Closed lemonsqueeze closed 6 years ago
This PR adds support for long options and adds two new time options which are nice on kgs:
--fuseki-time
specific time settings to use during fuseki-t =sims:max_sims
allow playing up to max_sims playouts if best move is unclearTried to make --help
more helpful. How about something like this:
./pachi: unrecognized option '--help'
Pachi version 11.99 (Genjo-devel)
Usage: ./pachi [OPTIONS] [ENGINE_ARGS]
Options:
-c, --chatfile FILE set kgs chatfile
-d, --debug-level LEVEL set debug level
-D don't log board diagrams
-e, --engine ENGINE select engine: (default uct)
random|replay|montecarlo|uct|distributed|dcnn|patternplay
-f, --fbook FBOOKFILE use opening book
-g, --gtp-port [HOST:]GTP_PORT read gtp commands from network instead of stdin.
listen on given port if HOST not given, otherwise
connect to remote host.
-l, --log-port [HOST:]LOG_PORT log to remote host instead of stderr
-r, --rules RULESET rules to use: (default chinese)
japanese|chinese|aga|new_zealand|simplified_ing
-s, --seed RANDOM_SEED set random seed
-t, --time TIME_SETTINGS force basic time settings (override kgs/gtp time settings)
--fuseki-time TIME_SETTINGS specific time settings to use during fuseki
-u, --unit-test FILE run unit tests
TIME_SETTINGS:
=SIMS fixed number of Monte-Carlo simulations per move
Pachi will play fast on a fast computer, slow on a slow computer,
but strength will remain the same.
=SIMS:MAX_SIMS same but allow playing up-to MAX_SIMS simulations if best move is unclear.
useful to avoid blunders when playing with very low number of simulations.
SECS fixed number of seconds per move
Pachi will spend a little less to allow for network latency and other
unexpected slowdowns. This is the same as one-period japanese byoyomi.
_SECS absolute time: use fixed number of seconds for the whole game
Examples: pachi -t =5000 5000 simulations per move
pachi -t =5000:15000 max 15000 simulations per move
pachi -t 20 20s per move
pachi -t _600 10min game, sudden death
I'm impressed that time settings make such a big difference, -t =5000:15000
is almost half a stone stronger than -t =5000
on kgs while still playing pretty fast on low-end machines. I guess it avoids many blunders.
Hi,
Some time ago i thought it might be good to cleanup the parts of Pachi i was involved with. It started with a few lines here and there, and next thing i knew it was spring cleanup in my tree.
Here's what i got so far: there are a few extra functionality / fixes, but other than that none of the cleanup changes are absolutely necessary. For the most part I'm happy with them though, they do make hacking Pachi more enjoyable =)
Many of the changes i've had in my tree for a long time and didn't notice anything bad so i'm confident with them, but the latest ones haven't been tested extensively yet. So not ready to merge at this point, but comments welcome.
I'll try to split them into different PRs to make it more manageable.