Training the model about imitation-learning HOT 4 OPEN

Hey James (@jcyriac2 ), Yes "run_CIL.py" load a trained model and test the model with a benchmark.

For your second question, the answer is yes. However, I'm not sure if I understood what do you want by starting the agent? You can find the code in "run_CIL.py" how to load a pre-trained model. If you're going to benchmark your own model, you only need to replace those checkpoint files.

They didn't release the code yet. You can start with my code for training. However, there is some little bug in that code which you can find in TODO. Please feel free to pull a request in my code repository.

from imitation-learning.

HI @mvpcom, thank you a lot for your code! Could you share, please, your results, did you manage to reproduce it and train your agent successfully?

from imitation-learning.

@merryHunter Hey Ivan, sorry for late answer. I was busy with my recent idea. Honestly, as you can see in the TODO parts, it needs some little changes to reproduce the same results. Hopefully, I managed to do that, but I need time to fix that repo and based on that code style. If you are in a hurry, you have to re-implement the loss function part and make it one loss in total for all branches and use a mask for performing specific operations for each branch. And in case you want to make the training process faster, you have to use a queue and new input methods in Tensorflow to reduce the pipeline bottleneck problems.

from imitation-learning.

hi, @mvpcom, thanks a lot for your code!

from imitation-learning.