def post_milp(self, x, nn, output_flag, t, template):
"""milp method"""
post = []
for chosen_action in range(2):
gurobi_model = grb.Model()
gurobi_model.setParam('OutputFlag', output_flag)
gurobi_model.setParam('Threads', 2)
input = Experiment.generate_input_region(gurobi_model, template, x,
self.env_input_size)
max_theta, min_theta, max_theta_dot, min_theta_dot = self.get_theta_bounds(
gurobi_model, input)
sin_cos_table = self.get_sin_cos_table(max_theta,
min_theta,
max_theta_dot,
min_theta_dot,
action=chosen_action)
feasible_action = CartpoleExperiment.generate_nn_guard(
gurobi_model, input, nn, action_ego=chosen_action)
if feasible_action:
thetaacc, xacc = CartpoleExperiment.generate_angle_milp(
gurobi_model, input, sin_cos_table)
# apply dynamic
x_prime = self.apply_dynamic(
input,
gurobi_model,
thetaacc=thetaacc,
xacc=xacc,
env_input_size=self.env_input_size)
gurobi_model.update()
gurobi_model.optimize()
found_successor, x_prime_results = self.h_repr_to_plot(
gurobi_model, template, x_prime)
if found_successor:
post.append(tuple(x_prime_results))
return post
def post_milp(self, x, nn, output_flag, t, template):
"""milp method"""
post = []
for chosen_action in range(2):
observable_template = self.observable_templates[chosen_action]
observable_result = self.observable_results[chosen_action]
gurobi_model = grb.Model()
gurobi_model.setParam('OutputFlag', output_flag)
gurobi_model.setParam('Threads', 2)
input = Experiment.generate_input_region(gurobi_model, template, x,
self.env_input_size)
observation = gurobi_model.addMVar(shape=(2, ),
lb=float("-inf"),
ub=float("inf"),
name="input")
gurobi_model.addConstr(
observation[1] <= input[0] - input[1] + self.input_epsilon / 2,
name=f"obs_constr21")
gurobi_model.addConstr(
observation[1] >= input[0] - input[1] - self.input_epsilon / 2,
name=f"obs_constr22")
gurobi_model.addConstr(
observation[0] <= input[2] - input[3] + self.input_epsilon / 2,
name=f"obs_constr11")
gurobi_model.addConstr(
observation[0] >= input[2] - input[3] - self.input_epsilon / 2,
name=f"obs_constr12")
# feasible_action = Experiment.generate_nn_guard(gurobi_model, observation, nn, action_ego=chosen_action)
# feasible_action = Experiment.generate_nn_guard(gurobi_model, input, nn, action_ego=chosen_action)
Experiment.generate_region_constraints(gurobi_model,
observable_template,
observation,
observable_result, 2)
gurobi_model.optimize()
feasible_action = gurobi_model.status == 2
if feasible_action:
# apply dynamic
# x_prime_results = self.optimise(template, gurobi_model, input) # h representation
# x_prime = Experiment.generate_input_region(gurobi_model, template, x_prime_results, self.env_input_size)
x_second = StoppingCarExperiment.apply_dynamic(
input,
gurobi_model,
action=chosen_action,
env_input_size=self.env_input_size)
gurobi_model.update()
gurobi_model.optimize()
found_successor, x_second_results = self.h_repr_to_plot(
gurobi_model, template, x_second)
if found_successor:
post.append(tuple(x_second_results))
return post
def post_milp(self, x, nn, output_flag, t, template):
"""milp method"""
post = []
for chosen_action in range(2):
observable_template = self.observable_templates[chosen_action]
observable_result = self.observable_results[chosen_action]
if USE_GUROBI:
gurobi_model = grb.Model()
gurobi_model.setParam('OutputFlag', output_flag)
gurobi_model.setParam('Threads', 2)
input = Experiment.generate_input_region(
gurobi_model, template, x, self.env_input_size)
Experiment.generate_region_constraints(
gurobi_model,
observable_template,
input,
observable_result,
env_input_size=self.env_input_size)
gurobi_model.optimize()
feasible_action = gurobi_model.status == 2
if feasible_action:
max_theta, min_theta, max_theta_dot, min_theta_dot = self.get_theta_bounds(
gurobi_model, input)
sin_cos_table = self.get_sin_cos_table(
max_theta,
min_theta,
max_theta_dot,
min_theta_dot,
action=chosen_action,
step_thetaacc=100)
x_prime_results = self.optimise(template, gurobi_model,
input) # h representation
x_prime = Experiment.generate_input_region(
gurobi_model, template, x_prime_results,
self.env_input_size)
thetaacc, xacc = CartpoleExperiment.generate_angle_milp(
gurobi_model, x_prime, sin_cos_table)
# apply dynamic
x_second = self.apply_dynamic(
x_prime,
gurobi_model,
thetaacc=thetaacc,
xacc=xacc,
env_input_size=self.env_input_size)
gurobi_model.update()
gurobi_model.optimize()
found_successor, x_second_results = self.h_repr_to_plot(
gurobi_model, template, x_second)
if found_successor:
post.append(tuple(x_second_results))
else:
model = pyo.ConcreteModel()
input = Experiment.generate_input_region_pyo(
model, template, x, self.env_input_size)
feasible_action = ORACartpoleExperiment.generate_nn_guard_pyo(
model, input, nn, action_ego=chosen_action, M=1e04)
if feasible_action: # performs action 2 automatically when battery is dead
max_theta, min_theta, max_theta_dot, min_theta_dot = self.get_theta_bounds_pyo(
model, input)
sin_cos_table = self.get_sin_cos_table(
max_theta,
min_theta,
max_theta_dot,
min_theta_dot,
action=chosen_action,
step_thetaacc=100)
x_prime_results = self.optimise_pyo(template, model, input)
x_prime = Experiment.generate_input_region_pyo(
model,
template,
x_prime_results,
self.env_input_size,
name="x_prime_input")
thetaacc, xacc = ORACartpoleExperiment.generate_angle_milp_pyo(
model, x_prime, sin_cos_table)
model.del_component(model.obj)
model.obj = pyo.Objective(expr=thetaacc,
sense=pyo.maximize)
result = Experiment.solve(model,
solver=Experiment.use_solver)
assert (result.solver.status == SolverStatus.ok) and (
result.solver.termination_condition
== TerminationCondition.optimal
), f"LP wasn't optimally solved {x}"
# apply dynamic
x_second = self.apply_dynamic_pyo(
x_prime,
model,
thetaacc=thetaacc,
xacc=xacc,
env_input_size=self.env_input_size,
action=chosen_action)
x_second_results = self.optimise_pyo(
template, model, x_second)
found_successor = x_prime_results is not None
if found_successor:
post.append((tuple(x_second_results)))
return post
