sess.run(dqn.trainStep, feed_dict={dqn.yInput: y_batch, dqn.actionInput: action_batch,dqn.currentQNet.stateInput: state_batch})
state = next_state
# if step % 100 == 0:
# m, opt = sess.run([dqn.merged, dqn.trainStep],
# {dqn.yInput: y_batch, dqn.actionInput: action_batch,
# dqn.currentQNet.stateInput: state_batch})
# summary_writer.add_summary(m, step)
# if step % UPDATE_TIME == 0:
# sess.run(dqn.copyCurrentToTargetOperation())
if game_over:
frame_stack.empty()
game +=1
game_scores.append(score)
if game % BACKUP_RATE == 0 and SAVE_NETWORK:
saver.save(sess, 'saved_networks/' + ENVIRONMENT + '-dqn', global_step=game)
print('Network backup done')
if (game % 20) == 0:
print("The average score of the last 20 games is:", np.mean(game_scores[-20:]),
" currently at game ", game, " , step ", step)
summary_scores = sess.run(avg_Score_l20, {avg_Score_l20_plhldr: np.mean(game_scores[-20:])})
summary_writer.add_summary(summary_scores, step)
print("The average score of all games is:", np.mean(game_scores))
# else:
# print('Game %s finished with score %s' % (game, score))
