def n_samp_wrap(n, individual, propDB, Kinp):
numpy.random.seed(n)
ncomp, spc_names, propvec = make_property_vector_all_sample_cost(propDB)
this_ron = blend(individual, propvec, 'RON')
this_s = blend(individual, propvec, 'S')
this_HoV = blend(individual, propvec, 'HoV')
# this_SL = blend(individual, propvec, 'SL')
this_AFR = blend(individual, propvec, 'AFR_STOICH')
this_LFV150 = blend(individual, propvec, 'LFV150')
this_PMI = blend(individual, propvec, 'PMI')
cost_f = blend(individual, propvec, 'COST')
# merit_f = mmf_single(RON=this_ron, S=this_s,
# HoV=this_HoV, SL=this_SL, K=Kinp)
merit_vec = mmf_single(RON=this_ron,
S=this_s,
HoV=this_HoV,
AFR=this_AFR,
PMI=this_PMI,
K=Kinp)
cost_vec = cost_f
# cost_vec = numpy.random.rand(1)[0]
return cost_vec # merit_vec
def eval_mo(individual, propvec, Kinp):
this_ron = blend(individual, propvec, 'RON')
this_s = blend(individual, propvec, 'S')
this_HoV = blend(individual, propvec, 'HoV')
this_SL = blend(individual, propvec, 'SL')
this_AFR = blend(individual, propvec, 'AFR_STOICH')
this_LFV150 = blend(individual, propvec, 'LFV150')
this_PMI = blend(individual, propvec, 'PMI')
cost_f = blend(individual, propvec, 'COST')
merit_f = mmf_single(RON=this_ron,
S=this_s,
HoV=this_HoV,
AFR=this_AFR,
PMI=this_PMI,
K=Kinp)
g_val = fraction_constraint(individual)
if (numpy.abs(g_val - 1) <= 1.0e-3):
penalty_mmf = 0
penalty_cost = 0
else: # theoretically, this part of the condition will never be entered
raise 'Your parameters dont add up to 1 '
# penalty_cost = 0 if nothing is changed about creation of individuals
c_p_p = cost_f + penalty_cost
# penalty_mmf = 0 if nothing is changed about creation of individuals
mmf_p_p = (merit_f + penalty_mmf) # minimize -Merit
# add here more objective functions as needed --
# add 1/-1 in line 125, weights (-1 = minimize, 1 = maximize)
# obj3 = numpy.prod(individual)
return mmf_p_p, c_p_p # , obj3
def obj_fun(model):
this_ron = blend_linear_pyomo(model, 'RON')
this_s = blend_linear_pyomo(model, 'S')
this_HoV = blend_linear_pyomo(model, 'HoV')
this_SL = blend_linear_pyomo(model, 'SL')
this_LFV150 = blend_linear_pyomo(model, 'LFV150')
this_PMI = blend_linear_pyomo(model, 'PMI')
return mmf_single(RON=this_ron, S=this_s, HoV=this_HoV,
SL=this_SL, K=KK)
def obj_fun(model):
this_ron = blend_linear_pyomo(model, 'RON')
this_s = blend_linear_pyomo(model, 'S')
this_HoV = blend_linear_pyomo(model, 'HoV')
this_SL = blend_linear_pyomo(model, 'SL')
this_LFV150 = blend_linear_pyomo(model, 'LFV150')
this_PMI = blend_linear_pyomo(model, 'PMI')
if(ref is None):
return mmf_single(RON=this_ron, S=this_s, HoV=this_HoV,
SL=this_SL, K=KK)
else:
return mmf_single_param(ref, sen, RON=this_ron, S=this_s,
HoV=this_HoV, SL=this_SL, K=KK)
def objective(individual, propvec, Kinp):
# compute miles function
this_ron = blend(individual, propvec, 'RON')
this_s = blend(individual, propvec, 'S')
this_HoV = blend(individual, propvec, 'HoV')
this_SL = blend(individual, propvec, 'SL')
this_AFR = blend(individual, propvec, 'AFR_STOICH')
this_LFV150 = blend(individual, propvec, 'LFV150')
this_PMI = blend(individual, propvec, 'PMI')
cost_f = blend(individual, propvec, 'COST')
merit_f = mmf_single(RON=this_ron, S=this_s,
HoV=this_HoV, AFR=this_AFR, PMI=this_PMI, K=Kinp)
return -merit_f # maximize merit = minimize -merit
def comp_to_cost_mmf(comp, propDB, k):
# Evaluate resulting mmf value
# print comp
prop_list = ['RON', 'S', 'HoV', 'SL', 'LFV150', 'PMI']
props = {}
for p in prop_list:
props[p] = blend_linear_propDB(p, propDB, comp)
mmf = mmf_single(RON=props['RON'], S=props['S'],
HoV=props['HoV'], SL=props['SL'],
LFV150=props['LFV150'], PMI=props['PMI'], K=k)
cost = blend_linear_propDB('COST', propDB, comp)
# print "Cost: {}, cstar = {} ".format(cost,cstar)
# This to make a radar plot of the composition for each solution
# cpt.plot_comp_radar(propDB,comp,savefile="{}_mfK{}".format(cstar,KK),
# title="Composition, "\
# "C*={}, mmf={:.2f}".format(cstar,mmf))
return cost, mmf
def comp_to_mmf(comp, propDB, k, ref=None, sen=None):
# Evaluate resulting mmf value
# print comp
prop_list = ['RON', 'S', 'HoV', 'SL', 'LFV150', 'PMI']
props = {}
for p in prop_list:
props[p] = blend_linear_propDB(p, propDB, comp)
if ref is None and sen is None:
mmf = mmf_single(RON=props['RON'], S=props['S'],
HoV=props['HoV'], SL=props['SL'],
LFV150=props['LFV150'], PMI=props['PMI'], K=k)
else:
mmf = mmf_single_param(ref, sen, RON=props['RON'], S=props['S'],
HoV=props['HoV'], SL=props['SL'],
LFV150=props['LFV150'], PMI=props['PMI'],
K=k)
cost = blend_linear_propDB('COST', propDB, comp)
return mmf
def objective_function(x, propvec, Kinp):
w = np.zeros(len(x) + 1)
w[0:len(x)] = x
w[-1] = 1 - np.sum(x)
this_ron = blend(w, propvec, 'RON')
this_s = blend(w, propvec, 'S')
this_HoV = blend(w, propvec, 'HoV')
this_SL = blend(w, propvec, 'SL')
this_AFR = blend(w, propvec, 'AFR_STOICH')
this_LFV150 = blend(w, propvec, 'LFV150')
this_PMI = blend(w, propvec, 'PMI')
cost_f = blend(w, propvec, 'COST')
f0 = -mmf_single(RON=this_ron,
S=this_s,
HoV=this_HoV,
AFR=this_AFR,
PMI=this_PMI,
K=Kinp)
return f0
def objective_function_exp(x, propvec, Kinp):
# needs to be generalized later -- add a component to make the sum = 1
if len(x) == 21:
w = np.zeros(len(x)+1)
w[0:len(x)] = x
w[-1] = 1 - np.sum(x)
else:
w = x
this_ron = blend(w, propvec, 'RON')
this_s = blend(w, propvec, 'S')
this_HoV = blend(w, propvec, 'HoV')
this_SL = blend(w, propvec, 'SL')
this_AFR = blend(w, propvec, 'AFR_STOICH')
this_LFV150 = blend(w, propvec, 'LFV150')
this_PMI = blend(w, propvec, 'PMI')
cost_f = blend(w, propvec, 'COST')
f0 = mmf_single(RON=this_ron, S=this_s, HoV=this_HoV,
AFR=this_AFR, PMI=this_PMI, K=Kinp)
# f1 = cost_f
return -f0 # , f1 #minimize negative merit
def sample_mmf_k(bplist, kmean, kvar, nsamples):
# Now let some parameters be random variables
# First look at K
mlist = []
klist = []
for bp in bplist:
ksamp = np.random.normal(kmean, kvar, nsamples)
msamp = np.zeros([nsamples])
for ns in range(nsamples):
msamp[ns] = mmf_single(RON=bp['RON'],
S=bp['S'],
HoV=bp['HoV'],
SL=bp['SL'],
LFV150=bp['LFV150'],
PMI=bp['PMI'],
K=ksamp[ns])
klist.append(ksamp)
mlist.append(msamp)
return mlist, klist
def eval_MMF_gp(individual, propvec, Kinp, GP, scal):
# print(individual)
# individual = [RON, S, HOV, SL, LFV150, PMI, CA50, IAT, KI]
# OG bounds CA50, IAT, KI, RON, S,HOV
low = numpy.array([6.7, 35., 2., 99.1, 0., 303.])
# OG bounds CA50, IAT, KI, RON, S,HOV
up = numpy.array([23.8, 90., 10.5, 105.6, 12.2, 595.])
CA50 = low[0] + (up[0] - low[0]) * individual[22]
IAT = low[1] + (up[1] - low[1]) * individual[23]
KI = low[2] + (up[2] - low[2]) * individual[24]
# print(propvec)
this_ron = blend(individual[0:22], propvec, 'RON')
this_s = blend(individual[0:22], propvec, 'S')
this_HoV = blend(individual[0:22], propvec, 'HoV')
this_SL = blend(individual[0:22], propvec, 'SL')
# this_AFR = blend(individual, propvec, 'AFR_STOICH')
this_LFV150 = blend(individual[0:22], propvec, 'LFV150')
this_PMI = blend(individual[0:22], propvec, 'PMI')
cost_f = blend(individual[0:22], propvec, 'COST')
merit_f = mmf_single(RON=this_ron,
S=this_s,
HoV=this_HoV,
SL=this_SL,
K=-1.25)
RON = this_ron
S = this_s
HOV = this_HoV
# CA50, IAT, KI, RON, S,HOV
gp_in = numpy.array([[CA50, IAT, KI, RON, S, HOV]]).reshape(1, -1)
pred_mean, pred_std = predict_GP(GP, scal, gp_in)
return merit_f, pred_mean[0]
def eval_MMF_gp_opt(individual, propvec, Kinp, GP, scal):
# print(individual)
# individual = [RON, S, HOV, SL, LFV150, PMI, CA50, IAT, KI]
# OG bounds CA50, IAT, KI, RON, S,HOV
low = numpy.array([6.7, 35., 2., 99.1, 0., 303.])
# OG bounds CA50, IAT, KI, RON, S,HOV
up = numpy.array([23.8, 90., 10.5, 105.6, 12.2, 595.])
# CA50 = low[0]+(up[0]-low[0])*individual[22]
# IAT = low[1]+(up[1]-low[1])*individual[23]
# KI = low[2]+(up[2]-low[2])*individual[24]
# print(propvec)
this_ron = blend(individual[0:22], propvec, 'RON')
this_s = blend(individual[0:22], propvec, 'S')
this_HoV = blend(individual[0:22], propvec, 'HoV')
this_SL = blend(individual[0:22], propvec, 'SL')
# this_AFR = blend(individual, propvec, 'AFR_STOICH')
this_LFV150 = blend(individual[0:22], propvec, 'LFV150')
this_PMI = blend(individual[0:22], propvec, 'PMI')
cost_f = blend(individual[0:22], propvec, 'COST')
merit_f = mmf_single(RON=this_ron,
S=this_s,
HoV=this_HoV,
SL=this_SL,
K=-1.25)
RON = this_ron
S = this_s
HOV = this_HoV
# if the values for RON, S, HOV are outside the
# GP training domani, assign bad fitness values
if ((RON < 99.1) or (RON > 105.6) or (S < 0) or (S > 12.2) or (HOV < 303)
or (HOV > 595)):
merit_f = -100
mean_out = -100
else:
# print(RON, S, HOV)
# RON,S,HOV fixed, now optimize over IAT, CA50, KI
bound_list = [(6.7, 23.8), (35., 90.), (2., 10.5)]
# fbest = numpy.inf
# xnew = None
# opti_exit = None
# number of local searches that we want to do -- bump this up
# if you want to do several
# local searches (may lead to different local optima)
# n_t = 5
# for jj in range(n_t):
x0 = (numpy.array([23.8, 90., 10.5]) +
numpy.array([6.7, 35., 2.])) / 2.
# numpy.asarray([6.7, 35., 2.]) + numpy.asarray(numpy.array([23.8, 90., 10.5])-numpy.array([6.7, 35., 2.])) * numpy.asarray(numpy.random.rand(1,3)) #random starting point
res = minimize(f_gp,
x0,
bounds=bound_list,
args=(
RON,
S,
HOV,
GP,
scal,
))
mean_out = -f_gp(res.x, RON, S, HOV, GP, scal)
return merit_f, mean_out
