In PhilippScheller/VisBayesOpt: Visualizing model based optimization runs from package mlrMBO

MboPlotProgress

Description
The plot shows the cumulative minimum (maximum) value of the objective function evaluations $f(x)$ at design points $x \in D$ after $n$ iterations, leading to a monotone decreasing (increasing) function for a minimization (maximization) problem.

Interpretation
* If the graph shows a decrease (increase), the optimization of the minimization (maximization) problem improves.
* If the graphs stays (more or less) constant, the optimizer might have converged.

How to use
* If the plot still shows major (no) improvement at larger $n$ one may adjust the constraints of the optimization to grant a larger (lower) number of maximum iterations (to save cost).
* If the plot still shows a steep improvement for $n \to n_{max}$, increasing the maximum number of iterations could yield further gain in optimization performance.

PhilippScheller/VisBayesOpt documentation built on Sept. 14, 2020, 12:47 p.m.