def _step(self, action):
if action == 1:
self.state += 0.05
elif action == 2:
self.state -= 0.05
self.state = np.mod(self.state, 2 * np.pi)
gt = self.gt_now
if self.in_period('fixation'):
reward = 0
else:
reward =\
np.max((self.rewards['correct']-tasktools.circular_dist(self.state-gt),
self.rewards['fail']))
norm_rew = (reward - self.rewards['fail']) / (
self.rewards['correct'] - self.rewards['fail'])
self.performance += norm_rew / self.dec_per_dur
return self.ob_now, reward, False, {'new_trial': False}
def _step(self, action):
ob = self.ob_now
ob[16:32] = np.cos(self.theta - self.state)
if action == 1:
self.state += 0.05
elif action == 2:
self.state -= 0.05
self.state = np.mod(self.state, 2 * np.pi)
gt = self.gt_now
reward = 0
if self.in_period('go1') or self.in_period('go2'):
reward =\
np.max((self.rewards['correct']-tasktools.circular_dist(self.state-gt),
self.rewards['fail']))
norm_rew = (reward - self.rewards['fail']) / (
self.rewards['correct'] - self.rewards['fail'])
self.performance += norm_rew / self.dec_per_dur
return ob, reward, False, {'new_trial': False}