def __init__(self, s, metrics_variables, metrics_objective_direction, decision_variables=[], options=GuidedImprovementAlgorithmOptions()):
self.s = s
self.metrics_variables = metrics_variables
self.metrics_objective_direction = metrics_objective_direction
self.decision_variables = decision_variables
self.options = options
self.verbosity = self.options.verbosity
self.GIALogger = GIALogging(self.options, self.metrics_variables, self.decision_variables)
class GuidedImprovementAlgorithm(object):
def __init__(self, s, metrics_variables, metrics_objective_direction, decision_variables=[], options=GuidedImprovementAlgorithmOptions()):
self.s = s
self.metrics_variables = metrics_variables
self.metrics_objective_direction = metrics_objective_direction
self.decision_variables = decision_variables
self.options = options
self.verbosity = self.options.verbosity
self.GIALogger = GIALogging(self.options, self.metrics_variables, self.decision_variables)
def _setAndLogZ3RandomSeeds(self):
"""
Will set random seeds for z3 and store them on the file
"""
random.seed()
z3_random_seed_val = random.randint(0, 500000000)
z3_arith_random_seed_val = random.randint(0, 500000000)
set_option(random_seed=z3_random_seed_val)
set_option(arith_random_seed=z3_arith_random_seed_val)
self.options.randomSeedsFil = open(self.options.writeRandomSeedsFilename, "a")
self.options.randomSeedsFil.write("%s, %s, %s\n" % \
( self.options.writeLogFilename ,"random_seed", z3_random_seed_val ))
self.options.randomSeedsFil.write("%s, %s, %s\n" % \
( self.options.writeLogFilename ,"arith_random_seed", z3_arith_random_seed_val ))
self.options.randomSeedsFil.close()
def ExecuteGuidedImprovementAlgorithm(self, outfilename, recordpointfile=""):
"""
Ran the Guided Improvement Algorithm.
"""
self._setAndLogZ3RandomSeeds()
count_paretoPoints = 0
ParetoFront = []
initial_start_time = time()
start_time = time()
count_sat_calls = 0
count_unsat_calls = 0
#self.GIALogger.logStartCall()
if self.s.check() == sat:
count_sat_calls += 1
prev_solution = self.s.model()
#self.GIALogger.logEndCall(True, model=prev_solution, statistics = self.s.statistics())
self.s.push()
FirstParetoPoint, local_count_sat_calls, local_count_unsat_calls = self.ranToParetoFront(prev_solution)
end_time = time()
count_sat_calls += local_count_sat_calls
count_unsat_calls += local_count_unsat_calls
count_paretoPoints += 1
ParetoFront.append(FirstParetoPoint)
# RecordPoint
strNextParetoPoint = list((d.name(), str(FirstParetoPoint[d])) for d in FirstParetoPoint.decls())
if RECORDPOINT:
outputFile = open(recordpointfile, 'a')
try:
outputFile.writelines(str(count_paretoPoints) + ',' +
str(count_sat_calls) + ',' +
str(end_time - start_time) + ',' +
str(strNextParetoPoint) + ',' +
'\n')
finally:
outputFile.close()
##
self.s.pop()
#self.s.add(self.edstest)
tmpNotDominatedByFirstParetoPoint = self.ConstraintNotDominatedByX(FirstParetoPoint)
# self.GIALogger.logStartCall(tmpNotDominatedByFirstParetoPoint)
self.s.add(tmpNotDominatedByFirstParetoPoint)
start_time = time()
while(self.s.check() == sat):
count_sat_calls += 1
# self.GIALogger.logEndCall(True, model = self.s.model(), statistics = self.s.statistics())
prev_solution = self.s.model()
self.s.push()
NextParetoPoint, local_count_sat_calls, local_count_unsat_calls = self.ranToParetoFront(prev_solution)
end_time = time()
count_sat_calls += local_count_sat_calls
count_unsat_calls += local_count_unsat_calls
count_paretoPoints += 1
ParetoFront.append(NextParetoPoint)
# RecordPoint
strNextParetoPoint = list((d.name(), str(FirstParetoPoint[d])) for d in FirstParetoPoint.decls())
if RECORDPOINT:
outputFile = open(recordpointfile, 'a')
try:
outputFile.writelines(str(count_paretoPoints) + ',' +
str(count_sat_calls) + ',' +
str(end_time - start_time) + ',' +
str(strNextParetoPoint) + ',' +
'\n')
finally:
outputFile.close()
##
self.s.pop()
#self.s.add(self.edstest)
# ParetoFront.append(NextParetoPoint)
# if self.options.incrementallyprintparetopoints == True:
# self.print_solution(NextParetoPoint)
tmpNotDominatedByNextParetoPoint = self.ConstraintNotDominatedByX(NextParetoPoint)
# self.GIALogger.logStartCall(tmpNotDominatedByNextParetoPoint)
self.s.add(tmpNotDominatedByNextParetoPoint)
start_time = time()
# self.GIALogger.logEndCall(False, statistics = self.s.statistics())
# else :
count_unsat_calls += 1
#self.GIALogger.logEndCall(False, statistics = self.s.statistics())
end_time = time()
# if OUTPUT_PARETO_FRONT:
# # print count_paretoPoints, count_sat_calls, count_sat_calls, end_time - start_time
# outputFile2 = open("ERS_ParetoFront.csv", 'a')
# ParetoPointsList2 = []
# try:
# for paretopoint in ParetoFront:
# strParetoPoint = list((d.name(), str(paretopoint[d])) for d in paretopoint.decls())
# dict = {}
# for item in strParetoPoint:
# if( item[0] == 'total_batteryusage' or item[0] == 'total_cost' or item[0] == 'total_deploymenttime' or
# item[0] == 'total_developmenttime' or item[0] == 'total_rampuptime' or item[0] == 'total_reliability' or
# item[0] == 'total_responsetime' ):
# # print item
# outputFile2.writelines(str(item) + ';')
# dict[str(item[0])] = str(item[1])
# ParetoPointsList2.append(dict)
# outputFile2.writelines('\n')
# finally:
# outputFile2.close()
# totalpoints = count_paretoPoints
# for i in xrange(len(ParetoPointsList2)):
# isDominated = False
# for j in xrange(i+1, len(ParetoPointsList2)):
# strlist1 = ParetoPointsList2[i]['total_reliability'].split("/")
# #print strlist1
# strlist2 = ParetoPointsList2[j]['total_reliability'].split("/")
# #print strlist2
# if (int(ParetoPointsList2[i]['total_cost']) >= int(ParetoPointsList2[j]['total_cost']) and
# int(ParetoPointsList2[i]['total_responsetime']) >= int(ParetoPointsList2[j]['total_responsetime']) and
# int(ParetoPointsList2[i]['total_batteryusage']) >= int(ParetoPointsList2[j]['total_batteryusage']) and
# int(ParetoPointsList2[i]['total_deploymenttime']) >= int(ParetoPointsList2[j]['total_deploymenttime']) and
# int(ParetoPointsList2[i]['total_developmenttime']) >= int(ParetoPointsList2[j]['total_developmenttime']) and
# int(ParetoPointsList2[i]['total_rampuptime']) >= int(ParetoPointsList2[j]['total_rampuptime']) and
# # precision of reliability is 1/1000
# round(float(strlist1[0])/float(strlist1[1]),12) >= round(float(strlist2[0])/float(strlist2[1]), 12)):
# isDominated = True
# # print str(i) + " dominated by " + str(j)
# # print str(int(ParetoPointsList2[i]['total_cost'])) + " cost >=" + str(int(ParetoPointsList2[j]['total_cost']))
# # print str(int(ParetoPointsList2[i]['total_responsetime'])) + "response >=" + str(int(ParetoPointsList2[j]['total_responsetime']))
# # print str(int(ParetoPointsList2[i]['total_batteryusage'])) + "battery >=" + str(int(ParetoPointsList2[j]['total_batteryusage']))
# # print str(int(ParetoPointsList2[i]['total_deploymenttime'])) + "deploy >=" + str(int(ParetoPointsList2[j]['total_deploymenttime']))
# # print str(int(ParetoPointsList2[i]['total_developmenttime'])) + "develp >=" + str(int(ParetoPointsList2[j]['total_developmenttime']))
# # print str(int(ParetoPointsList2[i]['total_rampuptime'])) + "rampup >=" + str(int(ParetoPointsList2[j]['total_rampuptime']))
# # print str(strlist1[0]) + "/" + str(strlist1[1]) + "reliability >=" + str(strlist2[0]) + "/" + str(strlist2[1])
# # print str(round(float(strlist1[0])/float(strlist1[1]),3)) +"reliability >=" + str(round(float(strlist2[0])/float(strlist2[1]), 3))
# break;
# if isDominated == True:
# totalpoints -= 1
# print totalpoints
# print count_paretoPoints, count_sat_calls, count_sat_calls, end_time - start_time
outputFile = open(outfilename, 'a')
try:
outputFile.writelines(str(count_paretoPoints) + ',' +
str(count_sat_calls) + ',' +
str(count_unsat_calls) + ',' +
str(end_time - initial_start_time) + ',' +
'\n')
finally:
outputFile.close()
# TotalUniqueParetoFront = len(ParetoFront)
# print TotalUniqueParetoFront
# ParetoPointsList = []
# for i in xrange(len(ParetoFront)):
# point = ParetoFront[i]
# strPoint = list((d.name(), str(point[d])) for d in point.decls())
# dict = {}
# for item in strPoint:
# dict[str(item[0])] = str(item[1])
# ParetoPointsList.append(dict)
#
# for i in xrange(len(ParetoPointsList)):
# isDominated = False
# for j in xrange(i+1, len(ParetoPointsList)):
# strlist1 = ParetoPointsList[i]['total_reliability'].split("/")
# #print strlist1
# strlist2 = ParetoPointsList[j]['total_reliability'].split("/")
# #print strlist2
# if (int(ParetoPointsList[i]['total_cost']) < int(ParetoPointsList[j]['total_cost']) and
# int(ParetoPointsList[i]['total_responsetime']) < int(ParetoPointsList[j]['total_responsetime']) and
# int(ParetoPointsList[i]['total_batteryusage']) < int(ParetoPointsList[j]['total_batteryusage']) and
# int(ParetoPointsList[i]['total_deploymenttime']) < int(ParetoPointsList[j]['total_deploymenttime']) and
# int(ParetoPointsList[i]['total_developmenttime']) < int(ParetoPointsList[j]['total_developmenttime']) and
# int(ParetoPointsList[i]['total_rampuptime']) < int(ParetoPointsList[j]['total_rampuptime']) and
# float(int(strlist1[0])/int(strlist1[1])) < float(int(strlist2[0])/int(strlist2[1]))):
# isDominated = True
# if isDominated == True:
# TotalUniqueParetoFront = TotalUniqueParetoFront - 1
# print TotalUniqueParetoFront
# if self.verbosity > consts.VERBOSE_NONE:
# print "ParetoFront has size of %s " % len(ParetoFront)
# print "Time taken is %s " % (end_time - start_time)
#
# for ParetoPoint in ParetoFront:
# self.print_solution(ParetoPoint)
#
# if self.options.useSummaryStatsFile == True:
# self.print_stats_info(end_time, start_time, ParetoFront)
#
# self.options.timefilelog = open(self.options.writeTotalTimeFilename, "a")
# self.options.timefilelog.write("%s, %s\n" % ( self.options.writeLogFilename , (end_time - start_time)))
# self.options.timefilelog.close()
#for entry in self.GIALogger.getLog():
# print entry.statistics
def print_stats_info(self, end_time, start_time, ParetoFront):
statsfile_fd = open(self.args.statsfile, "a")
self.print_header_if_file_is_empty(statsfile_fd)
statsfile_fd.write( "%s, %s,%s\n" % (self.optimizationmodelname , (end_time - start_time), len(ParetoFront)))
statsfile_fd.close()
def print_header_if_file_is_empty(self, statsfile_fd):
if statsfile_fd.tell() == 0:
statsfile_fd.write("%s, %s, %s\n" % ('model name' , 'time to compute pareto front', 'length of pareto front'))
def print_solution(self, solution):
"""
Prints the objective value for the solution in one line, followed by the decision variables in the next line.
"""
SolutionSpacePoint = []
for metric in self.metrics_variables:
SolutionSpacePoint.append(solution[metric])
print SolutionSpacePoint
DecisionSpacePoint = []
for instVariable in solution.decls() :
if instVariable.name() not in self.exclude_variables_in_printout:
DecisionSpacePoint.append("%s=%s" % (instVariable.name(), solution[instVariable]))
print DecisionSpacePoint
def ranToParetoFront(self, prev_solution):
"""
It iterates getting close and closer to the pareto front.
"""
# if self.verbosity > consts.VERBOSE_NONE:
# self.print_solution(prev_solution)
local_count_sat_calls = 0
local_count_unsat_calls = 0
tmpConstraintMustDominateX= self.ConstraintMustDominatesX(prev_solution)
# self.GIALogger.logStartCall(tmpConstraintMustDominateX)
self.s.add(tmpConstraintMustDominateX)
while (self.s.check() == sat):
local_count_sat_calls += 1
# self.GIALogger.logEndCall(True, model = self.s.model(), statistics = self.s.statistics())
# if self.verbosity > consts.VERBOSE_NONE:
# self.print_solution(prev_solution)
prev_solution = self.s.model()
tmpConstraintMustDominateX = self.ConstraintMustDominatesX(prev_solution)
self.s.add(tmpConstraintMustDominateX)
self.GIALogger.logStartCall(tmpConstraintMustDominateX)
local_count_unsat_calls += 1
# self.edstest= self.edstestExtension(prev_solution)
# self.GIALogger.logEndCall(False, statistics = self.s.statistics())
# if self.verbosity > consts.VERBOSE_NONE:
# self.print_solution(prev_solution)
return prev_solution, local_count_sat_calls, local_count_unsat_calls
def ConstraintNotDominatedByX(self, model):
"""
Returns a Constraint that a new instance, can't be dominated by the instance represented by model.
(it can't be worst in any objective).
"""
DisjunctionOrLessMetrics = list()
for i in range(len(self.metrics_variables)):
if self.metrics_objective_direction[i] == consts.METRICS_MAXIMIZE:
DisjunctionOrLessMetrics.append(self.metrics_variables[i] > model[self.metrics_variables[i]])
else :
DisjunctionOrLessMetrics.append(self.metrics_variables[i] < model[self.metrics_variables[i]])
return Or(DisjunctionOrLessMetrics)
# add for EPOAL
def EtractConstraintListNotDominatedByX(self, model):
"""
Returns a Constraint that a new instance, can't be dominated by the instance represented by model.
(it can't be worst in any objective).
"""
DisjunctionOrLessMetrics = list()
for i in range(len(self.metrics_variables)):
if self.metrics_objective_direction[i] == consts.METRICS_MAXIMIZE:
DisjunctionOrLessMetrics.append(self.metrics_variables[i] > model[self.metrics_variables[i]])
else :
DisjunctionOrLessMetrics.append(self.metrics_variables[i] < model[self.metrics_variables[i]])
return DisjunctionOrLessMetrics
def edstestExtension(self, model):
'''seeing if this extra constraint improves performance.'''
dominationConjunction = []
i = 0
for dominatedByMetric in self.metrics_variables:
if self.metrics_objective_direction[i] == consts.METRICS_MAXIMIZE :
dominationConjunction.append(dominatedByMetric > model[dominatedByMetric])
else:
dominationConjunction.append(dominatedByMetric < model[dominatedByMetric])
i = 1 + i
constraintDominateX = Not(And(dominationConjunction))
return constraintDominateX
def ConstraintMustDominatesX(self, model):
"""
Returns a constraint that a new instance has to be better than the instance represented by model in at least one dimension,
and better or equal in all the other ones.
"""
dominationDisjunction= []
i = 0
for dominatedByMetric in self.metrics_variables:
dominationConjunction = []
j = 0
if self.metrics_objective_direction[i] == consts.METRICS_MAXIMIZE :
dominationConjunction.append(dominatedByMetric > model[dominatedByMetric])
else:
dominationConjunction.append(dominatedByMetric < model[dominatedByMetric])
for AtLeastEqualInOtherMetric in self.metrics_variables:
if j != i:
if self.metrics_objective_direction[j] == consts.METRICS_MAXIMIZE:
dominationConjunction.append(AtLeastEqualInOtherMetric >= model[AtLeastEqualInOtherMetric])
else:
dominationConjunction.append(AtLeastEqualInOtherMetric <= model[AtLeastEqualInOtherMetric])
j = 1 + j
i = 1 + i
dominationDisjunction.append(And(dominationConjunction))
constraintDominateX = Or(dominationDisjunction)
return constraintDominateX