def convergence_test():
model = md.DeterministicModel(10,
10,
1,
sensor=Sensor.DeterministicSensor(3),
episode_length=20)
iterations = 20000
policy, c1, p1 = q_learning(model,
-2,
2,
iterations,
epsilon=0.2,
alpha=.9,
return_convergence=True)
policy, c2, p2 = q_learning(model,
-3,
3,
iterations,
epsilon=0.2,
alpha=.9,
return_convergence=True)
policy, c3, p3 = q_learning(model,
-6,
6,
iterations,
epsilon=0.2,
alpha=.9,
return_convergence=True)
model = md.DeterministicModel(25,
10,
1,
sensor=Sensor.DeterministicSensor(3),
episode_length=50)
iterations = 20000
policy, c4, p4 = q_learning(model,
-2,
2,
iterations,
epsilon=0.2,
alpha=.9,
return_convergence=True)
policy, c5, p5 = q_learning(model,
-3,
3,
iterations,
epsilon=0.2,
alpha=.9,
return_convergence=True)
policy, c6, p6 = q_learning(model,
-6,
6,
iterations,
epsilon=0.2,
alpha=.9,
return_convergence=True)
model.reset()
model.print_policy(policy)
model.update(0, True)
model.n_update(model, n=5)
model.print_policy(policy)
print(model)
print(policy.pi)
utils.plot_convergence([c1, c2, c3, c4, c5, c6], [
"action space = [-2, 2], state space 10x10",
"action space = [-3, 3], state space 10x10",
"action space = [-6, 6], state space 10x10",
"action space = [-2, 2], state space 25x10",
"action space = [-3, 3], state space 25x10",
"action space = [-6, 6], state space 25x10"
])
utils.plot_convergence([p1, p2, p3, p4, p5, p6], [
"action space = [-2, 2], state space 10x10",
"action space = [-3, 3], state space 10x10",
"action space = [-6, 6], state space 10x10",
"action space = [-2, 2], state space 25x10",
"action space = [-3, 3], state space 25x10",
"action space = [-6, 6], state space 25x10"
])
