CMA-ES Covariance Ellipse

What I worked on

Set up a CMA-ES optimization loop using the cma library. Tried to visualize sampling points, means, and standard deviation ellipses as the search converged.

covariance ellipse in 2D finding optimal weights

What I noticed

• CMA-ES samples solutions within a Gaussian around the mean
• The ellipse shows the covariance spread per iteration
• Changing initial std affects convergence behavior
• Step size controls exploration radius
• Even after convergence, sampling continues for improvement

”Aha” Moment

That step size defines how wide the search distribution is and the ellipse shows the evolving uncertainty.

What still feels messy

n/a

Next step

Create a policy using CMA-ES