search

DeepwaterDiscord / morl

Multi-Objective Reinforcement Learning
MIT License
5 stars 1 forks source link

Filter Methods #1

Open MatKallada opened 6 years ago

MatKallada commented 6 years ago

Hey guys!,

Firstly, I'm impressed by the level of commitment that you guys are putting into this project -- well done! I was working on a similar project, of incorporating filter methods into MORL, and wanted to see if anyone had already accomplished this and ran into this repository. I just had a remark.

You mention that you're using "filter methods" for multi-objective optimization. However, I think you might have mixed up the terminology in this project.

Filter methods are terms that appear both in a feature selection context and a multi-objective optimization context, but mean different things. In multi-objective optimization, filter methods mean that we have a "filter" that iteratively screens potential candidate solutions and actively maintains the Pareto front of those iterates. In a feature selection context from supervised learning, filter methods means eliminating some of the features in feature space.

Your project seems to be using filter methods from feature selection -- but I think you meant to use the one from multi-objective optimization as described above.

Let me know if this helps at all or if you have further questions!,

Best, Mat

frazierbaker commented 6 years ago

Hi Mat, Thanks for the comment!

A couple of weeks ago we actually delved further into the various meanings of filter methods, as we realized that, exactly as you have stated, there are many meanings to the term "filter methods" and we mixed up the terminology as it applies to this project. We wrote up a review of the terminology and an explanation of the direction we wanted to take so that we could present it to our advisor. We'll probably be uploading that document as we move forward with the project once we have his feedback. I think I can speak for both @jgreer013 and I when I say we really appreciate your comment and words of support.

Thanks, Frazier

MatKallada commented 6 years ago

No Problem! :) I was worried that you guys might fall into a rabbit hole; I'm glad we're not in that territory! Chapter 15.4 from Nocedal & Wright is the suggested reading on OpenAI's Requests for Research, which I've found particularly helpful for understanding filter methods; it might be helpful to have a look if you haven't already.

Feel free to drop me a line if you have further questions or if there's any other way I can help with the project.

Best, Mat