Manual Numerical Optimization Competition
This is a fun competition! Participation takes approximately ten minutes.
What's the Challenge?
You call yourself an optimization expert? Well... really? I am talking about you! Not about your algorithm. Here you can prove what you're made of!
Did you ever wonder how it feels to be a black-box optimizer? What it means to be lost in the vastness of space, wandering alone under search points, and to feel the sparseness of information on your skin? Do you have the courage to explore unknown landscapes? And do you have what it takes to hunt down an optimum? Then join the competition today!
Yes, you will optimize an actual black-box with your own hands. We send you out to locate the best search point you can find. You get only one chance. Trust your intuition. This is about you, not about uncaring algorithms.
How does it work?
You will optimize a black-box objective function
We'll drop you into an unknown landscape, with only four evaluated points. It is up to you to place your crosshair at the next point to be evaluated, and to submit it to the server for evaluation. A split second later, your bullet hits its target and you obtain a new search point, including its rank. Keep moving until you have used up your entire budget of 100 function evaluations.
How to run a competition?
You can run your own competition by filling the following fields.
| name of the competition | |
| opening date/time | |
| closing date/time | |
| objective function |
The objective function (to be minimized) is defined as a formula depending on the variables x and y. A formula consist of numbers, constants (pi, e), variables (x, y), basic functions (sign, sin, cos, sqrt, abs, exp, log, min, max), arithmetic operators +, -, *, / and ^ (power), and parentheses (...) or [...]. Examples:
1 + 1/2 + 1/3[sin(x)+1]^2 - [cos(y)^2 / 4]^3abs(y * e^(-2 * pi * x))(1-x)^2 + 10 * [y - sin(2.4 * x)]^2
Note that multiplication must be written as 7*x, not as 7x.
Important: Make sure that the smallest function value is zero, or very slightly positive. Do not place the optimum too close to x=-2 and y=2 because that is the starting point provided to users. Note that each user receives a transformed (shifted, rotated and scaled) version of the objective.