def epsilon_greedy_linear_constant(self, state, eps_ = 0.1):
"""
epsilon greedy exploration with constant exploration epsilon
"""
eps = eps_
if random() < eps or state.terminal:
return Action.getRandomAction()
else:
actionHit_value = sum(self.get_feature_vector(state, Action.hit) * self.weights)
actionStick_value = sum(self.get_feature_vector(state, Action.stick) * self.weights)
action = Action.hit if actionHit_value > actionStick_value else Action.stick
return action
def epsilon_greedy(self, state):
"""
epsilon greedy exploration
"""
if state.terminal:
min_num_action = 0
else:
min_num_action = min(self.N[state.dealer_card - 1, state.player_sum - 1, :])
eps = self.N0 / (self.N0 + min_num_action)
# print (eps)
if random() < eps:
return Action.getRandomAction()
else:
action_value = np.argmax(self.Q[state.dealer_card - 1, state.player_sum - 1,:])
return Action.get_action(action_value)
