A personal, experimental dive learning how to build C++ based machine learning systems.
- I ultimately want to explore the use of deep reinforcement learning to democratise the manufacturing domain
- I'm fascinated by the work of Deepmind / David Silver (loved his UCL lecture series)
- I'm enjoying reading Reinforcement Learning: An Introduction: Sutton, Barto (2018)
- I'm enjoying reading Deep Reinforcement Learning Hands-On: Maxim Lapan (2018)
- I think that https://gym.openai.com/ is cool
- I like bullet as a simulation environement, maybe even try godot?
- I learn by doing e.g. re-writing things from python -> C++
- PyTorch now has a wonderful C++ API
- I want my GPU to do something interesting
- I want to build the whole thing as a container native / ready to scale
- https://github.com/Omegastick/pytorch-cpp-rl (such as linking to torch C++ <-> openai gym with ZMQ)
- https://github.com/Shmuma/ptan / https://github.com/PacktPublishing/Deep-Reinforcement-Learning-Hands-On
- https://github.com/lupoglaz/GodotGymAI
- Build Godot gym module https://github.com/thomas-gale/godot-gym-cpp
- Attempt to roughly follow exercises in Deep Reinforcement Learning Hands-On: Maxim Lapan (2018) but using godot game approximations and rewritten in C++
- Possibily give learn library similar structure to ptan?
- Finally, do something cool (maybe with b2a - build in a learning agent into the service providers of the virtual factory marketplace with the reward function set to maximise profit...)
git clone
(this repo)git submodule update --init --recursive
- Open directory in VSCode
- Follow the 'Reopen in container prompt'
- Enable things that pop up - like allowing C++ intellisense to use CMake
- VSCode command pallette (ctrl+shft+p)
CMake: Configure
- Choose kit
GCC for x86_64-linux-gnu 7.5.0
Cmake: Build
- The target selection, debug and run commands are available at the bottom of the screen in VSCode