fa2cd0c7f0d382d2bb3726e9cec392962b9e35a0,mushroom/algorithms/batch_td.py,DeepFQI,draw_action,#DeepFQI#Any#,139

Before Change

        else:
            extended_state = self._buffer.get()

            action = super(DeepFQI, self).draw_action(extended_state)

        self._episode_steps += 1

After Change

        else:
            extended_state = self._buffer.get()

            if not np.random.uniform() < self.policy._epsilon(extended_state):
                q = np.ones(self.mdp_info["action_space"].n)
                for i in xrange(q.size):
                    features = self._extractor.models[i].predict(
                        np.expand_dims(extended_state, axis=0))
                    q[i] = self.approximator.predict(features)
                return np.array(
                    [np.random.choice(np.argwhere(q == np.max(q)).ravel())])
            return self.mdp_info["action_space"].sample()

        self._episode_steps += 1

        return action

Instances