Unnecessary iterations in parameter optimization loop with Hillclimbing method

KNIME Analytics Platform
1

I observed a strange behavior of the Parameter Optimization Loop. When using Hillclimbing with only one Parameter and step size = 0.01, it searches for the optimum like this:

(i) 0.5 -> look at the neighbors 0.49 and 0.51 -> best value is found at 0.49

(ii) look at the neighbors 0.48 and 0.5 -> best value is found at 0.48

(iii) look at the neighbors 0.47 and 0.49 -> and so on...

So why each point gets inspected twice?

 

Edit: I use KNIME Analytics 3.4.1

2

Hi, 

the algorithm is greedy and does not know about previous iterations. But you are right, for one dimensional feature optimization this would make quite a speed up. Let me open a ticket for you.

Cheers, Iris

3

Thanks a lot, Iris!