Hi all,
I had a call with a chemical/process development customer (organic chemistry, heavy DOE user) who asked something I couldn't answer with confidence.
They want to close the loop around Bayesian Optimization: JMP suggests the next run or batch, their system runs the experiment automatically, the responses are fed back into the JMP table, and JMP returns the next suggestion. The trigger would come from their own application.
I found the Discovery Summit presentation where JSL closes the loop with a PID controller (Autotuning a PID Controller using Bayesian Optimization and JSL), and I know Python can act as an intermediary, so the concept clearly works. What I'm less sure about is how hard this actually is in practice, and which route is the sensible one to recommend:
- JSL driving the loop, with HTTP requests out to their system
- Python as the integration layer between their backend and JMP
- Something else I haven't thought of
So, I wanted to ask you:
- Has anyone been asked this before, and how did you position it?
- Does anyone have a customer who has actually built something like this, even partially?
- Any sense of the effort involved, especially for programmatically pulling the selected batch out of the platform rather than going manually through Make Table?
- Is JSL the route we should be steering people toward, or is there a cleaner pattern?
I've also asked the customer to send the question to Support, but I also wanted to ask here if you have had any experiences with this.
Thanks,
Manlio
