Reinforcement Learning in TensorFlow with TF-Agents (TF Dev Summit '19)

[@dinoscheidt]

Unity was mentioned briefly on a slide for future environments. As soon unity3d is pluggable, this is going to be huge (under i.e. distributed learning).

[@yesyouarerightbut3200]

Unity Sucks. Go for UE4.

[@NN-qj4sk]

@@yesyouarerightbut3200 Why?

[@siddharthmittal9355]

with the unity support i guess it will create some innovative handshake between game development and deep learning/reinforcement learning.

[@AryanSingh-rq1nw]

Will have to see whether this streamlines the learning algorithms. Otherwise all this can be achieved via vanilla python code

[@EnderDeaD14]

wow very nice lib !

[@alihusham1560]

does your graphics intros build by AI?

[@ryashaum]

Where can I get the slides

[@matejracinsky5324]

What exactly is Tensorflow 1.14? I can't see thiws version anywhere in the Github versions.

[@eugenebrevdo36]

It hasn't been released yet. As of today the best snapshots of Tensorflow 1.14 are the tf-nightly and tf-nightly-gpu pip packages.

[@deep.space.12]

Thanks for open sourcing a Google-patented algorithm (DQN). Like... what's the point? Who even to contact for licensing?

[@jonabirdd]

@@ahmadayazamin3313 Lol, you can release open source code for patented ideas. It just means google can technically sue you for using their open source libraries in your commercial product. Double win lol.

[@karanbirchahal3268]

Nice !

[@oliverli9630]

if only it was released earlier lol

[@jg9193]

fancy

[@bruhaspati560]

Stable-baselines is better than this!