| TensorFlow |
YES |
YES |
YES |
|
-Implementation of RL algorithms, -Simple RL with TensorFlow, -TensorFlow ML tutorials |
Python (C++, Java, Go...) |
Developed by Google. |
| Theano |
YES |
YES |
YES |
- Atari implementation |
-Theano tutorial, -Theano tutorial (deeplearning.net), -Theano exercises, -AgentNet DRL library |
Python |
|
| Keras |
YES |
YES |
YES |
-Flappy bird with Keras |
- keras-rl, -qlearning4k |
Python |
High-level library on top of TensorFlow/Theano. |
| Torch |
YES |
YES |
YES |
-Nature Atari Deep Q Learner |
-Tutorials for learning Torch, -Awesome-torch (extensive list of nice tutorials), -Twitter RL with Torch , -Machine Learning (Nando de Freitas) |
LuaJIT |
Learn Lua here. |
| PyTorch |
YES |
YES |
YES |
- |
Tutorial, Examples |
Python |
"While still a new framework with lots of ground to cover to close the gap with its competitors, PyTorch already has a lot to offer." |
| Caffe |
YES |
YES |
YES |
-DQN-in-the-Caffe (Atari 2600), -Space Invaders (Atari 2600, First OS implementation) |
|
Python |
They believe that Caffe is the fastest convnet implementation available. |
| Neon |
Required, only works with latest NVIDIA Maxwell cards. |
YES |
YES |
-Model Zoo, -DQN Atari |
-Tutorials, -Resources |
Python |
Developed by nervana. |
| rllab |
? |
? |
? |
-Several implementations in GitHub |
|
Python |
Framework for developing and evaluating reinforcement learning algorithms (uses OpenAI, Theano and has support for TensorFlow). It includes a wide range of continuous control tasks. |
| OpenAI Gym |
- |
NO |
YES |
- |
Official Docs |
Python |
Framework to test, evaluate and compare reinforcement learning algorithms. |