Skip to content
Snippets Groups Projects

Bug fix and improve MCTS and others

Merged Bug fix and improve MCTS and others
bug-fix-and-improve-MCTS-and-others into master
6 unresolved threads
Merged Jae Young Lee requested to merge bug-fix-and-improve-MCTS-and-others into master
6 unresolved threads
  • Fixed error with mean and std
  • Improved MCTS.
    1. Fixed a bug of starting with start_node;
    2. Fixed the miscalculation of mean and std;
    3. Added timeout to prevent infinite-loop.
    4. Changed output format in mcts.py
    5. updated and tested mcts.py
  • Refactoring (env -> worlds)
  • Change AP ('before_but_close_to_stop_region' --> 'close_to_stop_region')

Merge request reports

Loading
Loading

Activity

Filter activity
  • Approvals
  • Assignees & reviewers
  • Comments (from bots)
  • Comments (from users)
  • Commits & branches
  • Edits
  • Labels
  • Lock status
  • Mentions
  • Merge request status
  • Tracking
high_level_policy_main.py
129 129 print("Termination reason(s):")
130 130 for reason, count_list in termination_reason_list.items():
131 131 count_list = np.array(count_list)
132 while count_list.size != nb_trials:
133 count_list = np.append(count_list,0)
134
  • Jae Young Lee
    Jae Young Lee @jy63lee started a thread on the diff
    • mcts.py
    73 74 debug=debug)
    74 75
    75 76 # Evaluate
    76 print("\nConducting {} trials of {} episodes each".format(
    77 nb_trials, nb_episodes))
    77 print("\nConducting {} trials of {} episodes each".format(nb_trials, nb_episodes))
    78 timeout = 40 # 40 sec. timeout for each episode
  • Jae Young Lee
    Jae Young Lee @jy63lee started a thread on the diff
    • mcts.py
    96 step_current_node(visualize_low_level_steps=visualize)
    97 episode_reward += R
    98 # print('Intermediate Reward: %f (ego x = %f)' %
    99 # (R, options.env.vehs[0].x))
    100 # print('')
    101 if terminal:
    102 if 'episode_termination_reason' in info:
    103 termination_reason = info['episode_termination_reason']
    104 if termination_reason in trial_termination_reason_counter:
    105 trial_termination_reason_counter[termination_reason] += 1
    106 else:
    107 trial_termination_reason_counter[termination_reason] = 1
    108 if options.controller.can_transition():
    109 options.controller.do_transition()
    90 t = 0
    91 while True:
  • Jae Young Lee
    Jae Young Lee @jy63lee started a thread on the diff
    • mcts.py
    106 else:
    107 trial_termination_reason_counter[termination_reason] = 1
    108 if options.controller.can_transition():
    109 options.controller.do_transition()
    90 t = 0
    91 while True:
    92 options.controller.do_transition()
    93
    94 features, R, terminal, info = options.controller.step_current_node(visualize_low_level_steps=visualize)
    95 episode_reward += R
    96 t += DT
    97 # print('Intermediate Reward: %f (ego x = %f)' %
    98 # (R, options.env.vehs[0].x))
    99 # print('')
    100
    101 if terminal or t > timeout:
  • Jae Young Lee
    Jae Young Lee @jy63lee started a thread on the diff
    • mcts.py
    127 129 print("Trial {} Termination reason(s):".format(num_tr))
    128 for reason, count_list in trial_termination_reason_counter.items():
    129 count_list = np.array(count_list)
    130 print("{}: Avg: {}, Std: {}".format(reason, np.mean(count_list),
    131 np.std(count_list)))
    130 for reason, count in trial_termination_reason_counter.items():
    131 print("{}: {}".format(reason, count))
    132 132 print("\n")
    133 133
    134 134 overall_reward_list += reward_list
    135 overall_success_accuracy += [num_successes * 1.0 / nb_episodes]
    135 overall_success_percent_list += [num_successes * 100.0 / nb_episodes]
    136 136
    137 137 print("===========================")
    138 print('Overall: Reward = (Avg: {}, Std: {}), Success = (Avg: {}, Std: {})\n'.\
    138 print('Overall: Reward = (Avg: {:.2f}, Std: {:.2f}), Success = (Avg: {:.2f}, Std: {:.2f})\n'.\
  • Jae Young Lee
    Jae Young Lee @jy63lee started a thread on the diff
    • mcts.py
    138 print('Overall: Reward = (Avg: {}, Std: {}), Success = (Avg: {}, Std: {})\n'.\
    138 print('Overall: Reward = (Avg: {:.2f}, Std: {:.2f}), Success = (Avg: {:.2f}, Std: {:.2f})\n'.\
    139 139 format(np.mean(overall_reward_list), np.std(overall_reward_list),
    140 np.mean(overall_success_accuracy), np.std(overall_success_accuracy)))
    140 np.mean(overall_success_percent_list), np.std(overall_success_percent_list)))
    141 141
    142 142 print("Termination reason(s):")
    143 143 for reason, count_list in overall_termination_reason_list.items():
    144 144 count_list = np.array(count_list)
    145 print("{}: Avg: {}, Std: {}".format(reason, np.mean(count_list),
    146 np.std(count_list)))
    145 while count_list.size != nb_trials:
    146 count_list = np.append(count_list, 0)
    147
    148 print("{}: Avg: {:.2f}, Std: {:.2f}".format(reason, np.mean(count_list), np.std(count_list)))
    149
  • Aravind Balakrishnan
    Aravind Balakrishnan @a4balakr ·
    Maintainer

    Looks good.

  • Ashish Gaurav
    Ashish Gaurav @a5gaurav ·
    Contributor

    Looks good to me

  • Jae Young Lee merged

    merged

  • Jae Young Lee mentioned in commit 6b50d1d1

    mentioned in commit 6b50d1d1

    • Please register or sign in to reply