def large_neighborhood(model):
"""
Solves a set cover instance with large-neighborhood search.
In each call to the MIP solver we exclude a fixed ratio of the sets that
are currently unused.
Args:
model: The set cover MIP model as created by mip.create_model().
"""
# Time limit per MIP call in s
TIMELIMIT = 3
# Ratio of unused nodes to exclude
FIX_RATIO = 0.9
SAMPLE_SIZE = 30
FEASIBLE_COUNT = 0
INFEASIBLE_COUNT = 0
g_model, vars = model
g_model.setParam("OutputFlag", 0)
g_model.setParam("TimeLimit", len(vars)*0.05)
print("Processing " + g_model.getAttr("ModelName"))
# Warmup
g_model.optimize()
print("Initial solution: {0}".format(g_model.objval))
m.write(model)
#g_model.setParam("SolutionLimit", 2147483647)
g_model.setParam("TimeLimit", TIMELIMIT)
best_obj = g_model.objval
best_sol = {var:int(var.x) for var in vars}
obj_constraint = g_model.addConstr(g_model.getObjective() <= best_obj - 1)
if g_model.status != grb.GRB.OPTIMAL:
while g_model.status != grb.GRB.INTERRUPTED:
# Add the additional constraints as described above
added_constraints = []
for var in vars:
if best_sol[var] == 0 and random.random() < FIX_RATIO:
added_constraints.append(g_model.addConstr(var == best_sol[var]))
g_model.optimize()
# Remove the additional constraints again
for constraint in added_constraints:
g_model.remove(constraint)
#print g_model.status
if g_model.status != grb.GRB.INFEASIBLE:
FEASIBLE_COUNT += 1;
if g_model.objval < best_obj:
sys.stdout.write("\nNext solution: {0}\n".format(g_model.objval))
m.write(model, False)
best_obj = g_model.objval
best_sol = {var:int(var.x) for var in vars}
g_model.remove(obj_constraint)
obj_constraint = g_model.addConstr(g_model.getObjective() <= best_obj - 1)
FEASIBLE_COUNT = 0
INFEASIBLE_COUNT = 0
else:
sys.stdout.write('_')
else:
INFEASIBLE_COUNT += 1;
sys.stdout.write('.')
sys.stdout.flush()
if (FEASIBLE_COUNT+INFEASIBLE_COUNT) == SAMPLE_SIZE:
if FEASIBLE_COUNT > SAMPLE_SIZE * 0.2:
TIMELIMIT = TIMELIMIT*1.1
sys.stdout.write("(T {0})".format(TIMELIMIT))
else:
FIX_RATIO = FIX_RATIO - 0.01
sys.stdout.write("(R {0})".format(FIX_RATIO))
FEASIBLE_COUNT = 0
INFEASIBLE_COUNT = 0
#g_model.setParam("TimeLimit", float('inf'))
#g_model.optimize()
#if g_model.status == grb.GRB.OPTIMAL:
# print("Optimal solution: {0}".format(g_model.objval))
m.write(model)
def large_neighborhood(model):
"""
Solves a set cover instance with large-neighborhood search.
In each call to the MIP solver we exclude a fixed ratio of the sets that
are currently unused. (Maybe I should favor excluding
the sets with the highest cost/size ratio?)
Args:
model: The set cover MIP model as created by mip.create_model().
"""
g_model, vars = model
g_model.setParam("OutputFlag", 0)
g_model.setParam("TimeLimit", TIMELIMIT)
g_model.setParam("SolutionLimit", 1)
print ("Processing " + g_model.getAttr("ModelName"))
# Warmup
g_model.optimize()
print ("Initial solution: {0}".format(g_model.objval))
m.write(model)
g_model.setParam("SolutionLimit", 2147483647)
best_obj = g_model.objval
best_sol = {var: int(var.x) for var in vars}
obj_constraint = g_model.addConstr(g_model.getObjective() <= best_obj - 1)
if g_model.status != grb.GRB.OPTIMAL:
while g_model.status != grb.GRB.INTERRUPTED:
# Add the additional constraints as described above
added_constraints = []
for var in vars:
if best_sol[var] == 0 and random.random() < FIX_RATIO:
added_constraints.append(g_model.addConstr(var == 0))
g_model.optimize()
# Remove the additional constraints again
for constraint in added_constraints:
g_model.remove(constraint)
# print g_model.status
if g_model.status != grb.GRB.INFEASIBLE:
if g_model.objval < best_obj:
print ("\nNext solution: {0}".format(g_model.objval))
m.write(model, False)
best_obj = g_model.objval
best_sol = {var: int(var.x) for var in vars}
g_model.remove(obj_constraint)
obj_constraint = g_model.addConstr(g_model.getObjective() <= best_obj - 1)
else:
print "(ERROR)"
else:
sys.stdout.write(".")
sys.stdout.flush()
g_model.setParam("TimeLimit", float("inf"))
g_model.optimize()
if g_model.status == grb.GRB.OPTIMAL:
print ("Optimal solution: {0}".format(g_model.objval))
m.write(model)
def large_neighborhood(model):
"""
Solves a set cover instance with large-neighborhood search.
In each call to the MIP solver we exclude a fixed ratio of the sets that
are currently unused. (Maybe I should favor excluding
the sets with the highest cost/size ratio?)
Args:
model: The set cover MIP model as created by mip.create_model().
"""
g_model, vars = model
g_model.setParam("OutputFlag", 0)
g_model.setParam("TimeLimit", TIMELIMIT)
g_model.setParam("SolutionLimit", 1)
print("Processing " + g_model.getAttr("ModelName"))
# Warmup
g_model.optimize()
print("Initial solution: {0}".format(g_model.objval))
m.write(model)
g_model.setParam("SolutionLimit", 2147483647)
best_obj = g_model.objval
best_sol = {var: int(var.x) for var in vars}
obj_constraint = g_model.addConstr(g_model.getObjective() <= best_obj - 1)
if g_model.status != grb.GRB.OPTIMAL:
while g_model.status != grb.GRB.INTERRUPTED:
# Add the additional constraints as described above
added_constraints = []
for var in vars:
if best_sol[var] == 0 and random.random() < FIX_RATIO:
added_constraints.append(g_model.addConstr(var == 0))
g_model.optimize()
# Remove the additional constraints again
for constraint in added_constraints:
g_model.remove(constraint)
#print g_model.status
if g_model.status != grb.GRB.INFEASIBLE:
if g_model.objval < best_obj:
print("\nNext solution: {0}".format(g_model.objval))
m.write(model, False)
best_obj = g_model.objval
best_sol = {var: int(var.x) for var in vars}
g_model.remove(obj_constraint)
obj_constraint = g_model.addConstr(
g_model.getObjective() <= best_obj - 1)
else:
print "(ERROR)"
else:
sys.stdout.write('.')
sys.stdout.flush()
g_model.setParam("TimeLimit", float('inf'))
g_model.optimize()
if g_model.status == grb.GRB.OPTIMAL:
print("Optimal solution: {0}".format(g_model.objval))
m.write(model)
def large_neighborhood(model):
"""
Solves a set cover instance with large-neighborhood search.
In each call to the MIP solver we exclude a fixed ratio of the sets that
are currently unused.
Args:
model: The set cover MIP model as created by mip.create_model().
"""
# Time limit per MIP call in s
TIMELIMIT = 3
# Ratio of unused nodes to exclude
FIX_RATIO = 0.9
SAMPLE_SIZE = 30
FEASIBLE_COUNT = 0
INFEASIBLE_COUNT = 0
g_model, vars = model
g_model.setParam("OutputFlag", 0)
g_model.setParam("TimeLimit", len(vars) * 0.05)
print("Processing " + g_model.getAttr("ModelName"))
# Warmup
g_model.optimize()
print("Initial solution: {0}".format(g_model.objval))
m.write(model)
#g_model.setParam("SolutionLimit", 2147483647)
g_model.setParam("TimeLimit", TIMELIMIT)
best_obj = g_model.objval
best_sol = {var: int(var.x) for var in vars}
obj_constraint = g_model.addConstr(g_model.getObjective() <= best_obj - 1)
if g_model.status != grb.GRB.OPTIMAL:
while g_model.status != grb.GRB.INTERRUPTED:
# Add the additional constraints as described above
added_constraints = []
for var in vars:
if best_sol[var] == 0 and random.random() < FIX_RATIO:
added_constraints.append(
g_model.addConstr(var == best_sol[var]))
g_model.optimize()
# Remove the additional constraints again
for constraint in added_constraints:
g_model.remove(constraint)
#print g_model.status
if g_model.status != grb.GRB.INFEASIBLE:
FEASIBLE_COUNT += 1
if g_model.objval < best_obj:
sys.stdout.write("\nNext solution: {0}\n".format(
g_model.objval))
m.write(model, False)
best_obj = g_model.objval
best_sol = {var: int(var.x) for var in vars}
g_model.remove(obj_constraint)
obj_constraint = g_model.addConstr(
g_model.getObjective() <= best_obj - 1)
FEASIBLE_COUNT = 0
INFEASIBLE_COUNT = 0
else:
sys.stdout.write('_')
else:
INFEASIBLE_COUNT += 1
sys.stdout.write('.')
sys.stdout.flush()
if (FEASIBLE_COUNT + INFEASIBLE_COUNT) == SAMPLE_SIZE:
if FEASIBLE_COUNT > SAMPLE_SIZE * 0.2:
TIMELIMIT = TIMELIMIT * 1.1
sys.stdout.write("(T {0})".format(TIMELIMIT))
else:
FIX_RATIO = FIX_RATIO - 0.01
sys.stdout.write("(R {0})".format(FIX_RATIO))
FEASIBLE_COUNT = 0
INFEASIBLE_COUNT = 0
#g_model.setParam("TimeLimit", float('inf'))
#g_model.optimize()
#if g_model.status == grb.GRB.OPTIMAL:
# print("Optimal solution: {0}".format(g_model.objval))
m.write(model)
