Hi All,
This example is built off of a recent conversation from a customer in the chemical industry. They create a lot of mixture/formulations, but in their examples they are doing mixtures with hard constraint rules, such as "we have 5 possible additives, but can only use 2 at a time and need to find the best settings".
This demo uses Bayesian Optimisation to show how this process can be done with Bayesian Optimisation and candidate sets. The initial data is generated with a custom design from the candidate dataset, and BayesOpt is ran against the candidate set to only suggets runs with 2 out of 5 additives used. It also has a nice pitch for combining LLMs to solve a challenge with scripting.
How to use:
1. Open up the "dt" button
2. Hit 'Candidate Set Generation' to create the reference-able candidate data set
3. Hit 'Run Bayes Opt' - this will prompt you to select the candidate set, and then produce the next rows.
4. Keep going until you reach 'Repeat Best' stage and voila.
If you're asking about how candidate sets are made, rather than showing them the script (and terrifying them), you can leverage LLMs, and the fact that you can transform natural language into the script for them
1. Hit the "LearnBot Prompt for Candidate Set", this will open up LearnBot and copy a prompt to your clipboard that you can paste in.
2. Run the script to generate the candidate set. This is working well (for now), the customer only needs to change the second paragraph to describe their experiments.
3. The 'Another LearnBot Candidate Set Prompt' is a more complex example from Bradley Bowens GTET presentation/journal.
