def test_estimate():
from environment import GridWorldEnv
env = GridWorldEnv(grid=[
[0, 0, 0, 1],
[0, 0, 0, 0],
[0, -1, 0, 0],
[0, 0, 0, 0],
])
# Train Teacher
teacher = PolicyIterationPlanner(env)
teacher.plan()
trajectories = []
print("Gather demonstrations of teacher.")
for i in range(20):
s = env.reset()
done = False
steps = [s]
while not done:
a = teacher.act(s)
n_s, r, done, _ = env.step(a)
steps.append(n_s)
s = n_s
trajectories.append(steps)
print("Estimate reward.")
irl = MaxEntIRL(env)
rewards = irl.estimate(trajectories, epoch=100)
print(rewards)
env.plot_on_grid(rewards)
def test_plan():
from environment import GridWorldEnv
env = GridWorldEnv(grid=[
[0, 0, 0, 1],
[0, 0, 0, 0],
[0, -1, 0, 0],
[0, 0, 0, 0],
])
print("Value Iteration.")
vp = ValuteIterationPlanner(env)
v = vp.plan()
print(v.reshape(env.shape))
print("Policy Iteration.")
pp = PolicyIterationPlanner(env)
v = pp.plan()
print(v.reshape(env.shape))
q = pp.policy_to_q(v, 0.9)
print(np.sum(q, axis=1).reshape(env.shape))