pygo/gomill/examples/clop_example.ctl

79 lines
1.8 KiB
Plaintext

# The following settings are supported:
#
# - all _common settings_
#
# - all _game settings_
#
# - tuning event settings (cf mcts_tuner):
# - candidate_colour
# - opponent
# - parameters
# - make_candidate
#
# - settings for experiment control
# - parallel -- number of games to run in parallel
# - stop_on_error -- boolean
#
# - regression parameters:
# - clop_H -- float
# - correlations -- 'all' (default) or 'none'
#
## <<
# clop_H: 3 is recommended (it is the default value)
# correlations:
# Even if variables are not correlated "all" should work well. The problem is
# that the regression might become very costly if the number of variables is
# high. So use "correlations none" only if you are certain parameters are
# independent or you have so many variables that "all" is too costly.
## >>
#
# The available parameter types are:
# LinearParameter
# IntegerParameter
# GammaParameter
# IntegerGammaParameter
# For GammaParameter, quadratic regression is performed on log(x)
competition_type = "clop_tuner"
description = """\
Sample control file for CLOP integration.
"""
def gnugo(level):
return Player("gnugo --mode=gtp --chinese-rules --capture-all-dead "
"--level=%d" % level)
def pachi(playouts, policy):
return Player(
"~/src/pachi/pachi "
"-d 0 " # silence stderr
"-t =%d "
"threads=1,max_tree_size=2048 "
"policy=%s "
% (playouts, policy))
players = {
'gnugo-l7' : gnugo(7),
}
parameters = [
Parameter('equiv_rave',
type = "GammaParameter",
min = 40,
max = 32000),
]
def make_candidate(equiv_rave):
return pachi(2000, policy="ucb1amaf:equiv_rave=%f" % equiv_rave)
board_size = 19
komi = 7.5
opponent = 'gnugo-l7'
candidate_colour = 'w'
parallel = 2