While working on Embla, I let some revisions play constant tournaments using cutechess-cli. I have one landmark; tscp 1.81 (1703 elo) and my "parishilton" bot which tries to play the worst move possible for a position (as an "absolute" 0 value).
Now running this through bayeselo gives a good idea of how things perform compared to other versions but I was hoping to get new insights so I hacked a python script which processes all pgn-files and then emits a graphviz-script as a result. A png-file of it can be viewed here: https://www.vanheusden.com/Embla/revisions-dotty.png. Your browser may stutter for a bit because it is huge.
The numbers beside each arrow mean the win/draw/loose percentage for the games player between the engine left of the arrow and the one on the right (which is the one where to arrow points to).
The colors are an indication of the strength.
Now what does it tell me?
Not much. Nothing at all really.
But it was a fun project and I like looking at graphs.
displaying elo ratings in a more interesting way
Moderators: hgm, Dann Corbit, Harvey Williamson