def run_IPs_per_forecast(numforecasts,
mC,
tsteps,
ipoutageattr,
useThisT=None):
if useThisT == None:
useThisT = tsteps - 1
IPSAAModel = [None for nf in xrange(numforecasts)]
ipobjval = [None for nf in xrange(numforecasts)]
sys.stdout.write("Processing forecast ")
for nf in xrange(numforecasts):
sys.stdout.write(str(nf) + ".. ")
IPSAAModel[nf] = exeADP.create_models_temp_framework(tsteps, mC)
nx.set_edge_attributes(mC["netobj"].dinet,
"t" + str(useThisT) + "_capacity",
ipoutageattr[nf])
exeADP.update_models_orig_scens(IPSAAModel[nf], mC, numscens=len(ipoutageattr[nf].values()[0]), \
mrange=[useThisT], reset=True, updateConstr=True)
# restricting the number of arcs to install as in the training procedure
#ASSUMPTION: even if we observe a problem at a time step other than T, we will allow a full install budget
IPSAAModel[nf][useThisT].linear_constraints.add( \
lin_expr = [cplex.SparsePair(ind = mC["temp_var_insts"], \
val = [1]*len(mC["temp_var_insts"]) ) ], \
senses = ["L"], rhs = [mC["pertinstbudget"]*tsteps], names = ["PerTInstConstr"] )
try:
IPSAAModel[nf][useThisT].solve()
ipobjval[nf] = IPSAAModel[nf][
useThisT].solution.get_objective_value()
#DEBUG begin
## slks = IPSAAModel[nf][useThisT].solution.get_values(mC["slak_var_names"])
## insts = [int(round(i)) for i in IPSAAModel[nf][useThisT].solution.get_values(mC["temp_var_insts"]) ]
## idx = [ i for i in xrange(len(insts)) if int(round(insts[i]))]
## instnames = [mC["temp_var_insts"][i] for i in idx]
## ltdict = dict()
## for ins in instnames:
## crds = exeADP.convert_cplex_structs(cpobj=ins)
## if mC["netobj"].tnet.edge[crds[0]][crds[1]]["linetype"] not in ltdict:
## ltdict[mC["netobj"].tnet.edge[crds[0]][crds[1]]["linetype"]] = 1
## else:
## ltdict[mC["netobj"].tnet.edge[crds[0]][crds[1]]["linetype"]] += 1
## print "\nLinetype counts"
## for k,v in ltdict.iteritems():
## print k+": "+str(v)
## # len of capacities => # of scenarios in the SAA
## print "avg # of slacks: "+str(len([i for i in slks if int(round(i,4)) != 0]) / \
## float(len(ipoutageattr[nf].values()[0])) )
#DEBUG end
except:
print "testADP: test IP SAA model infeasible"
# this is INCREDIBLY slow, and there's tsteps*numforecasts models, so we'll sacrifice memory for now
## exeADP.trash_model([k for l in IPSAAModel for k in l])
return ipobjval
def run_IPs_per_forecast(numforecasts, mC, tsteps, ipoutageattr, useThisT=None):
if useThisT == None:
useThisT = tsteps-1
IPSAAModel = [None for nf in xrange(numforecasts)]
ipobjval = [None for nf in xrange(numforecasts)]
sys.stdout.write("Processing forecast ")
for nf in xrange(numforecasts):
sys.stdout.write(str(nf)+".. ")
IPSAAModel[nf] = exeADP.create_models_temp_framework(tsteps, mC)
nx.set_edge_attributes(mC["netobj"].dinet, "t"+str(useThisT)+"_capacity", ipoutageattr[nf])
exeADP.update_models_orig_scens(IPSAAModel[nf], mC, numscens=len(ipoutageattr[nf].values()[0]), \
mrange=[useThisT], reset=True, updateConstr=True)
# restricting the number of arcs to install as in the training procedure
#ASSUMPTION: even if we observe a problem at a time step other than T, we will allow a full install budget
IPSAAModel[nf][useThisT].linear_constraints.add( \
lin_expr = [cplex.SparsePair(ind = mC["temp_var_insts"], \
val = [1]*len(mC["temp_var_insts"]) ) ], \
senses = ["L"], rhs = [mC["pertinstbudget"]*tsteps], names = ["PerTInstConstr"] )
try:
IPSAAModel[nf][useThisT].solve()
ipobjval[nf] = IPSAAModel[nf][useThisT].solution.get_objective_value()
#DEBUG begin
## slks = IPSAAModel[nf][useThisT].solution.get_values(mC["slak_var_names"])
## insts = [int(round(i)) for i in IPSAAModel[nf][useThisT].solution.get_values(mC["temp_var_insts"]) ]
## idx = [ i for i in xrange(len(insts)) if int(round(insts[i]))]
## instnames = [mC["temp_var_insts"][i] for i in idx]
## ltdict = dict()
## for ins in instnames:
## crds = exeADP.convert_cplex_structs(cpobj=ins)
## if mC["netobj"].tnet.edge[crds[0]][crds[1]]["linetype"] not in ltdict:
## ltdict[mC["netobj"].tnet.edge[crds[0]][crds[1]]["linetype"]] = 1
## else:
## ltdict[mC["netobj"].tnet.edge[crds[0]][crds[1]]["linetype"]] += 1
## print "\nLinetype counts"
## for k,v in ltdict.iteritems():
## print k+": "+str(v)
## # len of capacities => # of scenarios in the SAA
## print "avg # of slacks: "+str(len([i for i in slks if int(round(i,4)) != 0]) / \
## float(len(ipoutageattr[nf].values()[0])) )
#DEBUG end
except:
print "testADP: test IP SAA model infeasible"
# this is INCREDIBLY slow, and there's tsteps*numforecasts models, so we'll sacrifice memory for now
## exeADP.trash_model([k for l in IPSAAModel for k in l])
return ipobjval
def run_one_big_IPSAA(numforecasts, mC, tsteps, ipoutageattr, samplesize):
IPSAAModel = exeADP.create_models_temp_framework(tsteps, mC)
nx.set_edge_attributes(mC["netobj"].dinet, "t"+str(tsteps-1)+"_capacity", ipoutageattr)
exeADP.update_models_orig_scens(IPSAAModel, mC, numscens=numforecasts*samplesize, mrange=[tsteps-1], reset=True,\
updateConstr=True)
IPSAAModel[tsteps-1].linear_constraints.add( \
lin_expr = [cplex.SparsePair(ind = mC["temp_var_insts"], \
val = [1]*len(mC["temp_var_insts"])) ], \
senses = ["L"], rhs = [tsteps*mC["pertinstbudget"]], names = ["TInstConstr"] )
try:
IPSAAModel[tsteps-1].solve()
vcontr = IPSAAModel[tsteps-1].solution.get_objective_value()
except:
print "testADP: control IP SAA model infeasible"
__dbg = raw_input("execution halted, press ENTER to exit")
sys.exit()
exeADP.trash_model(IPSAAModel)
sys.stdout.write("Done.\n")
return vcontr
def run_one_big_IPSAA(numforecasts, mC, tsteps, ipoutageattr, samplesize):
IPSAAModel = exeADP.create_models_temp_framework(tsteps, mC)
nx.set_edge_attributes(mC["netobj"].dinet,
"t" + str(tsteps - 1) + "_capacity", ipoutageattr)
exeADP.update_models_orig_scens(IPSAAModel, mC, numscens=numforecasts*samplesize, mrange=[tsteps-1], reset=True,\
updateConstr=True)
IPSAAModel[tsteps-1].linear_constraints.add( \
lin_expr = [cplex.SparsePair(ind = mC["temp_var_insts"], \
val = [1]*len(mC["temp_var_insts"])) ], \
senses = ["L"], rhs = [tsteps*mC["pertinstbudget"]], names = ["TInstConstr"] )
try:
IPSAAModel[tsteps - 1].solve()
vcontr = IPSAAModel[tsteps - 1].solution.get_objective_value()
except:
print "testADP: control IP SAA model infeasible"
__dbg = raw_input("execution halted, press ENTER to exit")
sys.exit()
exeADP.trash_model(IPSAAModel)
sys.stdout.write("Done.\n")
return vcontr
def run_LPs_per_forecast(numforecasts, mC, tsteps, adpftdmgattr, adpoutageattr, features, lpoutageattr, \
poleval="adp"):
policyEvalModels = [None for nf in xrange(numforecasts)]
LPSAAModel = [None for nf in xrange(numforecasts)]
lpobjval = [None for nf in xrange(numforecasts)]
polpicks = [None for nf in xrange(numforecasts)]
# find multiple policy evaluations for robustness
for nf in xrange(numforecasts):
# step 1: evaluation of policy
if poleval == "alt":
netobj = copy.deepcopy(mC["netobj"])
samplesize = len(lpoutageattr[0].values()[0])
rankedarcs = exeADP.alt_vfa_calculation(tsteps, mC, netobj, features, numforecasts, adpftdmgattr)
policies = exeADP.alt_control_step(tsteps, mC, netobj, samplesize, rankedarcs)
polpicks[nf] = policies[-1]
elif poleval == "adp":
policyEvalModels[nf] = exeADP.create_models_temp_framework(tsteps, mC)
# override netobj's info with the stored scenario data
for t in xrange(tsteps):
nx.set_edge_attributes(mC["netobj"].dinet, "t"+str(t)+"_dmg_pct", adpftdmgattr[nf][t])
nx.set_edge_attributes(mC["netobj"].dinet, "t"+str(t)+"_capacity", adpoutageattr[nf][t])
# solve policy search problem to find install arcs
mC["mcsim"].eval_adp_basis_functions(mC["netobj"], tsteps, mC)
featurelist = mC["mcsim"].calc_feature_ctrmassquad(mC["netobj"], tsteps, features)
if "ud_vfa_combined" in mC["polmethod"] or "node_constr" in mC["polmethod"]:
exeADP.update_models_orig_scens(policyEvalModels[nf], mC, mrange=xrange(len(policyEvalModels[nf])), \
reset=True, updateConstr=True)
elif "ud_vfa_separate" in mC["polmethod"]:
exeADP.update_models_orig_scens(policyEvalModels[nf], mC, mrange=xrange(len(policyEvalModels[nf])), \
reset=True, updateConstr=False)
else:
print "run_LPs_per_forecast: invalid policy method specified."
__dbg = raw_input("execution halted, press ENTER to exit")
sys.exit()
mC["instchoices"] = [ [0]*len(mC["temp_var_insts"]) for i in xrange(tsteps)]
mC["previnsts"] = [[] for i in xrange(tsteps)]
for t in xrange(tsteps):
optpolicy,tflows,slks,bov,vfav = exeADP.adp_policy_search(t, featurelist, policyEvalModels[nf], mC,\
randpol=False)
# this is all we care about -- the policy evaluation at the last time step (i.e. all installed arcs)
polpicks[nf] = optpolicy
#DEBUG begin
## idx = [ i for i in xrange(len(optpolicy)) if int(round(optpolicy[i]))]
## instnames = [mC["temp_var_insts"][i] for i in idx]
## ltdict = dict()
## for ins in instnames:
## crds = exeADP.convert_cplex_structs(cpobj=ins)
## if mC["netobj"].tnet.edge[crds[0]][crds[1]]["linetype"] not in ltdict:
## ltdict[mC["netobj"].tnet.edge[crds[0]][crds[1]]["linetype"]] = 1
## else:
## ltdict[mC["netobj"].tnet.edge[crds[0]][crds[1]]["linetype"]] += 1
## print "\nLinetype counts for forecast "+str(nf)
## for k,v in ltdict.iteritems():
## print k+": "+str(v)
## print "policy eval slacks: "+str(len([i for i in slks if int(round(i,4)) != 0]))
#DEBUG end
# step 2: fix the temp arcs to the decisions and run an LP SAA over these samples at time T
LPSAAModel[nf] = exeADP.create_models_temp_framework(tsteps, mC)
nx.set_edge_attributes(mC["netobj"].dinet, "t"+str(tsteps-1)+"_capacity", lpoutageattr[nf])
exeADP.update_models_orig_scens(LPSAAModel[nf], mC, numscens=len(lpoutageattr[nf].values()[0]), \
mrange=[tsteps-1], reset=True, updateConstr=True)
LPSAAModel[nf][tsteps-1].variables.set_types([ (mC["temp_var_insts"][j], \
LPSAAModel[nf][tsteps-1].variables.type.continuous) \
for j in xrange(len(mC["temp_var_insts"])) ])
LPSAAModel[nf][tsteps-1].variables.set_lower_bounds( \
zip(mC["temp_var_insts"], polpicks[nf]) )
LPSAAModel[nf][tsteps-1].variables.set_upper_bounds( \
zip(mC["temp_var_insts"], polpicks[nf]) )
LPSAAModel[nf][tsteps-1].set_problem_type(LPSAAModel[nf][tsteps-1].problem_type.LP)
try:
LPSAAModel[nf][tsteps-1].solve()
lpobjval[nf] = LPSAAModel[nf][tsteps-1].solution.get_objective_value()
#DEBUG begin
## slaks = LPSAAModel[nf][tsteps-1].solution.get_values(mC["slak_var_names"])
## print "avg # of slacks: "+str(len([i for i in slaks if int(round(i,4)) != 0]) / \
## float(len(lpoutageattr[nf].values()[0])) )
#DEBUG end
except:
print "testADP: test LP SAA model infeasible"
LPSAAModel[nf][tsteps-1].conflict.refine(LPSAAModel[nf][tsteps-1].conflict.all_constraints())
conflicts = LPSAAModel[nf][tsteps-1].conflict.get()
conflicts = [i for i in xrange(len(conflicts)) if conflicts[i] != -1]
cgs = LPSAAModel[nf][tsteps-1].conflict.get_groups(conflicts)
ubcs=[j[0][1] for i,j in cgs if j[0][0] == 2]
lbcs=[j[0][1] for i,j in cgs if j[0][0] == 1]
lccs=[j[0][1] for i,j in cgs if j[0][0] == 3]
constrByVar = exeADP.find_constr_4_vars(LPSAAModel[nf][tsteps-1], \
LPSAAModel[nf][tsteps-1].variables.get_names())
conflConstrs = exeADP.find_constr_4_vars(LPSAAModel[nf][tsteps-1], \
LPSAAModel[nf][tsteps-1].linear_constraints.get_names(lccs), \
vartype="constraint")
__dbg = raw_input("execution halted, press ENTER to exit")
sys.exit()
# this is INCREDIBLY slow, and there's tsteps*numforecasts models, so we'll sacrifice memory for now
## exeADP.trash_model([k for l in LPSAAModel for k in l])
## exeADP.trash_model([k for l in policyEvalModels for k in l])
return polpicks, lpobjval
def run_one_big_LPSAA(numforecasts, mC, tsteps, adpftdmgattr, adpoutageattr, features, lpoutageattr):
policyEvalModels = [None for nf in xrange(numforecasts)]
polpicks = [None for nf in xrange(numforecasts)]
lpobjval = [None for nf in xrange(numforecasts)]
LPSAAModel = exeADP.create_models_temp_framework(tsteps, mC)
nx.set_edge_attributes(mC["netobj"].dinet, "t"+str(tsteps-1)+"_capacity", lpoutageattr)
exeADP.update_models_orig_scens(LPSAAModel, mC, numscens=len(lpoutageattr.values()[0]), mrange=[tsteps-1], \
reset=True, updateConstr=True)
LPSAAModel[tsteps-1].variables.set_types([ (mC["temp_var_insts"][j], \
LPSAAModel[tsteps-1].variables.type.continuous) \
for j in xrange(len(mC["temp_var_insts"])) ])
LPSAAModel[tsteps-1].set_problem_type(LPSAAModel[tsteps-1].problem_type.LP)
# find multiple policy evaluations for robustness
for nf in xrange(numforecasts):
# step 5: evaluation of policy
policyEvalModels[nf] = exeADP.create_models_temp_framework(tsteps, mC)
# override netobj's info with the stored scenario data
for t in xrange(tsteps):
nx.set_edge_attributes(mC["netobj"].dinet, "t"+str(t)+"_dmg_pct", adpftdmgattr[nf][t])
nx.set_edge_attributes(mC["netobj"].dinet, "t"+str(t)+"_capacity", adpoutageattr[nf][t])
mC["mcsim"].eval_adp_basis_functions(mC["netobj"], tsteps, mC)
featurelist = mC["mcsim"].calc_feature_ctrmassquad(mC["netobj"], tsteps, features)
exeADP.update_models_orig_scens(policyEvalModels[nf], mC, mrange=xrange(len(policyEvalModels[nf])), \
reset=True, updateConstr=True)
# solve policy search problem to find install arcs
mC["instchoices"] = [ [0]*len(mC["temp_var_insts"]) for i in xrange(tsteps)]
mC["previnsts"] = [[] for i in xrange(tsteps)]
for t in xrange(tsteps):
#TODO6: this may be semantically out of date
optpolicy,tflows,slks,bov,vfav = exeADP.adp_policy_search(t, featurelist, policyEvalModels[nf], mC, \
randpol=False)
# this is all we care about -- the policy evaluation at the last time step (i.e. all installed arcs)
polpicks[nf] = optpolicy
# step 6: fix the temp arcs to the decisions made in 5 and run an LP SAA over these samples at time T
LPSAAModel[tsteps-1].variables.set_lower_bounds( \
[ (mC["temp_var_insts"][i], polpicks[nf][i]) for i in xrange(len(mC["temp_var_insts"])) ])
LPSAAModel[tsteps-1].variables.set_upper_bounds( \
[ (mC["temp_var_insts"][i], polpicks[nf][i]) for i in xrange(len(mC["temp_var_insts"])) ])
try:
LPSAAModel[tsteps-1].solve()
lpobjval[nf] = LPSAAModel[tsteps-1].solution.get_objective_value()
except:
print "testADP: test LP SAA model infeasible"
LPSAAModel[tsteps-1].conflict.refine(LPSAAModel[tsteps-1].conflict.all_constraints())
conflicts = LPSAAModel[tsteps-1].conflict.get()
conflicts = [i for i in xrange(len(conflicts)) if conflicts[i] != -1]
cgs = LPSAAModel[tsteps-1].conflict.get_groups(conflicts)
ubcs=[j[0][1] for i,j in cgs if j[0][0] == 2]
lbcs=[j[0][1] for i,j in cgs if j[0][0] == 1]
lccs=[j[0][1] for i,j in cgs if j[0][0] == 3]
constrByVar = exeADP.find_constr_4_vars(LPSAAModel[tsteps-1], \
LPSAAModel[tsteps-1].variables.get_names())
conflConstrs = exeADP.find_constr_4_vars(LPSAAModel[tsteps-1], \
LPSAAModel[tsteps-1].linear_constraints.get_names(lccs), \
vartype="constraint")
__dbg = raw_input("execution halted, press ENTER to exit")
sys.exit()
exeADP.trash_model(LPSAAModel)
exeADP.trash_model([k for l in policyEvalModels for k in l])
return polpicks, lpobjval
def run_LPs_per_forecast(numforecasts, mC, tsteps, adpftdmgattr, adpoutageattr, features, lpoutageattr, \
poleval="adp"):
policyEvalModels = [None for nf in xrange(numforecasts)]
LPSAAModel = [None for nf in xrange(numforecasts)]
lpobjval = [None for nf in xrange(numforecasts)]
polpicks = [None for nf in xrange(numforecasts)]
# find multiple policy evaluations for robustness
for nf in xrange(numforecasts):
# step 1: evaluation of policy
if poleval == "alt":
netobj = copy.deepcopy(mC["netobj"])
samplesize = len(lpoutageattr[0].values()[0])
rankedarcs = exeADP.alt_vfa_calculation(tsteps, mC, netobj,
features, numforecasts,
adpftdmgattr)
policies = exeADP.alt_control_step(tsteps, mC, netobj, samplesize,
rankedarcs)
polpicks[nf] = policies[-1]
elif poleval == "adp":
policyEvalModels[nf] = exeADP.create_models_temp_framework(
tsteps, mC)
# override netobj's info with the stored scenario data
for t in xrange(tsteps):
nx.set_edge_attributes(mC["netobj"].dinet,
"t" + str(t) + "_dmg_pct",
adpftdmgattr[nf][t])
nx.set_edge_attributes(mC["netobj"].dinet,
"t" + str(t) + "_capacity",
adpoutageattr[nf][t])
# solve policy search problem to find install arcs
mC["mcsim"].eval_adp_basis_functions(mC["netobj"], tsteps, mC)
featurelist = mC["mcsim"].calc_feature_ctrmassquad(
mC["netobj"], tsteps, features)
if "ud_vfa_combined" in mC["polmethod"] or "node_constr" in mC[
"polmethod"]:
exeADP.update_models_orig_scens(policyEvalModels[nf], mC, mrange=xrange(len(policyEvalModels[nf])), \
reset=True, updateConstr=True)
elif "ud_vfa_separate" in mC["polmethod"]:
exeADP.update_models_orig_scens(policyEvalModels[nf], mC, mrange=xrange(len(policyEvalModels[nf])), \
reset=True, updateConstr=False)
else:
print "run_LPs_per_forecast: invalid policy method specified."
__dbg = raw_input("execution halted, press ENTER to exit")
sys.exit()
mC["instchoices"] = [[0] * len(mC["temp_var_insts"])
for i in xrange(tsteps)]
mC["previnsts"] = [[] for i in xrange(tsteps)]
for t in xrange(tsteps):
optpolicy,tflows,slks,bov,vfav = exeADP.adp_policy_search(t, featurelist, policyEvalModels[nf], mC,\
randpol=False)
# this is all we care about -- the policy evaluation at the last time step (i.e. all installed arcs)
polpicks[nf] = optpolicy
#DEBUG begin
## idx = [ i for i in xrange(len(optpolicy)) if int(round(optpolicy[i]))]
## instnames = [mC["temp_var_insts"][i] for i in idx]
## ltdict = dict()
## for ins in instnames:
## crds = exeADP.convert_cplex_structs(cpobj=ins)
## if mC["netobj"].tnet.edge[crds[0]][crds[1]]["linetype"] not in ltdict:
## ltdict[mC["netobj"].tnet.edge[crds[0]][crds[1]]["linetype"]] = 1
## else:
## ltdict[mC["netobj"].tnet.edge[crds[0]][crds[1]]["linetype"]] += 1
## print "\nLinetype counts for forecast "+str(nf)
## for k,v in ltdict.iteritems():
## print k+": "+str(v)
## print "policy eval slacks: "+str(len([i for i in slks if int(round(i,4)) != 0]))
#DEBUG end
# step 2: fix the temp arcs to the decisions and run an LP SAA over these samples at time T
LPSAAModel[nf] = exeADP.create_models_temp_framework(tsteps, mC)
nx.set_edge_attributes(mC["netobj"].dinet,
"t" + str(tsteps - 1) + "_capacity",
lpoutageattr[nf])
exeADP.update_models_orig_scens(LPSAAModel[nf], mC, numscens=len(lpoutageattr[nf].values()[0]), \
mrange=[tsteps-1], reset=True, updateConstr=True)
LPSAAModel[nf][tsteps-1].variables.set_types([ (mC["temp_var_insts"][j], \
LPSAAModel[nf][tsteps-1].variables.type.continuous) \
for j in xrange(len(mC["temp_var_insts"])) ])
LPSAAModel[nf][tsteps-1].variables.set_lower_bounds( \
zip(mC["temp_var_insts"], polpicks[nf]) )
LPSAAModel[nf][tsteps-1].variables.set_upper_bounds( \
zip(mC["temp_var_insts"], polpicks[nf]) )
LPSAAModel[nf][tsteps - 1].set_problem_type(
LPSAAModel[nf][tsteps - 1].problem_type.LP)
try:
LPSAAModel[nf][tsteps - 1].solve()
lpobjval[nf] = LPSAAModel[nf][tsteps -
1].solution.get_objective_value()
#DEBUG begin
## slaks = LPSAAModel[nf][tsteps-1].solution.get_values(mC["slak_var_names"])
## print "avg # of slacks: "+str(len([i for i in slaks if int(round(i,4)) != 0]) / \
## float(len(lpoutageattr[nf].values()[0])) )
#DEBUG end
except:
print "testADP: test LP SAA model infeasible"
LPSAAModel[nf][tsteps - 1].conflict.refine(
LPSAAModel[nf][tsteps - 1].conflict.all_constraints())
conflicts = LPSAAModel[nf][tsteps - 1].conflict.get()
conflicts = [
i for i in xrange(len(conflicts)) if conflicts[i] != -1
]
cgs = LPSAAModel[nf][tsteps - 1].conflict.get_groups(conflicts)
ubcs = [j[0][1] for i, j in cgs if j[0][0] == 2]
lbcs = [j[0][1] for i, j in cgs if j[0][0] == 1]
lccs = [j[0][1] for i, j in cgs if j[0][0] == 3]
constrByVar = exeADP.find_constr_4_vars(LPSAAModel[nf][tsteps-1], \
LPSAAModel[nf][tsteps-1].variables.get_names())
conflConstrs = exeADP.find_constr_4_vars(LPSAAModel[nf][tsteps-1], \
LPSAAModel[nf][tsteps-1].linear_constraints.get_names(lccs), \
vartype="constraint")
__dbg = raw_input("execution halted, press ENTER to exit")
sys.exit()
# this is INCREDIBLY slow, and there's tsteps*numforecasts models, so we'll sacrifice memory for now
## exeADP.trash_model([k for l in LPSAAModel for k in l])
## exeADP.trash_model([k for l in policyEvalModels for k in l])
return polpicks, lpobjval
def run_one_big_LPSAA(numforecasts, mC, tsteps, adpftdmgattr, adpoutageattr,
features, lpoutageattr):
policyEvalModels = [None for nf in xrange(numforecasts)]
polpicks = [None for nf in xrange(numforecasts)]
lpobjval = [None for nf in xrange(numforecasts)]
LPSAAModel = exeADP.create_models_temp_framework(tsteps, mC)
nx.set_edge_attributes(mC["netobj"].dinet,
"t" + str(tsteps - 1) + "_capacity", lpoutageattr)
exeADP.update_models_orig_scens(LPSAAModel, mC, numscens=len(lpoutageattr.values()[0]), mrange=[tsteps-1], \
reset=True, updateConstr=True)
LPSAAModel[tsteps-1].variables.set_types([ (mC["temp_var_insts"][j], \
LPSAAModel[tsteps-1].variables.type.continuous) \
for j in xrange(len(mC["temp_var_insts"])) ])
LPSAAModel[tsteps - 1].set_problem_type(LPSAAModel[tsteps -
1].problem_type.LP)
# find multiple policy evaluations for robustness
for nf in xrange(numforecasts):
# step 5: evaluation of policy
policyEvalModels[nf] = exeADP.create_models_temp_framework(tsteps, mC)
# override netobj's info with the stored scenario data
for t in xrange(tsteps):
nx.set_edge_attributes(mC["netobj"].dinet,
"t" + str(t) + "_dmg_pct",
adpftdmgattr[nf][t])
nx.set_edge_attributes(mC["netobj"].dinet,
"t" + str(t) + "_capacity",
adpoutageattr[nf][t])
mC["mcsim"].eval_adp_basis_functions(mC["netobj"], tsteps, mC)
featurelist = mC["mcsim"].calc_feature_ctrmassquad(
mC["netobj"], tsteps, features)
exeADP.update_models_orig_scens(policyEvalModels[nf], mC, mrange=xrange(len(policyEvalModels[nf])), \
reset=True, updateConstr=True)
# solve policy search problem to find install arcs
mC["instchoices"] = [[0] * len(mC["temp_var_insts"])
for i in xrange(tsteps)]
mC["previnsts"] = [[] for i in xrange(tsteps)]
for t in xrange(tsteps):
#TODO6: this may be semantically out of date
optpolicy,tflows,slks,bov,vfav = exeADP.adp_policy_search(t, featurelist, policyEvalModels[nf], mC, \
randpol=False)
# this is all we care about -- the policy evaluation at the last time step (i.e. all installed arcs)
polpicks[nf] = optpolicy
# step 6: fix the temp arcs to the decisions made in 5 and run an LP SAA over these samples at time T
LPSAAModel[tsteps-1].variables.set_lower_bounds( \
[ (mC["temp_var_insts"][i], polpicks[nf][i]) for i in xrange(len(mC["temp_var_insts"])) ])
LPSAAModel[tsteps-1].variables.set_upper_bounds( \
[ (mC["temp_var_insts"][i], polpicks[nf][i]) for i in xrange(len(mC["temp_var_insts"])) ])
try:
LPSAAModel[tsteps - 1].solve()
lpobjval[nf] = LPSAAModel[tsteps -
1].solution.get_objective_value()
except:
print "testADP: test LP SAA model infeasible"
LPSAAModel[tsteps - 1].conflict.refine(
LPSAAModel[tsteps - 1].conflict.all_constraints())
conflicts = LPSAAModel[tsteps - 1].conflict.get()
conflicts = [
i for i in xrange(len(conflicts)) if conflicts[i] != -1
]
cgs = LPSAAModel[tsteps - 1].conflict.get_groups(conflicts)
ubcs = [j[0][1] for i, j in cgs if j[0][0] == 2]
lbcs = [j[0][1] for i, j in cgs if j[0][0] == 1]
lccs = [j[0][1] for i, j in cgs if j[0][0] == 3]
constrByVar = exeADP.find_constr_4_vars(LPSAAModel[tsteps-1], \
LPSAAModel[tsteps-1].variables.get_names())
conflConstrs = exeADP.find_constr_4_vars(LPSAAModel[tsteps-1], \
LPSAAModel[tsteps-1].linear_constraints.get_names(lccs), \
vartype="constraint")
__dbg = raw_input("execution halted, press ENTER to exit")
sys.exit()
exeADP.trash_model(LPSAAModel)
exeADP.trash_model([k for l in policyEvalModels for k in l])
return polpicks, lpobjval