keyboard_layout_optimizer VS cmaes

Regarding evaluating, comparing and potentially even optimizing layouts, I would recommend my keyboard layout optimizer.

I recently got a new ergonomic keyboard, after just typing on my laptop for years. I've used an ErgoDox before, so I went with the Moonlander. While configuring the keyboard and playing with the location of the spacebar, I realized that putting it on the right hand side resulted in a really bad balance between the two hands – it seemed like dvorak was way more right-hand dominant, and that was confirmed when I started looking into it. Went down the rabbit hole of alternate keyboard layouts, and found this repo: https://github.com/dariogoetz/keyboard_layout_optimizer . You can specify the physical configuration of your keyboard and provide a sample text, and it will generate layouts for you, using simulated annealing or a genetic algorithm. Here's a screenshot of the generated layout, along with Dvorak on the bottom:

That's very interesting. I tried a genetic algorithm myself for my optimizer. In the end, I removed the crossover part, but mainly because I did not understand, how exactly it worked :)

You can find the code for the evaluator here. Apart from my own set of metrics, it includes the ones used in KLAnext and also the ones from Oxey's layout playground (with some caveats).

Regarding an evaluator with support for multiple layers that can be user-configured, supports "classical" hold layers as well as one-shot layers, and modifiers on "any key" (think home-row mods), I can advertise my own evaluator/optimizer https://github.com/dariogoetz/keyboard_layout_optimizer Saves you some time on a new project ;)

Identified (amongst others), by tweaked version of https://github.com/dariogoetz/keyboard_layout_optimizer

I want to introduce a pet project of mine that I have been working on in the last year - a keyboard layout evaluator/optimizer: https://github.com/dariogoetz/keyboard_layout_optimizer

cmaes is an implementation of the CMA-ES black box numerical optimization algorithm. It's especially suited to solving high-dimension, non-linear, non-convex, ill-conditioned, and/or noisy problems.

If your function is truly black box then you can use a black box optimization algorithms (note that they likely will not be as good since they make less asumptions on the problem) such as the one provided by argmin or my own simplers_optimization (amongst others). The CMA-ES algorithm has a very good reputation for those use cases but I only found one implementation in Rust and it is not on crates.io.