gaussian process bayesian optimization visualization.ipynb
- use less samples to find the maximum value in test function(Branin Rosenbrock)
- generate 20 points in x y dimension
- in order to plot the 3D function, use meshgrid to get the points
- means 20*20 = 400 samples are necessary to plot the function
- all data = 400 samples = design_domain + train_set
- use train_set -> GP to get a surrogate model
- use design_design -> BO to find to good point and add it to train_set
- (train_set + good point) + (design_domain - good point) = all_data
- if in some epochs, the good point change little, stop iteration
- the first pic is Branin and the second pic is Rosenbrock function
- the green points are train set
- the gray plane is the true function branin, use 400 points by meshgrid()
- the red plane is the prediction of GP on design_domain(all data train set)
- the red vertical line and the blue star point tell us the next point that should be added
- after some iterations , the predction is close to the true function
- the maximum of acquistion function, actually, iteration8 is the best.
- how to draw the funciton by meshgrid
var ihubo = {
nickName : "Duke",
site : "https://github.com/WholeG/GPBO"
}