Training with NLG + NLU error

[randyhouse]

Hello

I am trying to train your DDQ model with NLG + NLU on.
I ran the following command:

python run.py --agt 9 --usr 1 --max_turn 40 
	      --movie_kb_path ./deep_dialog/data/movie_kb.1k.p 
	      --dqn_hidden_size 80 --experience_replay_pool_size 5000 
	      --episodes 500 
	      --simulation_epoch_size 100 
	      --run_mode 3 
	      --act_level 1
	      --slot_err_prob 0.0 
	      --intent_err_prob 0.00 
	      --batch_size 16 
	      --goal_file_path ./deep_dialog/data/user_goals_first_turn_template.part.movie.v1.p 
	      --warm_start 1 --warm_start_epochs 100 
	      --planning_steps 9 
	      --write_model_dir ./deep_dialog/checkpoints/DDQAgent
	      --torch_seed 100
	      --grounded 0
	      --boosted 1
	      --train_world_model 1

But I received this error during the warm_start phase:

Traceback (most recent call last):
  File "/Applications/PyCharm.app/Contents/helpers/pydev/pydevd.py", line 1668, in <module>
    main()
  File "/Applications/PyCharm.app/Contents/helpers/pydev/pydevd.py", line 1662, in main
    globals = debugger.run(setup['file'], None, None, is_module)
  File "/Applications/PyCharm.app/Contents/helpers/pydev/pydevd.py", line 1072, in run
    pydev_imports.execfile(file, globals, locals)  # execute the script
  File "run.py", line 535, in <module>
    run_episodes(num_episodes, status)
  File "run.py", line 458, in run_episodes
    warm_start_simulation()
  File "run.py", line 408, in warm_start_simulation
    episode_over, reward = dialog_manager.next_turn()
  File "deep_dialog/dialog_system/dialog_manager.py", line 117, in next_turn
    self.user_action)
  File "deep_dialog/usersims/usersim_model.py", line 304, in register_experience_replay_tuple
    action_idx = self.action_index(copy.deepcopy(user_a_t))
  File "deep_dialog/usersims/usersim_model.py", line 292, in action_index
    raise Exception("action index not found")
Exception: action index not found

It looks like when the user says the following, it is not inside the self.feasible_actions_users, do you know what is going on?

after: {'request_slots': {}, 'nl': 'I want to watch avengers.', 'turn': 2, 'diaact': 'request', 'inform_slots': {'moviename': 'avengers'}}
Turn 2 usr: request, inform_slots: {'moviename': 'avengers'}, request_slots: {}
Turn 2 usr: I want to watch avengers.

[Xiujun Li]

Thanks for this feedback.

The intent of NLU is wrong here, it should be an inform action. Just fixed it. You can check out the latest code.

Thanks

[randyhouse]

Hello @xjli, thanks for your reply.

I just git pull the latest code, and re-ran the command above, I received this the same error, but now, the error occurs at:

{'request_slots': {'ticket': 'UNK'}, 'inform_slots': {'numberofpeople': 'PLACEHOLDER'}, 'diaact': 'request'}

[Xiujun Li]

Thanks, let me check NLU part. The NLU is quite brittle.

You can run the code without NLU (--act_level 0) first.
I will let you know when we fix it.

Thanks

[Xiujun Li]

@randyhouse , I fixed the bugs in act_level = 1 mode.

You can check out the latest code.

Thanks so much,