def opt_cut(model, where):
if where == GRB.Callback.MIPSOL: # for a given integer first stage solution
startTime = time.time()
y0_param_preprocessed = model.cbGetSolution(y0) #{y[i, j, k]}
y0_param = {}
y0_key = "" # cant hash with a dict type (y0_param), so transform it to a str
for i, j in A:
for k in K:
y0_param[i, j, k] = round(y0_param_preprocessed[i, j, k])
y0_key += str(y0_param[i, j, k])
theta_tongo = model.cbGetSolution(theta)
# calculate FSC(y^)
FSC = sum(lc[(arco, k)] * y0_param[arco[0], arco[1], k] for arco in AF
for k in K)
# print("FSC: {}, theta tongo {}".format(FSC, theta_tongo))
if partial:
if y0_key not in next_gap_to_solve.keys():
next_gap_to_solve[y0_key] = gaps_to_assign[0]
# if for this solution Q hasnt been solved
if (not partial and y0_key not in int_Q_hash.keys()) or (
partial and next_gap_to_solve[y0_key] != None):
startTime = time.time()
Qs_int_calculated = {}
Qs_bound_int = {}
Qs_gaps_int = {}
for s in S:
#update
if not continuous_cargo:
ss6, ss7, ss10, ss11 = update_model_Qs(
const_ss, int_sat, y_sat, x_sat, w_sat, q_sat,
zplus_sat, r_sat, days, s, airports, Nint, Nf, Nl, N,
AF, AG, A, K, nav, n, av, air_cap, air_vol, Cargo, OD,
size, vol, ex, mand, cv, cf, ch, inc, lc, sc, delta, V,
gap, tv)
const_ss[6], const_ss[7], const_ss[10], const_ss[
11] = ss6, ss7, ss10, ss11
else:
ss5_list, ss7 = update_model_Qs_ContFlow(
const_ss, int_sat, y_sat, f_sat, zplus_sat, r_sat,
days, s, airports, Nint, Nf, Nl, N, AF, AG, A, K, nav,
n, av, air_cap, air_vol, Cargo, OD, size, vol, ex,
mand, cv, cf, ch, inc, lc, sc, delta, V, gap, tv,
volperkg, incperkg)
const_ss[5], const_ss[7] = ss5_list, ss7
#solve
if partial:
Qs_int_calculated[s], Qs_bound_int[s], Qs_gaps_int[
s] = solve_integer_Qs_SingleSat(
y0_param,
s,
A,
K,
int_r1,
int_sat,
partial_gap=next_gap_to_solve[y0_key])
else:
Qs_int_calculated[s], Qs_bound_int[
s] = solve_integer_Qs_SingleSat(
y0_param, s, A, K, int_r1, int_sat)
if partial:
Q_av_gaps_int = sum(Qs_gaps_int[s] for s in S) / len(S)
if Q_av_gaps_int <= gaps_to_assign[-1]: # av <= 1%
next_gap_to_solve[y0_key] = None # finish
else:
next_gap_to_solve[y0_key] = max(
gap for gap in gaps_to_assign
if gap < Q_av_gaps_int) # the next smaller gap
Q_int_calculated = sum(Qs_int_calculated[s] for s in S) / len(S)
Q_int_bound = sum(Qs_bound_int[s] for s in S) / len(S)
# print("Q_int_calculated: ", Q_int_calculated)
theta_Q.append((theta_tongo, Q_int_calculated))
int_Q_hash[y0_key] = Q_int_calculated
int_sol_hash[y0_key] = y0_param
comment = "QTbound didnt cut; solved Q"
# adding feasible solution (y tongo, Q(y tongo))
this_value = Q_int_calculated - FSC
if this_value > model._current_incumbent:
# print("*** solution better than incumbent ***")
model._sol_to_add = model.cbGetSolution(model.getVars())
model._theta_to_add = Q_int_calculated
model._current_incumbent = this_value
if Q_int_calculated > L:
print(
"\n########## ERROR: Q(y tongo)={:.0f} > {:.0f}=L ##########\n"
.format(Q_int_calculated, L))
if theta_tongo > Q_int_calculated + epsilon:
# print(" ##Theta tongo {:.0f} > Q(y tongo) {:.0f}, adding INTEGER optimality cut##".format(theta_tongo, Q_int_calculated))
SUP_y = [(i, j, k) for i, j in AF for k in K
if y0_param[i, j, k] > 0.5]
not_SUP_y = [(i, j, k) for i, j in AF for k in K
if y0_param[i, j, k] < 0.5]
delta_callback = quicksum(1 - y[0][i, j, k]
for i, j, k in SUP_y) + quicksum(
y[0][i, j, k]
for i, j, k in not_SUP_y)
if partial:
model.cbLazy(theta <= (L - Q_int_bound) * delta_callback +
Q_int_bound)
else:
model.cbLazy(
theta <= (L - Q_int_calculated) * delta_callback +
Q_int_calculated)
model._Q_cblazy_added += 1
else:
comment = comment + ", Q didnt cut; not added"
# print("**** Theta tongo {:.0f} <= Q(y tongo) {:.0f}, NO NEW INTEGER CUT TO ADD ****".format(theta_tongo, Q_int_calculated))
theta_int_values.append(theta_tongo)
Q_int_values.append(Q_int_calculated)
runTime = time.time() - startTime
int_cut_times.append(runTime)
if write_cuts:
wrt.write_cuts(logs_dir,
model.ModelName,
FSC,
theta_tongo,
Q_int_calculated,
runTime,
integer=True,
QT_bound=False,
comment=comment)
elif (not partial and y0_key in int_Q_hash.keys()) or (
partial and next_gap_to_solve[y0_key] == None):
Q_int_calculated = int_Q_hash[y0_key] # for plotting
if theta_tongo > Q_int_calculated + epsilon: # add the cut if the first theta wasnt the optimal for the schedule
SUP_y = [(i, j, k) for i, j in AF for k in K
if y0_param[i, j, k] > 0.5]
not_SUP_y = [(i, j, k) for i, j in AF for k in K
if y0_param[i, j, k] < 0.5]
delta_callback = quicksum(1 - y[0][i, j, k]
for i, j, k in SUP_y) + quicksum(
y[0][i, j, k]
for i, j, k in not_SUP_y)
model.cbLazy(theta <= (L - Q_int_calculated) * delta_callback +
Q_int_calculated)
model._Q_cblazy_added += 1
theta_int_values.append(theta_tongo)
Q_int_values.append(Q_int_calculated)
# print("***** VISITED SOLUTION TWICE, NOTHING TO SOLVE. Q(y) = {:.0f} *****".format(Q_int_calculated))
# adding the feasible solution as the incumbent for the node
elif where == GRB.Callback.MIPNODE:
if len(model._sol_to_add) > 0:
# print("*** ADDING FEASIBLE SOLUTION ***")
model.cbSetSolution(model.getVars(), model._sol_to_add)
model.cbSetSolution(theta, model._theta_to_add)
model._sol_to_add.clear()
model._theta_to_add = 0
#solve
Qs_bound_int = {}
Qs_int_calculated = {}
times_list = []
Q_list = []
for s in S_OUTSAMPLE:
this_StarTime = time.time()
ss6, ss7, ss10, ss11 = update_model_Qs(
const_ss, int_sat, y_sat, x_sat, w_sat, q_sat, zplus_sat, r_sat,
days, s, airports, Nint, Nf, Nl, N, AF, AG, A, K, nav, n, av,
air_cap, air_vol, Cargo_base, OD_base, size_OUTSAMPLE,
vol_OUTSAMPLE, ex_OUTSAMPLE, mand_OUTSAMPLE, cv, cf, ch,
inc_OUTSAMPLE, lc, sc_OUTSAMPLE, delta, V, gap, tv)
const_ss[6], const_ss[7], const_ss[10], const_ss[
11] = ss6, ss7, ss10, ss11
Qs_int_calculated[s], Qs_bound_int[s], _ = solve_integer_Qs_SingleSat(
y0_param, s, A, K, int_r1, int_sat, partial_gap=gap_outsample)
this_RunTime = round(time.time() - this_StarTime, 2)
times_list.append(this_RunTime)
Q_list.append(round(Qs_int_calculated[s], 2))
print(" Scenario OUTSAMPLE {}: Q = {:.2f}, time = {}".format(
s, Qs_int_calculated[s], this_RunTime))
Q_int_calculated = sum(Qs_int_calculated[s]
for s in S_OUTSAMPLE) / len(S_OUTSAMPLE)
performance_RunTime = round(time.time() - performance_StartTime, 2)
sd = round(st.stdev(Q_list), 2)
print(
f"Q mean : {st.mean(Q_list)}, min: {min(Q_list)}, max: {max(Q_list)}, SD: {sd}, Percentage Variation: {round(sd/st.mean(Q_list)*100, 2)}%"
)
print(
f"Real second stage value for the schedule: {round(Q_int_calculated, 2)}"
)
def opt_cut(model, where):
if where == GRB.Callback.MIPSOL: # for a given integer first stage solution
startTime = time.time()
y0_param_preprocessed = model.cbGetSolution(y0) #{y[i, j, k]}
y0_param = {}
y0_key = "" # cant hash with a dict type (y0_param), so transform it to a str
for i,j in A:
for k in K:
y0_param[i,j,k] = round(y0_param_preprocessed[i,j,k])
y0_key += str(y0_param[i,j,k])
theta_tongo = model.cbGetSolution(theta)
# calculate FSC(y^)
FSC = sum(lc[(arco, k)]*y0_param[arco[0], arco[1], k] for arco in AF for k in K)
# print("FSC: {}, theta tongo {}".format(FSC, theta_tongo))
if partial:
if y0_key not in next_gap_to_solve.keys():
next_gap_to_solve[y0_key] = gaps_to_assign[0]
# if for this solution Q hasnt been solved
if (not partial and y0_key not in int_Q_hash.keys()) or (partial and next_gap_to_solve[y0_key] != None):
# print("New solution, solving QTbound(y)")
# predict a bound from a previous solution
QT_bound = check_best_bound(y0_param, int_sol_hash, int_Q_bound_hash, sc, V, gap, tv, AF, K, S)
# print("QT_bound: ", QT_bound)
# hash the QTbound if its for a new sol or is better than the previous
if y0_key not in QTbound_hash.keys():
QTbound_hash[y0_key] = QT_bound
elif QT_bound < QTbound_hash[y0_key]:
QTbound_hash[y0_key] = QT_bound
comment = ""
# if doesnt cut current solution (y, theta), solve Q
if not theta_tongo > QT_bound + epsilon:
# print("QTbound didnt cut, solving Q")
Qs_int_calculated = {}
Qs_bound_int = {}
Qs_gaps_int = {}
for s in S:
if not continuous_cargo:
ss6, ss7, ss10, ss11 = update_model_Qs(const_ss, int_sat, y_sat, x_sat, w_sat, q_sat, zplus_sat, r_sat, days, s, airports, Nint, Nf, Nl, N, AF, AG, A, K, nav, n, av, air_cap, air_vol, Cargo, OD, size, vol, ex, mand, cv, cf, ch, inc, lc, sc, delta, V, gap, tv)
const_ss[6], const_ss[7], const_ss[10], const_ss[11] = ss6, ss7, ss10, ss11
else:
ss5_list, ss7 = update_model_Qs_ContFlow(const_ss, int_sat, y_sat, f_sat, zplus_sat, r_sat, days, s, airports, Nint, Nf, Nl, N, AF, AG, A, K, nav, n, av, air_cap, air_vol, Cargo, OD, size, vol, ex, mand, cv, cf, ch, inc, lc, sc, delta, V, gap, tv, volperkg, incperkg)
const_ss[5], const_ss[7] = ss5_list, ss7
if partial:
Qs_int_calculated[s], Qs_bound_int[s], Qs_gaps_int[s] = solve_integer_Qs_SingleSat(y0_param, s, A, K, int_r1, int_sat, partial_gap=next_gap_to_solve[y0_key])
else:
Qs_int_calculated[s], Qs_bound_int[s] = solve_integer_Qs_SingleSat(y0_param, s, A, K, int_r1, int_sat)
if partial:
Q_av_gaps_int = sum(Qs_gaps_int[s] for s in S)/len(S)
if Q_av_gaps_int <= gaps_to_assign[-1]: # av <= 1%
next_gap_to_solve[y0_key] = None # finish
else:
next_gap_to_solve[y0_key] = max(gap for gap in gaps_to_assign if gap < Q_av_gaps_int) # the next smaller gap
Q_int_calculated = sum(Qs_int_calculated[s] for s in S)/len(S)
Q_int_bound = sum(Qs_bound_int[s] for s in S)/len(S)
# print("Q_int_calculated: ", Q_int_calculated)
theta_Q.append( (theta_tongo, Q_int_calculated) )
int_Q_hash[y0_key] = Q_int_calculated
int_Q_bound_hash[y0_key] = Q_int_bound
int_sol_hash[y0_key] = y0_param # to calculate T(y02,y01)
comment = "QTbound didnt cut; solved Q"
# adding feasible solution (y tongo, Q(y tongo))
this_value = Q_int_calculated - FSC
if this_value > model._current_incumbent:
# print("*** solution better than incumbent ***")
model._sol_to_add = model.cbGetSolution(model.getVars())
model._theta_to_add = Q_int_calculated
model._current_incumbent = this_value
if Q_int_calculated > L:
print("\n########## ERROR: Q(y tongo)={:.0f} > {:.0f}=L ##########\n".format(Q_int_calculated, L))
if theta_tongo > Q_int_calculated + epsilon:
# print(" ##Theta tongo {:.0f} > Q(y tongo) {:.0f}, adding INTEGER optimality cut##".format(theta_tongo, Q_int_calculated))
SUP_y = [(i,j,k) for i,j in AF for k in K if y0_param[i, j, k] > 0.5]
not_SUP_y = [(i,j,k) for i,j in AF for k in K if y0_param[i, j, k] < 0.5]
delta_callback = quicksum(1 - y[0][i, j, k] for i,j,k in SUP_y) + quicksum(y[0][i, j, k] for i,j,k in not_SUP_y)
if partial:
model.cbLazy(theta <= (L - Q_int_bound)*delta_callback + Q_int_bound)
else:
model.cbLazy(theta <= (L - Q_int_calculated)*delta_callback + Q_int_calculated)
model._Q_cblazy_added += 1
else:
comment = comment + ", Q didnt cut; not added"
# print("**** Theta tongo {:.0f} <= Q(y tongo) {:.0f}, NO NEW INTEGER CUT TO ADD ****".format(theta_tongo, Q_int_calculated))
theta_int_values.append(theta_tongo)
Q_int_values.append(Q_int_calculated)
runTime = time.time() - startTime
int_cut_times.append(runTime)
if write_cuts:
wrt.write_cuts(logs_dir, model.ModelName, FSC, theta_tongo, Q_int_calculated, runTime, integer=True, QT_bound=False, comment=comment)
# GLOBAL CUTS
proactive_startTime = time.time()
# for every operated flight
for i_bar, j_bar in AF:
for k_bar in K:
if y0_param[i_bar, j_bar, k_bar] == 1:
# similar flights to the observed one
Similar = generate_similar_flights(y0_param, i_bar, j_bar, k_bar, AF, airports, K)
if Similar != []:
# QTb in this vicinity
Q_y2_p = Q_int_calculated
T_y2_y1_p = sum(sc[s][(i_bar, j_bar), k_bar] for s in S)/len(S)
QT_bound_p = Q_y2_p + T_y2_y1_p
#add global cut
Omega_1 = [(i,j,k) for i,j in AF for k in K if (y0_param[i, j, k] > 0.5 and i,j not in Similar)]
Omega_2 = [(i,j,k) for i,j in AF for k in K if (y0_param[i, j, k] < 0.5 and i,j not in Similar)]
Delta_1 = quicksum(1 - y[0][i, j, k] for i,j,k in Omega_1) + quicksum(y[0][i, j, k] for i,j,k in Omega_2)
Delta_2 = quicksum(y[0][i, j, k] for i,j in Similar for k in K if k != k_bar) + quicksum(y[0][i_bar, j_bar, k] for k in K)
a_hat, b_hat = bounds_for_similar(Similar)
Delta_3 = 1 - nu_two[a_hat][b_hat, k_bar]
# GC constraints
for i,j in Similar:
#GC1
model.cbLazy(y[0][i, j, k_bar] <= nu_one[a_hat][b_hat, k_bar])
#GC2
model.cbLazy(nu_one[a_hat][b_hat, k_bar] <= quicksum(y[0][i, j, k_bar] for i,j in Similar))
#GC3
model.cbLazy(2*nu_one[a_hat][b_hat, k_bar] - quicksum(y[0][i, j, k_bar] for i,j in Similar) <= nu_two[a_hat][b_hat, k_bar])
#global cut
model.cbLazy(theta <= QT_bound_p + (Delta_1 + Delta_2 + Delta_3)*L)
model._QTbound_global_cuts += 1
proactive_runTime = time.time() - proactive_startTime
proactive_bound_times.append(proactive_runTime)
# if it cuts the current solution, add the QTbound
else:
SUP_y = [(i,j,k) for i,j in AF for k in K if y0_param[i, j, k] > 0.5]
not_SUP_y = [(i,j,k) for i,j in AF for k in K if y0_param[i, j, k] < 0.5]
delta_callback = quicksum(1 - y[0][i, j, k] for i,j,k in SUP_y) + quicksum(y[0][i, j, k] for i,j,k in not_SUP_y)
model.cbLazy(theta <= (L - QT_bound)*delta_callback + QT_bound)
model._QTbound_cblazy_added += 1
runTime_bound = time.time() - startTime
bound_times.append(runTime_bound)
if write_cuts:
wrt.write_cuts(logs_dir, model.ModelName, FSC, theta_tongo, QT_bound, runTime_bound, integer=True, QT_bound=True, comment=comment)
elif (not partial and y0_key in int_Q_hash.keys()) or (partial and next_gap_to_solve[y0_key] == None):
Q_int_calculated = int_Q_hash[y0_key] # for plotting
if theta_tongo > Q_int_calculated + epsilon: # add the cut if the first theta wasnt the optimal for the schedule
SUP_y = [(i,j,k) for i,j in AF for k in K if y0_param[i, j, k] > 0.5]
not_SUP_y = [(i,j,k) for i,j in AF for k in K if y0_param[i, j, k] < 0.5]
delta_callback = quicksum(1 - y[0][i, j, k] for i,j,k in SUP_y) + quicksum(y[0][i, j, k] for i,j,k in not_SUP_y)
model.cbLazy(theta <= (L - Q_int_calculated)*delta_callback + Q_int_calculated)
model._Q_cblazy_added += 1
theta_int_values.append(theta_tongo)
Q_int_values.append(Q_int_calculated)
# print("***** VISITED SOLUTION TWICE, NOTHING TO SOLVE. Q(y) = {:.0f} *****".format(Q_int_calculated))
# adding the feasible solution as the incumbent for the node
elif where == GRB.Callback.MIPNODE:
if len(model._sol_to_add) > 0:
# print("*** ADDING FEASIBLE SOLUTION ***")
model.cbSetSolution(model.getVars(), model._sol_to_add)
model.cbSetSolution(theta, model._theta_to_add)
model._sol_to_add.clear()
model._theta_to_add = 0