def satisfied(self, sp):
from optTune.MOTA_code import MOTA_subproblem
assert isinstance(sp, MOTA_subproblem)
sp_int = len(sp.PFA_history.gamma_hist)
if sp_int <= self.averagingIterations:
return False
else:
PFA_c = paretoArchive2D()
for d in sp.PFA.designs:
f1 = self.f1_transform(d.fv[0])
f2 = self.f2_transform(d.fv[1])
PFA_c.inspect_design( None, numpy.array([f1,f2]) )
PFA_prev = paretoArchive2D()
PFA_prev_untransformed = sp.PFA_history.get_PFA_index(sp_int-self.averagingIterations-1, sp.PFA.__class__ )
f1_org = [ d.f1_val for d in PFA_prev_untransformed.designs]
U_org = [ d.U_i for d in PFA_prev_untransformed.designs]
for f1, U_i in zip(f1_org, U_org):
f2 = sp.scalarize_utility_values(U_i).mean()
PFA_prev.inspect_design( None, numpy.array([self.f1_transform(f1),self.f2_transform(f2)]) )
HV_ref = PFA_c.upper_bounds()
HV_curr = PFA_c.hyper_volume(HV_ref)
HV_prev = PFA_prev.hyper_volume(HV_ref)
ratio = (HV_curr - HV_prev) / HV_curr
if ratio < 0:
print("ratio of HV improvement negative! pickle.dumps follows")
print("PFA_c", pickle.dumps(PFA_c) )
print("self.PFAs[0]", pickle.dumps(self.PFAs[0]) )
raise RuntimeError, "ratio of HV improvement negative!"
if ratio <= self.ratioThreshold:
self.msg = "hyperVolume stagation occured : ratio %1.3f ( < %1.3f ) over %i iterations" % (ratio, self.ratioThreshold, self.averagingIterations)
return True
else:
#print(" hyperVolume yet to occur : ratio %1.3f ( limit %1.3f ) over %i iterations" % (ratio, self.ratioThreshold, self.averagingIterations ) )
return False
def satisfied(self, opt):
assert len(self.PFAs) <= opt.it
PFA_c = paretoArchive2D()
for d in opt.PFA.designs:
f1 = self.f1_transform(d.fv[0])
f2 = self.f2_transform(d.fv[1])
PFA_c.inspect_design(None, numpy.array([f1, f2]))
self.PFAs.append(PFA_c)
if len(self.PFAs) > self.averagingIterations:
del self.PFAs[0]
if len(self.PFAs) < self.averagingIterations:
return False
else:
HV_ref = PFA_c.upper_bounds()
HV_curr = PFA_c.hyper_volume(HV_ref)
HV_prev = self.PFAs[0].hyper_volume(HV_ref)
ratio = (HV_curr - HV_prev) / HV_curr
if ratio < 0:
print("ratio of HV improvement negative! pickle.dumps follows")
print("PFA_c", pickle.dumps(PFA_c))
print("self.PFAs[0]", pickle.dumps(self.PFAs[0]))
raise RuntimeError, "ratio of HV improvement negative!"
if ratio <= self.ratioThreshold:
self.msg = "hyperVolume stagation occured : ratio %1.3f ( < %1.3f ) over %i iterations" % (
ratio, self.ratioThreshold, self.averagingIterations)
return True
else:
#print(" hyperVolume yet to occur : ratio %1.3f ( limit %1.3f ) over %i iterations" % (ratio, self.ratioThreshold, self.averagingIterations ) )
return False
def satisfied(self, opt):
assert len(self.PFAs) <= opt.it
PFA_c = paretoArchive2D()
for d in opt.PFA.designs:
f1 = self.f1_transform(d.fv[0])
f2 = self.f2_transform(d.fv[1])
PFA_c.inspect_design( None, numpy.array([f1,f2]) )
self.PFAs.append(PFA_c)
if len(self.PFAs) > self.averagingIterations:
del self.PFAs[0]
if len(self.PFAs) < self.averagingIterations:
return False
else:
HV_ref = PFA_c.upper_bounds()
HV_curr = PFA_c.hyper_volume(HV_ref)
HV_prev = self.PFAs[0].hyper_volume(HV_ref)
ratio = (HV_curr - HV_prev) / HV_curr
if ratio < 0:
print("ratio of HV improvement negative! pickle.dumps follows")
print("PFA_c", pickle.dumps(PFA_c) )
print("self.PFAs[0]", pickle.dumps(self.PFAs[0]) )
raise RuntimeError, "ratio of HV improvement negative!"
if ratio <= self.ratioThreshold:
self.msg = "hyperVolume stagation occured : ratio %1.3f ( < %1.3f ) over %i iterations" % (ratio, self.ratioThreshold, self.averagingIterations)
return True
else:
#print(" hyperVolume yet to occur : ratio %1.3f ( limit %1.3f ) over %i iterations" % (ratio, self.ratioThreshold, self.averagingIterations ) )
return False
def satisfied(self, sp):
from optTune.MOTA_code import MOTA_subproblem
assert isinstance(sp, MOTA_subproblem)
sp_int = len(sp.PFA_history.gamma_hist)
if sp_int <= self.averagingIterations:
return False
else:
PFA_c = paretoArchive2D()
for d in sp.PFA.designs:
f1 = self.f1_transform(d.fv[0])
f2 = self.f2_transform(d.fv[1])
PFA_c.inspect_design(None, numpy.array([f1, f2]))
PFA_prev = paretoArchive2D()
PFA_prev_untransformed = sp.PFA_history.get_PFA_index(
sp_int - self.averagingIterations - 1, sp.PFA.__class__)
f1_org = [d.f1_val for d in PFA_prev_untransformed.designs]
U_org = [d.U_i for d in PFA_prev_untransformed.designs]
for f1, U_i in zip(f1_org, U_org):
f2 = sp.scalarize_utility_values(U_i).mean()
PFA_prev.inspect_design(
None,
numpy.array([self.f1_transform(f1),
self.f2_transform(f2)]))
HV_ref = PFA_c.upper_bounds()
HV_curr = PFA_c.hyper_volume(HV_ref)
HV_prev = PFA_prev.hyper_volume(HV_ref)
ratio = (HV_curr - HV_prev) / HV_curr
if ratio < 0:
print("ratio of HV improvement negative! pickle.dumps follows")
print("PFA_c", pickle.dumps(PFA_c))
print("self.PFAs[0]", pickle.dumps(self.PFAs[0]))
raise RuntimeError, "ratio of HV improvement negative!"
if ratio <= self.ratioThreshold:
self.msg = "hyperVolume stagation occured : ratio %1.3f ( < %1.3f ) over %i iterations" % (
ratio, self.ratioThreshold, self.averagingIterations)
return True
else:
#print(" hyperVolume yet to occur : ratio %1.3f ( limit %1.3f ) over %i iterations" % (ratio, self.ratioThreshold, self.averagingIterations ) )
return False
ratio, self.ratioThreshold, self.averagingIterations)
return True
else:
#print(" hyperVolume yet to occur : ratio %1.3f ( limit %1.3f ) over %i iterations" % (ratio, self.ratioThreshold, self.averagingIterations ) )
return False
if __name__ == '__main__':
print('Testing optTune termination critera')
terminator = hyperVolume_stagation(0.1, 5)
class optClass:
pass
opt = optClass()
opt.PFA = paretoArchive2D()
PFA_hist = []
opt.it = 0
N = 5
while True:
for i in range(N):
f1 = numpy.random.rand()
f2 = numpy.random.rand()
if f1 + f2 < 0.5:
f1 = 1 - f1
f2 = 1 - f2
opt.PFA.inspect_design(None, numpy.array([f1, f2]))
opt.it = opt.it + 1
print(' it %i complete, HV(opt.PFA) %f' %
(opt.it, opt.PFA.hyper_volume(numpy.array([1, 1]))))
print("self.PFAs[0]", pickle.dumps(self.PFAs[0]) )
raise RuntimeError, "ratio of HV improvement negative!"
if ratio <= self.ratioThreshold:
self.msg = "hyperVolume stagation occured : ratio %1.3f ( < %1.3f ) over %i iterations" % (ratio, self.ratioThreshold, self.averagingIterations)
return True
else:
#print(" hyperVolume yet to occur : ratio %1.3f ( limit %1.3f ) over %i iterations" % (ratio, self.ratioThreshold, self.averagingIterations ) )
return False
if __name__ == '__main__':
print('Testing optTune termination critera')
terminator = hyperVolume_stagation(0.1, 5)
class optClass:
pass
opt = optClass()
opt.PFA = paretoArchive2D()
PFA_hist = []
opt.it = 0
N = 5
while True:
for i in range(N):
f1 = numpy.random.rand()
f2 = numpy.random.rand()
if f1 + f2 < 0.5:
f1 = 1 - f1
f2 = 1 - f2
opt.PFA.inspect_design( None, numpy.array([f1,f2]) )
opt.it = opt.it + 1
print(' it %i complete, HV(opt.PFA) %f' % (opt.it,opt.PFA.hyper_volume(numpy.array([1,1])) ))
PFA_hist.append(opt.PFA.copy())
