212681dee8efa1ddfdcf6f31dbb590f739d107d8,agents/a3c.py,A3CThread,run,#A3CThread#,370

Before Change

                self.r: np.asarray(batch_r)
            })
            n_states = states.shape[0]
            feed_dict = {
                // self.master.reward: sum(trajectory.rewards),
                // self.master.episode_length: trajectory.steps
            }
            losses = zip(self.master.losses, map(lambda x: x / n_states, results))
            feed_dict.update(losses)
            summary = sess.run([self.master.summary_op], feed_dict)
            self.writer.add_summary(summary[0], results[-1])
            self.writer.flush()
            t += 1

After Change

            feature = trajectory.features[0]
            if isinstance(self, A3CDiscreteCNNRNN):
                feed_dict[self.ac_net.rnn_state_in] = feature
            results = sess.run(fetches, feed_dict)
            n_states = states.shape[0]
            feed_dict = dict(zip(self.master.losses, map(lambda x: x / n_states, results)))
            summary = sess.run([self.master.summary_op], feed_dict)
            self.writer.add_summary(summary[0], results[-1])

Instances