Minimum objective vs Number of function evaluations

I got this graph tuning the SVM with the bayesian optimization method the SVM gave me accuracy 100% but when I look at this graph it shows much differ between Minimum objective vs Number of function evaluations thats mean the optimizer didn't work that good if it didn't work then how it gave accuracy 100%.

could anyone explain the graph for me and what does these differences means and how it affect my system.

Answer 1

[Walter Roberson]

The optimizer observes a 7.5e-3 at roughly 12e7 function evaluations, and records that. But it does not assume that is going to be the best possible, and keeps trying. Part of that involves estimating a objective value in a manner that is different than a thorough evaluation; the estimated objective is plotted.

It is common for this kind of work that estimates would increase at some point: it is common that there is some kind of "barrier" that needs to be overcome. Like needing to climb to the lip of the current butte in order to find the canyon.

[Yousif Alaraji]

So in the attached plot do you think that this work in a good manner?