•1 min read•from Machine Learning
[R] Best practices for implementing and benchmarking a custom PyTorch RL algorithm?
Our take
Implementing and benchmarking a custom PyTorch reinforcement learning (RL) algorithm can be an exciting yet challenging endeavor. To ensure success, it’s essential to familiarize yourself with resources that guide the development of custom algorithms within PyTorch. Streamlining your code and establishing a well-organized directory structure can enhance readability and collaboration. It’s also important to identify target environments for benchmarking, including considerations for dockerization, especially if your development starts on a Mac but needs compatibility with Linux.
Hey, I'm working on a reinforcement learning algorithm. The theory is complete, and now I want to test it on some Gym benchmarks and compare it against a few other known algorithms. To that end, I have a few questions:
- Is there a good resource for learning how to build custom PyTorch algorithms?
- How optimized or clean does my code need to be? Should I spend time cleaning things up, creating proper directory structures, etc.?
- Is there a known target environment or standard? Do I need to dockerize my code? I'll likely be writing it on a Mac system. Do I also need to ensure it works on Linux?
[link] [comments]
Read on the original site
Open the publisher's page for the full experience