def community_test(DAG,k,threshold):
communities = comm.community(DAG, k, threshold)
census = []
#creates a list, 'census', where each item is a list of the nodes in a certain community
for community in communities:
comm_members = []
for box in box_list:
for node in box.nodes():
if box.node[node]['community'] == community:
comm_members.append(node)
census.append(comm_members)
success = 0. #initialise the 'success' variable
#iterate across each list of community members (the items in 'census')
for comm_members in census:
for box in box_list:
nodes = box.nodes()
#for a given community, count how many of the nodes come from each box...
counter = 0.
for node in nodes:
if node in comm_members:
counter += 1.
#If over half of the nodes in a given community come from the same box, these nodes are considered to have been successfully placed together...
#...whilst the rest are unsuccessfully placed in a community
if len(comm_members) > 0:
fraction = counter/float(len(comm_members)) #calculates the fraction of a community that come from a single box
if fraction > 0.5:
success += counter/(len(box_list)*len(nodes)) #adds to the success score the RHS, which is the fraction of nodes deemed successful from this community
return success
def simulate(c=None, T=80):
if c is None:
c = community(300, 3, 10)
Transfer = transfer(c)
# Data to store c.allIdeas, to then make animation
dataX = np.zeros((T, c.numberMembers))
dataY = np.zeros((T, c.numberMembers))
dataZ = np.zeros((T, c.numberMembers))
posX = np.zeros((T, c.numberMembers))
posY = np.zeros((T, c.numberMembers))
ideaDist = np.zeros((T, c.domainSize * 2 + 1))
# Iterate the idea transfer throughout community
for t in range(0, T):
np.random.seed()
dataX[t, :] = c.allIdeas[:, 0]
dataY[t, :] = c.allIdeas[:, 1]
dataZ[t, :] = c.allIdeas[:, 2]
posX[t, :] = c.allPositions[:, 0]
posY[t, :] = c.allPositions[:, 1]
ideaDist[t, :] = c.ideaDistribution[0][0]
#Transfer.deterministicMerge()
Transfer.probabilisticMerge()
data = [dataX, dataY, dataZ, posX, posY, ideaDist]
return data
def main():
search_conf_file = sys.argv[1]
original_conf_file = sys.argv[2]
partition_conf_file = sys.argv[3]
#partition_conf_file = "partition.json"
#
# get original config and score
#
original_conf = json.loads(open(original_conf_file, 'r').read())
#
# hierarchically type tuning
#
partition_tree = json.loads(open(partition_conf_file, 'r').read())
level = len(partition_tree) - 1
while level >= 0:
search_conf = json.loads(open(search_conf_file, 'r').read())
change_items = search_conf['config']
# create communities based on the current level of the partition tree
global community_obj
community_obj = com.community(partition_tree, change_items, level)
if __debug__:
fp = open('ivy.log', 'a')
print >> fp, "----level:", level
print >> fp, "community configs:"
print >> fp, "lens:", len(community_obj.confs)
print >> fp, community_obj.confs
fp.close()
level -= 1
def ETLNAUTIA(communityType):
data = DataNAUTIA(communityType)
com.community(communityType)
camp.camp(data.getBibliography(), data.getEntities())
ctr.country(data.getBibliography())
cd.campData(data.getBibliography(), data.getEntities(),
data.getLocalLeaders())
gd.generalData(data.getBibliography())
se.socioEconomic(data)
g.government(data.getBibliography())
fa.fisicoAmbiental(data.getBibliography(), data.getEntities())
u.urbanism(data.getEntities(), data.getGeneralForm(),
data.getPublicSpace())
inf.infrastructure(data)
s.services(data)
sh.shelter(data.getEntities(), data.getShelter(), data.getHouseHold())
fs.foodSecurity(data)
def simulate(rl, numTrials=1, maxIterations=100, verbose=False, sort=False):
T = maxIterations
totalRewards = [] # The rewards we get on each trial
dataX = np.zeros((T, rl.c.numberMembers))
dataY = np.zeros((T, rl.c.numberMembers))
dataZ = np.zeros((T, rl.c.numberMembers))
posX = np.zeros((T, rl.c.numberMembers))
posY = np.zeros((T, rl.c.numberMembers))
ideaDist = np.zeros((T, rl.c.domainSize * 2 + 1))
for trial in range(numTrials):
# Form a brand new community
c = community(rl.c.numberMembers, rl.c.numberIdeas, rl.c.domainSize)
# Form a brand new agent
agentMDP = mdp.agentMDP(c)
agentMDP.tau = rl.tau
# Update QLearning community to new community
rl.c = agentMDP.c
rl.index = agentMDP.index
Transfer = transfer(c)
state = agentMDP.startState()
sequence = [state]
totalDiscount = 1
totalReward = 0
for t in range(maxIterations):
action = rl.getAction(state)
transitions = agentMDP.succAndProbReward(state, action)
if trial == numTrials - 1:
dataX[t, :] = c.allIdeas[:, 0]
dataY[t, :] = c.allIdeas[:, 1]
dataZ[t, :] = c.allIdeas[:, 2]
posX[t, :] = c.allPositions[:, 0]
posY[t, :] = c.allPositions[:, 1]
ideaDist[t, :] = c.ideaDistribution[0][0]
i = sample([prob for newState, prob, reward in transitions])
newState, prob, reward = transitions[i]
for n in range(agentMDP.tau):
Transfer.probabilisticMerge()
newState = agentMDP.observeCommunity()
totalReward += totalDiscount * reward
totalDiscount *= agentMDP.discount()
state = newState
if verbose:
print(("Trial %d (totalReward = %s): %s" %
(trial, totalReward, sequence)))
totalRewards.append(totalReward)
data = [dataX, dataY, dataZ, posX, posY, ideaDist, totalRewards]
return data
def detect(filename, nop = -1, debug = False):
comdet_tmppath = 'comdet_tmp'
fileinput = open(filename, 'r')
lines = fileinput.readlines()
p1 = wrapper.Pool()
for line in lines:
line = line.strip()
if not line:
continue
split = line.split(' ')
n1 = p1.get_node(split[0])
if not n1:
n1 = p1.add_node(split[0])
n1.add_elem(split[1])
if os.path.exists(comdet_tmppath):
shutil.rmtree(comdet_tmppath)
os.mkdir(comdet_tmppath)
os.chdir(comdet_tmppath)
com_t = community.community()
result = com_t.start(p1, nop, debug)
os.chdir('..')
if os.path.exists(comdet_tmppath):
shutil.rmtree(comdet_tmppath)
return result
def detect(filename, nop=-1, debug=False):
comdet_tmppath = 'comdet_tmp'
fileinput = open(filename, 'r')
lines = fileinput.readlines()
p1 = wrapper.Pool()
for line in lines:
line = line.strip()
if not line:
continue
split = line.split(' ')
n1 = p1.get_node(split[0])
if not n1:
n1 = p1.add_node(split[0])
n1.add_elem(split[1])
if os.path.exists(comdet_tmppath):
shutil.rmtree(comdet_tmppath)
os.mkdir(comdet_tmppath)
os.chdir(comdet_tmppath)
com_t = community.community()
result = com_t.start(p1, nop, debug)
os.chdir('..')
if os.path.exists(comdet_tmppath):
shutil.rmtree(comdet_tmppath)
return result
def main():
bitcode = sys.argv[1]
search_conf_file = sys.argv[2]
original_conf_file = sys.argv[3]
partition_conf_file = sys.argv[4]
#partition_conf_file = "partition.json"
#
# delete log file if exists
#
try:
os.remove("log.dd")
except OSError:
pass
if __debug__:
try:
os.remove("ivy.log")
except OSError:
pass
#
# get original config and score
#
original_conf = json.loads(open(original_conf_file, 'r').read())
run_config(original_conf, original_conf, bitcode, 0)
original_score = utilities.get_dynamic_score()
if __debug__:
fp = open('ivy.log', 'a')
print >> fp, "---------"
print >> fp, "original score"
print >> fp, original_score
fp.close()
#
# hierarchically type tuning
#
partition_tree = json.loads(open(partition_conf_file, 'r').read())
curr_conf = None
curr_score = -1
level = len(partition_tree) - 1
while level >= 0:
if __debug__:
fp = open('ivy.log', 'a')
print >> fp, "partition level and config"
print >> fp, level, partition_tree[level]
fp.close()
search_conf = json.loads(open(search_conf_file, 'r').read())
change_items = search_conf['config']
# delete from the search space the configs that have been pruned
if curr_conf != None:
reach_bottom_label = 1
for i in range(len(curr_conf['config'])):
curr_type = curr_conf['config'][i].values()[0]["type"]
type_vector = change_items[i].values()[0]["type"]
index = type_vector.index(curr_type)
if index != 0:
reach_bottom_label = 0
del type_vector[index + 1:]
if change_items[i].keys()[0] == "call":
del change_items[i].values()[0]["switch"][index + 1:]
if reach_bottom_label == 1:
break
# create communities based on the current level of the partition tree
global community_obj
community_obj = com.community(partition_tree, change_items, level)
if __debug__:
fp = open('ivy.log', 'a')
print >> fp, "community configs:"
print >> fp, community_obj.confs
fp.close()
#
# record the change set
#
search_changes = community_obj.confs
change_set = []
type_set = []
switch_set = []
for i in range(len(search_changes)):
type_vector = search_changes[i]["type"]
if isinstance(type_vector, list):
if len(type_vector) == 1:
tmp_set = []
tmp_type_set = []
tmp_switch_set = []
tmp_set.append(search_changes[i])
tmp_type_set.append(type_vector)
if "switch" in community_obj.get_vars(
search_changes[i]["name"])[0].keys():
tmp_switch_set.append(
community_obj.get_vars(
search_changes[i]["name"])[0]["switch"])
else:
tmp_switch_set.append([])
to_highest_precision(tmp_set, tmp_type_set, tmp_switch_set)
else:
type_set.append(type_vector)
change_set.append(search_changes[i])
if (len(community_obj.get_vars(search_changes[i]["name"]))
== 1) & ("switch" in community_obj.get_vars(
search_changes[i]["name"])[0].keys()):
switch_set.append(
community_obj.get_vars(
search_changes[i]["name"])[0]["switch"])
else:
switch_set.append([])
#
# search for valid configuration
#
print "Searching for valid configuration using delta-debugging algorithm ..."
# get current score
if curr_score == -1:
curr_score = original_score ## 0.95
curr_conf = original_conf
# keep searching while the type set is not searched throughout
while not is_empty(type_set):
search_config(change_set, type_set, switch_set, search_conf,
original_conf, bitcode, original_score)
# get the score of modified program
run_config(search_conf, original_conf, bitcode, 0)
modified_score = utilities.get_dynamic_score()
if modified_score <= curr_score:
curr_conf = search_conf
curr_score = modified_score
if __debug__:
fp = open('ivy.log', 'a')
print >> fp, "---------"
print >> fp, "updating new score"
print >> fp, curr_score
fp.close()
# print the intermediate level configuration
utilities.print_config(
curr_conf,
"dd2_valid_level" + str(level) + "_" + bitcode + ".json")
fleveline = open("log.dd", "a")
fleveline.write(
"----------------------------------------------------------------------\n"
)
fleveline.close()
# check the modified configuration
'''
if is_bottom_type(change_set):
if __debug__ :
fp = open('ivy.log', 'a')
print>>fp, "stop searching : reach bottom types"
print>>fp, search_conf
print>>fp, "change set:"
print>>fp, change_set
fp.close()
break
'''
level -= 1
# print tuning result
if (curr_score <= original_score) & (curr_conf != None):
#print valid configuration file and diff file
diff = utilities.print_diff(curr_conf, original_conf,
"dd2_diff_" + bitcode + ".json")
if diff:
utilities.print_config(curr_conf, "dd2_valid_" + bitcode + ".json")
print "original_score: ", original_score
print "modified_score: ", curr_score
fp = open('time.txt', 'a')
print >> fp, curr_score, "/", original_score
fp.close()
print "Check valid_" + bitcode + ".json for the valid configuration file"
return
print "No configuration is found!"
from community import community
from transfer import transfer
import mdp, util, utilRL, copy, time, cloudpickle
import numpy as np
############################# Choose a simulation #############################
from simulateRL import simulate
#from controlSimulate import simulate
startTime = time.time()
######################### Take care of filename (pkl extension added later) ##########################
fn = 'ag_dist_rad_gre'
######################## community parameters ##############################
c = community(100, 4, 10)
c.allRadii = np.random.normal(0.4, 0.1, c.numberMembers)
######################### agent parameters ############################
agentMDP = mdp.agentMDP(community=c, index=0, sampleSize=4)
agentMDP.T = 100
agentMDP.tau = 10
rl = mdp.QLearningAlgorithm(agentMDP.index,
agentMDP.tau,
agentMDP.c,
agentMDP.actions,
agentMDP.discount(),
mdp.communityFeatureExtractor,
explorationProb=0.2,
alpha=0.1)