I think i introduced a bug somewhere that produces bad results. It worked with a previous commit.
Configure the main parameters directly in the main.rs file. Select your layout with the first generic parameter. The second ist the number of keys and must match the Layout, otherwise you will get a compiler error.
To adjust the training set update the training_set_path in main.rs.
Advanced setup can be found in the advanced_setup.rs file.
temperature = 500
epoch = 20
coolingRate = 0.99
num_iterations = 2500062.265114 seconds (26.52 M allocations: 1.370 GiB, 0.35% gc time, 1.85% compilation time: 3% of which was recompilation)
Duration: PT27.854251S
With even more optimizations in the implementation:
Duration: PT19.896563S
I added support for the german keys with the QWERTZ layout.
Here is an example layout i got:
original readme:
Supporting simulated annealing code for the Why I Made The World's Worst Keyboard YouTube video.
Written in Julia... because it's fast, easy to read, and annoys my labmates.
To run, clone the this repository and start Julia with
julia --project=. and run include("keyboardSA.jl").
It should start by benchmarking your training data (myBook.txt)
against QWERTY followed by building it's own optimal layout.
Change the number of iterations and cooling rates as
desired within the data block at the beginning of the file.
The terminal will give some indication of current progress
(also stored by a new text file will give a iteration-by-iteration record of progress),
and .png files of the current best solution will be saved to your same directory.
To train on your own custom dataset either point the "myBook.txt" somewhere else or just replace its contents.
Good luck!
Download at https://julialang.org/downloads/ and install the Julia language.
Assuming julia is in your path, run
git clone https://github.com/AtomicFrontierCode/keyboards.git
cd keyboards
julia -L 'keyboardSA.jl'If you want to play with the script, it's recommended to use the Revise.jl package to minimize latency.