Nonlinear TDC, GTD, GTD2

[Aisha Shokry]

Hi all,

Great work, many thanks Thomas!

I was just wondering if there is any implementation for the new gradient TD algorithms using non-linear representations, MLPs to be specific?

I have seen the results on RLGO on the main papers, but as I believe they used logistic linear model, something of that sort. So is there also any reported results also?

In my side, I have an implementation for it and the multiplication by hessian is inspired by the details in Bishop's book, pattern recognition. Though I would like to get code/ results to consult and verify.

Regards,

Aisha

[Thomas Degris]

Hello Aisha,

Thank you for your interest in RLPark. 

There is no implementation of TD with multi-layered perceptrons in RLPark. All the implemented TD algorithms use a linear combination of (non-linear) features. 

About RLGO results, you may be able to find an implementation at http://rlgo.sourceforge.net/

Furthermore, I would suggest you to contact Hamid Maei: perhaps Hamid has some Matlab implementation of non linear TD. 

I hope this will help. 

Thomas

--
 
---
You received this message because you are subscribed to the Google Groups "RLPark" group.
To unsubscribe from this group and stop receiving emails from it, send an email to githubrlpark...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

[Aisha Shokry]

Hi Thomas,

From my quick review I couldn't locate the algorithms, but will dig even more. From the papers, and as I believe, I could be wrong though, Maei has solved TD divergence in the Spiral problem, which constitutes only one variable, i.e. not quite as complicated as an MLP can get. If you have any further information please let me know, otherwise I will pop the question to Dr. Maei and see what can I get.

Many thanks for your help.

Aisha