Chapter 5. Recap and Next Steps

We have learned that Ray is a system for building distributed applications, born out of pragmatic challenges of modern, computationally intensive ML/AI systems yet broadly applicable to applications that need to scale to a cluster for performance and resilience.

If your team is building ML libraries or general-purpose applications, Ray provides high performance and an intuitive, concise core API.

If your team is building ML-based applications, Ray provides libraries for reinforcement learning (RLlib), hyperparameter tuning (Tune), distributed training (SGD), and model serving (Serve). The list of third-party libraries and systems based on Ray is also growing. Two examples of NLP (natural language processing) libraries built on Ray are spaCy and Hugging Face.

The Ray and Anyscale websites and blogs have stories from Ray users in different industries and communities.

Next Steps with Ray

To take the next steps with Ray, check out the following resources:

  • To learn more about the Ray API, go through my Ray class on the O’Reilly Learning Platform. With the material organized into notebooks, you can work with live Ray code, try the Ray replacements for several popular multiprocessing libraries, and try your hand at reinforcement learning with RLlib.

  • The Ray documentation provides a lot of details on the more advanced aspects of the Ray API and on all the included libraries, like RLlib, Tune, SGD, and Serve.

  • If you have Python code that uses Python’s multiprocessing.Pool API, try Ray’s drop-in replacement that breaks through the single-node boundary with a one-line change to your code, the import statement.

  • If you use scikit-learn, then try Ray’s drop-in replacement for joblib, which also lets you scale to a cluster.

  • If you like using coroutines with Python’s asyncio, you can also use Ray actors with this API.

  • Finally, check out Modin for a Ray-based way to spread your Pandas DataFrames over a cluster.

For More Information

The Ray home page provides links to additional information, such as these resources:

Anyscale is the company developing Ray. The Anyscale home page has links to Ray events, live training events, the company blog, and job openings.

Final Thoughts

Thanks for reading this report. I hope you now understand the unique computing challenges that Ray was designed to address. I hope you appreciate how concise and intuitive the Ray API is, while it hides most of the complexities of distributed computing. The Ray ML libraries, like RLlib, demonstrate how well Ray addresses the real-world challenges of distributed computing, which you can leverage, too.

How are you using Ray? I hope you’ll let me know.

..................Content has been hidden....................

You can't read the all page of ebook, please click here login for view all page.
Reset