search

Svalorzen / AI-Toolbox

A C++ framework for MDPs and POMDPs with Python bindings
GNU General Public License v3.0
647 stars 98 forks source link

Question about function approximation. #38

Closed ghost closed 5 years ago

ghost commented 5 years ago

Does the library provide any facilities for generalisation and value function approximation ? (e.g. simple Q-Learning with linear function approximation)

Svalorzen commented 5 years ago

The Factored part of the library does use linearly factored QFunctions. At the moment there's not that much stuff yet (Sparse Cooperative Q-Learning and the multi-agent Linear Programming are the main ones), but more stuff is to come. I already have some code for something similar to what you suggest, but I'm waiting to put it online until I can get it published :)

Svalorzen commented 5 years ago

If you have suggestions for methods feel free to add them to issue #7!

ghost commented 5 years ago

Awesome ! Really great. Are you also considering deep RL ?

Svalorzen commented 5 years ago

Not at the moment. I think deep RL is a bit outside the scope of this library for now, and as it stands there's quite a lot of other repositories which are probably better suited for the task, so I'm trying to avoid duplicating their work..

ghost commented 5 years ago

Thank you.