook on Github - STACKOOM

  •  

• Should we include XTB example?

  • PB + SB – Would be cool. We could clarify that this would be a precursor to using ANI or another ML method.

  • JH – Runtime for XTB should be just a few minutes, but fitting could take 10-15 minutes for QCA molecule. So let’s start the session by having people open the notebook and execute all.

  •  

• Slides

  • JH will share google slides, JW+PB+SB will review and comment

  • (General) – Add slide saying “We actually used this to fit Sage. Wow!”

• JH – Value in putting this on binder as well?

  • JW – I think it’s better to get things running locally if possible – It shows to the audience that it’s a bit more stability/deployability

  •  

Workshop schedule

• 10 min: Notebook setup and start running

• 20 min: Talk

• 30 min: Notebook walkthrough

• Up to 30 mins: followup, futzing around, and Q&A

• JW will run Zoom call, manage recording, moderate waiting room

• JW + KCJ will manage communication, invites, and file distribution.