This repo contains:
- a script to calculate SPRT results
- derivation/explanations of
All of this is, of course, in the context of chess. Some familiarity with what hypothesis testing does is assumed for the explanantions.
Run python3 sprt.py --help
for a full list of options.
SPRT:
python3 sprt.py --wins 4307 --losses 4063 --draws 8620 --elo0 0 --elo1 3
> LLR: 2.47 (-2.94, 2.94)
> Continue Playing
GSPRT:
python3 sprt.py --wins 4307 --losses 4063 --draws 8620 --elo0 0 --elo1 3 --gsprt
> LLR: 2.99 (-2.94, 2.94)
> H1 Accepted
There are additional options for alpha
and beta
, and by default the
calculation uses OpenBench's method, but passing the --cutechess
option
will calculate the CuteChess result instead (only for normal SPRT).
If you don't pass in certain arguments, here are the defaults:
Argument | Default |
---|---|
wins | 0 |
losses | 0 |
draws | 0 |
elo0 | 0 |
elo1 | 5 |
alpha | 0.05 |
beta | 0.05 |
cutechess | false |
gsprt | false |