I have a scheduling model formulation ( experimental setup) that takes in product states as input (sample points) and checks the model status (response) and returns feasible or infeasible.
My plan is to be able to represent the feasible space of the facility with an algebraic predictor and for this, I need to sample effectively. This is basically how I thought of the problem:
- Started with an Initial LHS sample for the whole input space generated 200 samples
- For each sample check if the model is feasible(-1) or infeasible (1).
- With that, we have an initial dataset
- Initiate an adaptive sampling procedure to generate more relevant points.
- Do this until we have a total of 1000 dataset (inputs and responses)
Here is my main problem: the responses are discrete, because of this I find it hard to perform the usual adaptive sampling procedure because they are all suited for cases where the response is continuous.
Are their ways to do adaptive sampling for cases where the response is binary? Also, are their ways to better approach the problem, such that the infeasibilities are described by continuous functions?