I was talking with someone much more experienced in stats than I am and they suggested the use of Monte Carlo Tree Search for a problem I am facing.
Problem Statement: I am collecting jitter measurements from an IC. there are 4 settings which control jitter. The output jitter is stochastic. So if I try settings [x,y,z,w], I will get a distribution of values. I want to find the optimal settings.
He mentioned I use Monte Carlo Search as it is often used in finding the optimal of a simulation. But from all the documents I have read on MCTS, its used for game AI making decisions on the next best move. If I have a sample space of X settings and a corresponding set of Y points, how would I go about using MCTS in finding the set of settings which produce the statistical minima of my output jitter? I am having a hard time in formulating the problem in my head.
I am also aware Bayesian Optimization can work for this, but apparently MCTS is more optimal for a smaller dataset like this (~300 measurements - ~300 outputs)
Further Points Much of my work is data analysis with large volumes of data where I need to find the optimal settings. This can usually be done visually, but I would like to run a script, within the script make a decision based on the data collected (ie from an optima or some other metric based on the problem), and then run further tests.
So I would like help in pointing me to the right direction in how I would optimize these stochastic processes. If someone gave me a 4 variable function on paper and ask me to optimize, I could do this by hand quiet easily. But I am having a hard time formulating how to approach problems when I collect data and try to find settings which minimize. I typically try and fit a surface to complex data and making a "good guess" on which points are closest to the min/max based on the surface. If someone could point me in the right direction that would be awesome (I also lack the proper jargon so it is difficult to google what I really want)