Closed Ruzihm closed 11 years ago
Hi ! I think I haven't got relative entropy to work yet. I have stopped working on this branch at the end of January, maintaining the code in its litteral form, embedded in .org files is in fact a bad idea. It is very cumbersome. I currently work on the "Soulless" branch, which use ipython notebooks.
I hope I'll get around to clean the code in master sometimes, but I believe it will not be before a few months.
Are you working on Inverse Reinforcement Learning ?
I'll be happy to share my new, functional code with you if you need it.
Cheers.
Yes, I am curious about using inverse reinforcement in situations where it is difficult to create a model. That is what makes the relative entropy approach appeal to me.
Rick Van Tassel
On Mon, Mar 4, 2013 at 3:45 PM, edouardklein notifications@github.comwrote:
Hi ! I think I haven't got relative entropy to work yet. I have stopped working on this branch at the end of January, maintaining the code in its litteral form, embedded in .org files is in fact a bad idea. It is very cumbersome. I currently work on the "Soulless" branch, which use ipython notebooks.
I hope I'll get around to clean the code in master sometimes, but I believe it will not be before a few months.
Are you working on Inverse Reinforcement Learning ?
I'll be happy to share my new, functional code with you if you need it.
Cheers.
— Reply to this email directly or view it on GitHubhttps://github.com/edouardklein/RL-and-IRL/issues/1#issuecomment-14404338 .
Hello Again,
I just pushed, on the branch Soulless, code that runs the Relative Entropy algorithm on the Highway driving problem : https://github.com/edouardklein/RL-and-IRL/blob/Soulless/Exp10.py
I hope this helps. If not, maybe contact one of the author of Relative Entropy : http://damas.ift.ulaval.ca/~boularia/ .
If you are interested in model-free IRL, may I suggest taking a look at the SCIRL algorithm that I co-autored : http://rdklein.fr/research/ second publication. I'd be happy to help you with applying it. The code is here : https://github.com/edouardklein/RL-and-IRL/blob/Soulless/Exp7.py (on the mountain car) https://github.com/edouardklein/RL-and-IRL/blob/Soulless/Exp12.py (on the Highway driving problem) This algorithms only needs expert data (up to the use of a simple heuristics, see the paper and the source code), to the best of my knowledge, it is the only published algorithm with such a feature.
Hi again,
The SCIRL link on your research page gives me a 404 error: http://rdklein.fr/papers/klein2012structured.pdf
My bad, all the links missed the "research" part. This is now corrected. The paper is at http://rdklein.fr/research/papers/klein2012structured.pdf .
In GridWorld/REntropy_Exp1.org there is a reference to DU.py that I can not find.