def generate_episode_from_Q(env: BlackjackEnv, Q, epsilon,
action_count) -> [tuple]:
"""
Generates an episode
@param env:
@param Q:
@param epsilon:
@param action_count
Returns
"""
episode = []
# stores the initial state. [sum of player cards, open dealer card, has usable Ace]
state = env.reset()
while True:
if state in Q:
# choose the action with the Q table in mind
action = np.random.choice(np.arange(action_count),
p=get_probs(Q[state], epsilon,
action_count))
else:
# if we have never visited this state before, just throw the dice
action = env.action_space.sample()
next_state, reward, done, info = env.step(action)
episode.append((state, action, reward))
state = next_state
if done:
break
return episode
def gen_episode_data(policy: DeterministicPolicy,
env: BlackjackEnv) -> List[Tuple[State, Action, Reward]]:
episode_history = []
state = env.reset()
done = False
while not done:
action = policy(state)
next_state, reward, done, _ = env.step(action)
episode_history.append((state, action, reward))
state = next_state
return episode_history
def gen_custom_s0_stochastic_episode(policy: Policy, env: BlackjackEnv, initial_state: State) \
-> List[Tuple[State, Action, Reward]]:
episode_history = []
state = initial_state
done = False
while not done:
A: ActionValue = policy[state]
action = np.random.choice([0, 1], p=A / sum(A))
next_state, reward, done, _ = env.step(action)
episode_history.append((state, action, reward))
state = next_state
return episode_history