10:01 < luffy1996> zoq: I would like to know what is aggregated_policy class in reinforcement learning

10:01 < luffy1996> It would be great if you can give a paper that describes the method

10:05 < luffy1996> https://github.com/mlpack/mlpack/blob/master/src/mlpack/methods/ann/layer/reinforce_normal_impl.hpp

10:05 < luffy1996> Is this the reinforce algorithm defined in reinforcement learning?

12:36 < zoq> luffy1996: I'm not sure there is a reference for the Aggregated policy class: "An aggregated policy will randomly select a child policy under a given distribution at each time step."

12:36 < zoq> luffy1996: The reinforce normal layer is mend to be used for the recurrent visual attention model.

12:36 < luffy1996> zoq: this reinforce is different from the policy gradient method. right?

12:37 < zoq> luffy1996: yes

12:40 < luffy1996> zoq: I am having confusion here. What do you define a child policy ?

12:43 < zoq> If you use the AggregatedPolicy class, you have to pass a collection of policies, each policy in that collection is a child.

12:47 < luffy1996> I think I will have to take a look at the implementation again.

12:48 < luffy1996> zoq: I am planning to go ahead with policy gradient methods implementation in mlpack as gsoc propsoal. How do you feel about this?

12:50 < zoq> luffy1996: Sounds fine for me, do you have specifc methods in mind?

12:51 < viraj> Hi I am new to the mlpack. I have gone through the methods that are included in mlpack. I want to know if you are interested to implement the SVM (Support Vector Machine) Algorithm because I didn't saw anything like that in the current methods.

12:51 < luffy1996> I was planning to go with Reinforce (with,without baseline) and actor critic method implementation

12:51 < luffy1996> What are your say in this?

12:53 < zoq> luffy1996: Sounds good :)

12:53 < zoq> viraj: Hello, if you are going to write an SVM implementation, keep in mind that it should outperform libsvm, which might be pretty difficult, but if you're up for a challenge, then that might be a good one :)

12:55 < luffy1996> zoq: I have an idea how important is a gsoc proposal because one has to complete the project in summers. How do you see my plans. Is it apt or should I add one or two algorithms more :)

12:57 < luffy1996> zoq: Moreover I would like to know your views on its implementation. How do you see that policy gradient might help the public . I would like to here your views :)

13:06 < zoq> Actually, that depends on the details, e.g. Shangtong already worked on A3C, so it is interesting to see in what way you like to extend the existing codebase. Each idea has its advantages, e.g. it might not return the best results, but it is able to achieve good results in fewer iterations.

13:10 < viraj> zoq: I don't know about libsvm before so I searched about libsvm and found that it is very developed library. It would be very difficult to outperform their implementations.

13:14 < zoq> viraj: yeah, definitely challenging.

13:15 < viraj> I would like to know that is it a good idea to implement Genetic Algorithms in mlpack for gsoc.

13:36 < luffy1996> zoq: I have been through the codes so I know that A3C has been implementated. What I wanted was to go for PPO in actor critic .

13:37 < luffy1996> How do you see this ? :)

14:35 < zoq> luffy1996: Ah I see, that would be a really neat addition.

14:36 < zoq> viraj: Absolutely, there is an open PR for NEAT, that needs some additional work.

14:37 < luffy1996> So I am going with this as my proposal. I hope this will meet will all the requirements that mlpack would need :)

16:07 < moksh> Hey, is it possible to run only some specific test cases? For example, I wrote code for a new variant of the adam optimizer and some test cases in the adam_test.cpp file. Is there a way to run only that test file?

16:08 < zoq> yes, bin/mlpack_test -t SuiteName/TestCase e.g. bin/mlpack_test -t AdamTest/AdamSphereFunctionTest

16:19 < moksh_> @zoq if I make any changes then do I have to build the entire library or it is possible to build just that adam module and its test case?

16:32 < zoq> moksh_: if you run make again it should also rebuild changes.

16:35 < moksh_> @zoq Okay. Thank you! :)