def choose_action(self):
"""Choose action by taking max over action cells.
"""
if len(self.striatum_activation) != len(self.env.actions):
print('something is wrong')
action_idx = utils.random_argmax(self.striatum_activation)
value = self.striatum_activation[action_idx]
return self.env.actions[action_idx], value * 120
def e_greedy_selection(self):
"""Epsilon greedy action selection: choose greedy action or random with probability epsilon.
:return (int): Index of chosen action
"""
if np.random.rand() < self.epsilon:
chosen_action = np.random.choice(range(self.actions.__len__()))
else:
chosen_action = utils.random_argmax(self.striatum_activation)
return chosen_action
def choose_action(self):
available_actions = self.get_available_actions()
action_goodness = []
for act in available_actions:
next_x, next_y = self.env.compute_new_position(act)
goal_cell_rate_diff = self.get_value(next_x, next_y)
action_goodness.append(goal_cell_rate_diff)
action_idx = utils.random_argmax(action_goodness)
allocentric_action = available_actions[action_idx]
# Back to egocentric reference frame
egocentric_action = get_relative_angle(allocentric_action,
self.env.curr_orientation)
return egocentric_action, action_goodness[action_idx]
def choose_action(self):
available_actions = self.get_available_actions()
action_goodness = []
for act in available_actions:
next_x, next_y = self.env.compute_new_position(act)
goal_cell_rate_diff = self.get_goal_cell_rate(
next_x, next_y) - self.goal_cell_rate
action_goodness.append(goal_cell_rate_diff)
#if np.all(np.array(action_goodness) < 0): # top reached but not goal
# self.weights *= 0
action_idx = utils.random_argmax(action_goodness)
allocentric_action = available_actions[action_idx]
# Back to egocentric reference frame
egocentric_action = get_relative_angle(allocentric_action,
self.env.curr_orientation)
return egocentric_action, action_goodness[action_idx]
def choose_action(self):
"""Choose action by taking max over action cells.
"""
action_idx = utils.random_argmax(self.striatum_activation)
value = self.striatum_activation[action_idx]
return self.env.actions[action_idx], value