HACKB

Online judge for black-box optimization

Implement algorithms against hidden objective functions and compete to find the best solution within a limited evaluation budget. Try anything from CMA-ES and Bayesian optimization to your own regression models.

What is black-box optimization?

Finding a good x for a function f(x) whose internals (formula, gradient) are unknown — using only a limited number of evaluations.

x
f(x)
y

You only see y. The function is a black box.

The internals are hidden

No formula, no gradients — only the scalar y = f(x) is returned for each input x.

Each evaluation is expensive

A single evaluation can be a real experiment, a simulation, or a model training run. You must search wisely under a strict budget.

Found everywhere in practice

Hyperparameter tuning, materials and drug discovery, engineering design — black-box optimization shows up across many real domains.

Optimization mode

Right now HACKB ships a single Optimization mode. We plan to introduce additional evaluation modes in the future.

Optimization mode

Find the x that minimizes f(x) within a fixed evaluation budget

Call the worker-provided evaluate(x) and search for the best point with any algorithm of your choice. The number of evaluate() calls per submission is capped per problem.

def solve(problem, evaluate):
    ...

How it works

From signup to scoring in four steps.

  1. 1

    Create an account

    Sign up with your email and a password.

  2. 2

    Pick a contest and a problem

    Open a problem from a public contest and review its details.

  3. 3

    Write submission.py

    Implement a Python function for each problem and submit it.

  4. 4

    See results and standings

    Check your judge results and ranking. Other participants' submissions are also visible.

Ready to take a shot?

Pick a problem from the live contests and submit right away.

Browse contests