def test_learn_play(d = 6, num_layers = 2, num_units = 100,
eps = 0.5, iters = 10000, draw=False,
tabular = True, batch=False, batch_epochs=10,
num_episodes = 10, episode_length = 100):
iters_per_value = 1 if iters <= 10 else int(iters / 10.0)
scores = []
def interact(q, iter=0):
if iter % iters_per_value == 0:
scores.append((iter, evaluate(game, num_episodes, episode_length,
lambda s: greedy(q, s))[0]))
print('score', scores[-1], flush=True)
game = No_Exit(d)
if tabular:
q = TabularQ(game.states, game.actions)
else:
q = NNQ(game.states, game.actions, game.state2vec, num_layers, num_units,
epochs=batch_epochs if batch else 1)
if batch:
qf = Q_learn_batch(game, q, iters=iters, episode_length = 100, n_episodes=10,
interactive_fn=interact)
else:
qf = Q_learn(game, q, iters=iters, interactive_fn=interact)
if scores:
print('String to upload (incude quotes): "%s"'%toHex(pickle.dumps([tabular, batch, scores], 0).decode()))
# Plot learning curve
plot_points(np.array([s[0] for s in scores]),
np.array([s[1] for s in scores]))
for i in range(num_episodes):
reward, _, animation = sim_episode(game, (episode_length if d > 5 else episode_length/2),
lambda s: greedy(qf, s), draw=draw)
print('Reward', reward)
return animation
def test_solve_play(d=6, draw=False,
num_episodes=10, episode_length=100):
game = TempSim()
qf = value_iteration(game, TabularQ(game.states, game.actions))
for i in range(num_episodes):
reward, _ = sim_episode(game, (episode_length if d > 5 else episode_length / 2),
lambda s: greedy(qf, s), draw=draw)
print('Reward', reward)
def test_solve_play(d = 5, draw=False,
num_episodes = 10, episode_length = 100):
game = No_Exit(d)
qf = value_iteration(game , TabularQ(game.states, game.actions))
for i in range(num_episodes):
reward, _, animation = sim_episode(game, (episode_length if d > 3 else episode_length/2),
lambda s: greedy(qf, s), draw=draw)
print('Reward', reward)
return animation
def emulate(game, q, episode_length, num_episodes=1):
for i in range(num_episodes):
reward, _ = sim_episode(game, episode_length,
lambda s: greedy(q, s), interactive_fn=print)
print('Reward', reward)
