Course: Learning Agents (5.4)

General Tutorial & Course Discussions
21

This usually means that the trainer didn’t get created successfully because of some other error. Look through the log and see if there is another error prior to this one.

22

Hi! This is a Great Job! But i have a few question. I followed your step about Tensorboard, but i can’t found the folder “TensorBoard\runs” in the path "Intermediate\LearningAgents". I wander why? I made sure I installed the tensorboard as the tutorial and set Use Tensorboard to true in the mananger.

23

@Deathcalibur Any goodies you can tell us about for 5.5?

1 Like
24

@moon-white

  1. Check the log to make sure there are no warnings or errors related to tensorboard. You should see “tensorboard: true” in the config print out
  2. Make sure you are checking under your project directory’s intermediate path and not somewhere in the engine.

@tomhalpin8
Here’s what is coming in 5.5:

  • Re-architected training code
    • Split choice of training algorithm from networking protocol and training environment
  • Additional networking protocols
    • TCP sockets in addition to shared memory
  • Bring your own algorithm
    • Implement your own class and Python code to replace PPO trainer
  • Mac and Linux training
    • Train your agents on non-Windows machines

Currently working on a solution for RL scale out (meaning multi-process data collection) for 5.6

Let me know if you have questions!

2 Likes
25

Great! Is this already in 5.5.0-Preview?

26

Is ActActor in PerformAgentAction supposed to be a variable or local variable? It’s not mentioned in the tutorial text, but it’s referenced by the PerformAgentAction script.

27

Was the LearningAgentsTrainer class replaced by LearningAgentsPPOTrainer in the last version? Though it does not contain the function GatherAgentReward.

28

Great! Is this already in 5.5.0-Preview?

Yes

Is ActActor in PerformAgentAction supposed to be a variable or local variable? It’s not mentioned in the tutorial text, but it’s referenced by the PerformAgentAction script.

When in doubt, you can usually make it a local variable and it will be fine.

Was the LearningAgentsTrainer class replaced by LearningAgentsPPOTrainer in the last version? Though it does not contain the function GatherAgentReward .

The trainer was split in “TrainingEnvironment” and “PPOTrainer”. This was done so that you could introduce your own training algorithm (or we could introduce others eventually) and you would not need to rewrite your reward function since that now exists in the “TrainingEnvironment” class.

29

(post deleted by author)