If I use RLlib, I can get the model summary through >>> trainer = PPOTrainer(env="CartPole-v0", config={"eager": True, "num_workers": 0}) >>> policy = trainer.get_policy() >>> policy.model.base_model.summary() Model: "model" _____________________________________________________________________ Laye…

The code you were using before will only work on a tensolfow model that is using keras and has stored that model as a “base_model” variable in the model class. If you are using torch you would need a different method to print the model. If you are using tf but not keras, which is what is happening i…

Hey @bug404 , unfortunately, there is no way of getting the Trainer object directly from the tune.run() results. But you could do the following after tune.run: results = tune.run("PPO", config=[some config], checkpoint_at_end=True) checkpoint = results.get_last_checkpoint() # Create a new Trainer.…

Actually, you don’t even have to restore, the model summary would be the same for a freshly init’d model.

Yeah, I think I can create a new PPOTrainer and pass the config parameter to it, and then call get_policy(). I asked the question was to see if there is a better way to do it.

Sorry, no, at this point, there isn’t. :confused:

@bug404 , You could also do something like this. I don’t know if it counts as “a better way” but it would work. I do not think that updating after_init on the Trainer will step on any other callbacks because I think those are defined in the policies rather than the trainer but @sven1977 would know b…

ray/custom_fast_model.py at master · ray-project/ray · GitHub , I modified this code as yours way, but it didn’t work. Can you have a look?

Wow, it’s very useful. Thank you very much.

# TODO: (sven): Get rid of `get_initial_state` once Trajectory Sven in this code refers to you? :open_mouth:

Yeah, I have seen it. Thank you very much.

How to get mode summary if I use tune.run()?

RLlib

sven1977 May 6, 2021, 8:49am 11

Thanks @mannyv for your detailed answer and the really cool hack into the after_init
This is great. @bug404 I answered your question on how to do this for torch in another thread here in the forum:

Topic		Replies	Views
How to get model summary using Pytorch backend? RLlib	6	1550	June 23, 2021
A little help for a novice RLlib	1	458	October 26, 2022
Use Policy_Trainer with TensorBoard RLlib	33	2487	November 13, 2021
How to make checkpoint by ray.tune.run and load it? RLlib	3	2814	July 7, 2022
How to print the TF model? RLlib	6	642	January 13, 2023

How to get mode summary if I use tune.run()?

Related topics