def __init__(self,
actions,
name="qlearner",
alpha=0.1,
gamma=0.99,
epsilon=0.2,
explore="uniform",
anneal=False):
'''
Args:
actions (list): Contains strings denoting the actions.
name (str): Denotes the name of the agent.
alpha (float): Learning rate.
gamma (float): Discount factor.
epsilon (float): Exploration term.
explore (str): One of {softmax, uniform}. Denotes explore policy.
'''
Agent.__init__(self, name=name, actions=actions, gamma=gamma)
# Set/initialize parameters and other relevant classwide data
self.alpha, self.alpha_init = alpha, alpha
self.epsilon, self.epsilon_init = epsilon, epsilon
self.step_number = 0
self.anneal = anneal
self.default_q = 0.0
self.q_func = defaultdict(lambda: self.default_q)
# Choose explore type. Can also be "uniform" for \epsilon-greedy.
self.explore = explore
def __init__(self, policy, name="fixed-policy"):
'''
Args:
policy (func: S ---> A)
'''
Agent.__init__(self, name=name, actions=[])
self.policy = policy
self.name = name
def __init__(self, actions, gamma=0.95, horizon=4, s_a_threshold=10):
Agent.__init__(self,
name="rmax-h" + str(horizon),
actions=actions,
gamma=gamma)
self.rmax = 1.0
self.horizon = horizon
self.s_a_threshold = s_a_threshold
self.reset()
def __init__(self,
actions,
env_model,
explore_param=m.sqrt(2),
rollout_depth=100,
num_rollouts_per_step=50,
name="mcts",
gamma=0.99):
self.env_model = env_model
self.rollout_depth = rollout_depth
self.num_rollouts_per_step = num_rollouts_per_step
self.value_total = defaultdict(float)
self.explore_param = explore_param
self.visitation_counts = defaultdict(lambda: 1)
Agent.__init__(self, name=name, actions=actions, gamma=gamma)
def __init__(self,
actions,
name="Q-learning",
alpha=0.1,
gamma=0.9,
epsilon=0.05,
explore="uniform",
anneal=False,
custom_q_init=None,
default_q=0):
'''
Args:
actions (list): Contains strings denoting the actions.
name (str): Denotes the name of the agent.
alpha (float): Learning rate.
gamma (float): Discount factor.
epsilon (float): Exploration term.
explore (str): One of {softmax, uniform}. Denotes explore policy.
custom_q_init (defaultdict{state, defaultdict{action, float}}): a dictionary of dictionaries storing the initial q-values. Can be used for potential shaping (Wiewiora, 2003)
default_q (float): the default value to initialize every entry in the q-table with [by default, set to 0.0]
'''
name_ext = "-" + explore if explore != "uniform" else ""
Agent.__init__(self,
name=name + name_ext,
actions=actions,
gamma=gamma)
# Set/initialize parameters and other relevant classwide data
self.alpha, self.alpha_init = alpha, alpha
self.epsilon, self.epsilon_init = epsilon, epsilon
self.step_number = 0
self.anneal = anneal
self.default_q = default_q # 0 # 1 / (1 - self.gamma)
self.explore = explore
self.custom_q_init = custom_q_init
self._action_history = [] # store actions taken
# Q Function:
if self.custom_q_init:
self.q_func = self.custom_q_init
else:
self.q_func = defaultdict(
lambda: defaultdict(lambda: self.default_q))
def __init__(self,
actions,
name="qlearner",
alpha=0.05,
gamma=0.95,
epsilon=0.01,
explore="softmax"):
'''
Args:
actions (list): Contains strings denoting the actions.
name (str): Denotes the name of the agent.
alpha (float): Learning rate.
gamma (float): Discount factor.
epsilon (float): Exploration term.
explore (str): One of {softmax, uniform}. Denotes explore policy.
'''
Agent.__init__(self, name=name, actions=actions, gamma=gamma)
# Set/initialize parameters and other relevant classwide data
self.alpha = alpha
self.epsilon = epsilon
# Choose explore type. Can also be "uniform" for \epsilon-greedy.
self.explore = explore
def __init__(self, actions, name=""):
name = "random" if name is "" else name
Agent.__init__(self, name=name, actions=actions)
def __init__(self, actions):
Agent.__init__(self, name="random", actions=actions)