def test2():
print()
print('blackjack')
print()
ex = Blackjack()
run_environment(
ex,
lambda: TransitionBOW(
[BlackjackFeatures()], #ex.width, ex.height)],
[AttendAt(lambda _: 0, 'blackjack')],
ex.n_actions),
BlackjackLoss(),
)
def test0():
print()
print('micro pocman')
print()
ex = MicroPOCMAN()
run_environment(
ex,
lambda: TransitionBOW(
[LocalPOCFeatures(history_length=4)], #ex.width, ex.height)],
[AttendAt(lambda _: 0, 'poc')],
4),
POCLoss(),
)
def test():
print('')
print('Cart Pole')
print('')
ex = CartPoleEnv()
run_environment(
ex,
lambda: TransitionBOW([CartPoleFeatures(
)], [AttendAt(lambda _: 0, 'cartpole')], ex.n_actions),
CartPoleLoss(),
rl_alg=reinforce,
n_epochs=100,
lr=0.1,
)
def test3():
print()
print('hex')
print()
board_size = 3
ex = Hex(Hex.BLACK, board_size)
run_environment(
ex,
lambda: TransitionBOW(
[HexFeatures(board_size)], #ex.width, ex.height)],
[AttendAt(lambda _: 0, 'hex')],
ex.n_actions),
HexLoss(),
)
def test():
print('')
print('Proximal Policy Optimization')
print('')
args = parse_arguments()
if args.task == 'mountaincar':
print('Mountain Car')
ex = MountainCar()
run_ppo(
ex,
lambda dy_model:
TransitionBOW(
[MountainCarFeatures()],
[AttendAt(lambda _: 0, 'mountain_car')],
ex.n_actions),
MountainCarLoss(),
args.eps,
args.learner,
)
elif args.task == 'cartpole':
print('Cart Pole')
ex = CartPoleEnv()
run_ppo(
ex,
lambda dy_model:
TransitionBOW(
[CartPoleFeatures()],
[AttendAt(lambda _: 0, 'cartpole')],
ex.n_actions),
CartPoleLoss(),
args.eps,
args.learner,
)
else:
print('Unsupported Task!')
exit(-1)
def test4():
print '\n===\n=== test4: big grid world, global features\n==='
ex = make_big_gridworld()
run_gridworld(
ex, lambda: TransitionBOW([GlobalGridFeatures(ex.width, ex.height)],
[AttendAt(lambda _: 0, 'grid')], 4))
def test3():
print '\n===\n=== test3: p_step_success=0.8, but local features only\n==='
ex = make_default_gridworld(p_step_success=0.8, start_random=True)
run_gridworld(
ex, lambda: TransitionBOW([LocalGridFeatures(ex.width, ex.height)],
[AttendAt(lambda _: 0, 'grid')], 4))
def test2():
print '\n===\n=== test2: p_step_success=0.8 and per_step_cost=0.1\n==='
ex = make_default_gridworld(per_step_cost=0.1, p_step_success=0.8)
run_gridworld(
ex, lambda: TransitionBOW([GlobalGridFeatures(ex.width, ex.height)],
[AttendAt(lambda _: 0, 'grid')], 4))
def test0():
print '\n===\n=== test0: p_step_success=1.0\n==='
ex = make_default_gridworld(p_step_success=1.0)
run_gridworld(
ex, lambda: TransitionBOW([GlobalGridFeatures(ex.width, ex.height)],
[AttendAt(lambda _: 0, 'grid')], 4))