# Config for the webserver to deploy for the distributed training run.
# Regular users should ignore this config.

# Match settings and logging--------------------------------------------------------------------

logSearchInfo = false
logMoves = false
maxMovesPerGame = 1600

# Fancy game selfplay settings--------------------------------------------------------------------

initGamesWithPolicy = true
policyInitAreaProp = 0.09
startPosesPolicyInitAreaProp = 0.0
compensateAfterPolicyInitProb = 0.55
forkSidePositionProb = 0.020

earlyForkGameProb = 0.04
earlyForkGameExpectedMoveProp = 0.025
forkGameProb = 0.01
forkGameMinChoices = 3
earlyForkGameMaxChoices = 12
forkGameMaxChoices = 36
sekiForkHackProb = 0.01

cheapSearchProb = 0.75
cheapSearchVisits = 240
cheapSearchTargetWeight = 0.0

reduceVisits = true
reduceVisitsThreshold = 0.9
reduceVisitsThresholdLookback = 3
reducedVisitsMin = 240
reducedVisitsWeight = 0.1

handicapAsymmetricPlayoutProb = 0.5
normalAsymmetricPlayoutProb = 0.01
maxAsymmetricRatio = 8.0
minAsymmetricCompensateKomiProb = 0.4

policySurpriseDataWeight = 0.5
valueSurpriseDataWeight = 0.1

scaleDataWeight = 1.0

estimateLeadProb = 0.20
estimateLeadVisits = 10
fancyKomiVarying = true

# Rules------------------------------------------------------------------------------------

koRules = SIMPLE,SIMPLE,POSITIONAL,SITUATIONAL
scoringRules = AREA,AREA,AREA,TERRITORY
taxRules = NONE,NONE,NONE,NONE,NONE,NONE,SEKI,SEKI,ALL
multiStoneSuicideLegals = false,false,true
hasButtons = false,true,true

bSizes = 7,9,11,13,15,17,19,  8,10,12,14,16,18
bSizeRelProbs = 1,4,3,10,13,17,200, 1,2,4,8,15,20
allowRectangleProb = 0.10

komiAuto = True
komiStdev = 1.0
handicapProb = 0.05
handicapCompensateKomiProb = 0.60
forkCompensateKomiProb = 0.80
sgfCompensateKomiProb = 0.85
komiBigStdevProb = 0.05
komiBigStdev = 12.0
komiBiggerStdevProb = 0.005
komiBiggerStdev = 45.0
handicapKomiInterpZeroProb = 0.05
sgfKomiInterpZeroProb = 0.15

drawRandRadius = 0.5
noResultStdev = 0.166666666

# Search limits-----------------------------------------------------------------------------------

maxVisits = 1200
numSearchThreads = 1

# Root move selection and biases------------------------------------------------------------------------------

chosenMoveTemperatureEarly = 0.75
chosenMoveTemperatureHalflife = 19
chosenMoveTemperature = 0.15
chosenMoveSubtract = 0
chosenMovePrune = 1

rootNoiseEnabled = true
rootDirichletNoiseTotalConcentration = 10.83
rootDirichletNoiseWeight = 0.25

rootDesiredPerChildVisitsCoeff = 2
rootNumSymmetriesToSample = 4

useLcbForSelection = true
lcbStdevs = 5.0
minVisitPropForLCB = 0.15

# Internal params------------------------------------------------------------------------------

winLossUtilityFactor = 1.0
staticScoreUtilityFactor = 0.05
dynamicScoreUtilityFactor = 0.30
dynamicScoreCenterZeroWeight = 0.25
dynamicScoreCenterScale = 0.50
noResultUtilityForWhite = 0.0
drawEquivalentWinsForWhite = 0.5

rootEndingBonusPoints = 0.5
rootPruneUselessMoves = true

rootPolicyTemperatureEarly = 1.55
rootPolicyTemperature = 1.1

cpuctExploration = 1.05
cpuctExplorationLog = 0.18
fpuReductionMax = 0.2
rootFpuReductionMax = 0.0
valueWeightExponent = 0.5

subtreeValueBiasFactor = 0.30
subtreeValueBiasWeightExponent = 0.8

useNonBuggyLcb = true
useGraphSearch = true

fpuParentWeightByVisitedPolicy = true
fpuParentWeightByVisitedPolicyPow = 2.0

mutexPoolSize = 64
numVirtualLossesPerThread = 1