def main():
searchParams= MCTSPARAMS.MCTSPARAMS()#MCTS::PARAMS
expParams= EXPERIMENTPARAMS.EXPERIMENTPARAMS()# EXPERIMENT::PARAMS
knowledge= KNOWLEDGE.KNOWLEDGE()# SIMULATOR::KNOWLEDGE
problem=""
outputfile=""
policy=""#
size=0
number=0
treeknowledge = 1
rolloutknowledge = 1
smarttreecount = 10# int
smarttreevalue = 1.0#
#real =NETWORK.NETWORK(size, number)#
#simulator =NETWORK.NETWORK(size, number)#
#
file="C:\projs\\40_weka_randomForest9.txt"
ei= .sfl.Diagnoser.diagnoserUtils.readPlanningFile(file)
real = DIAGNOSER.DIAGNOSER(ei, 0.7)#
simulator = DIAGNOSER.DIAGNOSER(ei.Copy(), 0.7)#
simulator.SetKnowledge(knowledge)#
experiment = EXPERIMENT.EXPERIMENT(real, simulator, outputfile, expParams, searchParams)#EXPERIMENT
experiment.DiscountedReturn()#
return 0#
def UnitTestSearch(depth):
testSimulator = TEST_SIMULATOR.TEST_SIMULATOR(3, 2, depth)
params = MCTSPARAMS.MCTSPARAMS()
params.MaxDepth = depth + 1
params.NumSimulations = pow(10, depth + 1)
mcts = MCTS(testSimulator, params)
mcts.UCTSearch()
rootValue = mcts.Root.Value.GetValue()
optimalValue = testSimulator.OptimalValue()
assert (abs(optimalValue - rootValue) < 0.1)
def main(ei):
searchParams= MCTSPARAMS.MCTSPARAMS()#MCTS::PARAMS
expParams= EXPERIMENTPARAMS.EXPERIMENTPARAMS()# EXPERIMENT::PARAMS
print "start", ei.calc_precision_recall()
real = DIAGNOSER.DIAGNOSER(ei, 0.6)#
simulator = DIAGNOSER.DIAGNOSER(ei.Copy(), 0.6)#
experiment = EXPERIMENT.EXPERIMENT(real, simulator, expParams, searchParams)#EXPERIMENT
print "running"
return experiment.RunMultiple()#
def UnitTestRollout():
testSimulator = TEST_SIMULATOR.TEST_SIMULATOR(2, 2, 0)
params = MCTSPARAMS.MCTSPARAMS()
params.NumSimulations = 1000
params.MaxDepth = 10
mcts = MCTS(testSimulator, params)
totalReward = 0.0
for n in range(mcts.Params.NumSimulations):
state = testSimulator.CreateStartState()
mcts.TreeDepth = 0
totalReward += mcts.Rollout(state)
rootValue = totalReward / mcts.Params.NumSimulations
meanValue = testSimulator.MeanValue()
assert (abs(meanValue - rootValue) < 0.1)
def UnitTestGreedy():
testSimulator = TEST_SIMULATOR.TEST_SIMULATOR(5, 5, 0)
params = MCTSPARAMS.MCTSPARAMS()
mcts = MCTS(testSimulator, params)
numAct = testSimulator.GetNumActions()
numObs = testSimulator.GetNumObservations()
vnode = mcts.ExpandNode(testSimulator.CreateStartState())
vnode.Value.Set(1, 0)
vnode.Child(0).Value.Set(0, 1)
for action in range(1, numAct):
vnode.Child(action).Value.Set(0, 0)
x = mcts.GreedyUCB(vnode, False)
print "x:", x
assert (mcts.GreedyUCB(vnode, False) == 0)
def UnitTestUCB():
testSimulator = TEST_SIMULATOR.TEST_SIMULATOR(5, 5, 0)
params = MCTSPARAMS.MCTSPARAMS()
mcts = MCTS(testSimulator, params)
numAct = testSimulator.GetNumActions()
numObs = testSimulator.GetNumObservations()
#// With equal value, action with lowest count is selected
vnode1 = mcts.ExpandNode(testSimulator.CreateStartState())
vnode1.Value.Set(1, 0)
for action in range(numAct):
if (action == 3):
vnode1.Child(action).Value.Set(99, 0)
else:
vnode1.Child(action).Value.Set(100 + action, 0)
assert (mcts.GreedyUCB(vnode1, True) == 3)
#// With high counts, action with highest value is selected
vnode2 = mcts.ExpandNode(testSimulator.CreateStartState())
vnode2.Value.Set(1, 0)
for action in range(numAct):
if (action == 3):
vnode2.Child(action).Value.Set(99 + numObs, 1)
else:
vnode2.Child(action).Value.Set(100 + numAct - action, 0)
assert (mcts.GreedyUCB(vnode2, True) == 3)
#// Action with low value and low count beats actions with high counts
vnode3 = mcts.ExpandNode(testSimulator.CreateStartState())
vnode3.Value.Set(1, 0)
for action in range(numAct):
if (action == 3):
vnode3.Child(action).Value.Set(1, 1)
else:
vnode3.Child(action).Value.Set(100 + action, 1)
assert (mcts.GreedyUCB(vnode3, True) == 3)
#// Actions with zero count is always selected
vnode4 = mcts.ExpandNode(testSimulator.CreateStartState())
vnode4.Value.Set(1, 0)
for action in range(numAct):
if (action == 3):
vnode4.Child(action).Value.Set(0, 0)
else:
vnode4.Child(action).Value.Set(1, 1)
assert (mcts.GreedyUCB(vnode4, True) == 3)