Unable to add new config parameters

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04):
• Ray installed from (source or binary): pip install ray
• Ray version:
• Python version: 3.6.5
• Exact command to reproduce: python train_ppo.py

Describe the problem

I am trying to train sonic-on-ray here When I execute python train_ppo.py, ray.rllib.utils.init gives me:

raise Exception("Unknown config parameter `{}` ".format(k))

for everything NEW inside config.update( { } ). I tried config[newKey] = value, nothing seems to allow me to add new keys.

1. How do I set new_keys_allowed to True? In agent.py _allow_unknown_configs is set to False. Am I suppose to modify this to True? And if I didn’t install ray from source, how would I change that?
2. And what is the up-to-date syntax on updating config using config.update(dict)? I tried looking at the documentation I’m wondering is "timesteps_per_batch" or sgd_stepsize changed under a different name? Since these keys are no longer under the DEFAULT_CONFIG inside either agent.py or ppo.py If so, what are the new names for these keys? or can someone from ray update the sonic-on-ray repo?

As always, thanks a lot

Source code / logs

Created LogSyncer for /home/peng00/ray_results/PPO_sonic_env_2018-10-29_22-07-56i9zwqmz5 -> None
Traceback (most recent call last):
  File "train_ppo.py", line 65, in <module>
    alg = ppo.PPOAgent(config=config, env=env_name)
  File "/home/peng00/miniconda3/envs/sonic/lib/python3.6/site-packages/ray/rllib/agents/agent.py", line 216, in __init__
    Trainable.__init__(self, config, logger_creator)
  File "/home/peng00/miniconda3/envs/sonic/lib/python3.6/site-packages/ray/tune/trainable.py", line 84, in __init__
    self._setup()
  File "/home/peng00/miniconda3/envs/sonic/lib/python3.6/site-packages/ray/rllib/agents/agent.py", line 253, in _setup
    self._allow_unknown_subkeys)
  File "/home/peng00/miniconda3/envs/sonic/lib/python3.6/site-packages/ray/rllib/utils/__init__.py", line 35, in deep_update
    raise Exception("Unknown config parameter `{}` ".format(k))
Exception: Unknown config parameter `timesteps_per_batch` 

[ericl]

That would be sample_batch_size.

On Tue, Oct 30, 2018 at 1:09 PM jhpenger [e-mail] wrote:

Yep, that looks right to me! … <#m_2581864536168091198_>

On Tue, Oct 30, 2018, 12:47 PM jhpenger @.***> wrote: Thanks for the info. Just to make sure, is this the correct translation to the current ray version? - timesteps_per_batch = train_batch_size - sgd_stepsize = lr — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#3160 (comment) https://github.com/ray-project/ray/issues/3160#issuecomment-434443299>, or mute the thread https://github.com/notifications/unsubscribe-auth/AAA6SgBENZ6i_mKW1-P5ZxPDBmL6C6Ehks5uqKy5gaJpZM4YAvtm .

@ericl https://github.com/ericl , sorry i updated the comment, can’t figure out what min_steps_per_task is

— You are receiving this because you were mentioned.

Reply to this email directly, view it on GitHub https://github.com/ray-project/ray/issues/3160#issuecomment-434450856, or mute the thread https://github.com/notifications/unsubscribe-auth/AAA6Sju3xNqIQWQiDLpip__Rt0kiHEzDks5uqLHhgaJpZM4YAvtm .

[jhpenger]

Yep, that looks right to me! … On Tue, Oct 30, 2018, 12:47 PM jhpenger @.***> wrote: Thanks for the info. Just to make sure, is this the correct translation to the current ray version? - timesteps_per_batch = train_batch_size - sgd_stepsize = lr — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#3160 (comment)>, or mute the thread https://github.com/notifications/unsubscribe-auth/AAA6SgBENZ6i_mKW1-P5ZxPDBmL6C6Ehks5uqKy5gaJpZM4YAvtm .

@ericl , sorry i updated the comment, can’t figure out what min_steps_per_task is