Some ideas to better approximate DRA's human-drawn notable maps and find "the frontier."
We currently "push" the points near the frontier of an ensemble, requiring that each mutation improve the (continuous) ratings on one or both dimensions.
Alternatively, we could require that each mutation improve only one dimension while the other stays w/in some threshold (e.g., 1%) of the original value. This is the "pin" functionality that I've implemented. So, this option would pin one dimension and optimize the other. We could push each point in both directions some number of times. By uncoupling the dimensions, we might get higher maxima.
We could also increase the threshold (elasticity) in the pinned dimension, the idea being giving the solver more flexibility to find solutions. For example, instead of a 1% threshold, we could use 5% or even 5 ratings units.
Some ideas to better approximate DRA's human-drawn notable maps and find "the frontier."
We currently "push" the points near the frontier of an ensemble, requiring that each mutation improve the (continuous) ratings on one or both dimensions.
Alternatively, we could require that each mutation improve only one dimension while the other stays w/in some threshold (e.g., 1%) of the original value. This is the "pin" functionality that I've implemented. So, this option would pin one dimension and optimize the other. We could push each point in both directions some number of times. By uncoupling the dimensions, we might get higher maxima.
We could also increase the threshold (elasticity) in the pinned dimension, the idea being giving the solver more flexibility to find solutions. For example, instead of a 1% threshold, we could use 5% or even 5 ratings units.