Connecting Reinforcement Learning to Simulation Software

The Data Exchange Podcast: Max Pumperla on why he’s bullish on Tune, training data scientists, and contributing to open source.

SubscribeiTunesAndroidSpotifyStitcherGoogle, and RSS.

In this episode of the Data Exchange I speak with Max Pumperla, deep learning engineer at Pathmind1 and a contributor to many open source projects in data science and machine learning.  Max is speaking on applications of reinforcement learning to simulation problems at the upcoming Ray Summit, a free virtual conference scheduled for Sep 30th and Oct 1st.  Earlier this year I had Pathmind’s CEO Chris Nicholson on this podcast and he described how reinforcement learning might play a role in simulation problems. In this episode, Max provides an update and a technical description of how Pathmind uses reinforcement learning, RLLib, and Tune, to help users of AnyLogic, a widely used software for simulations in business applications.

Scalable machine learning, scalable Python, for everyone: Join Max Pumperla, David Patterson, Michael Jordan, Oriol Vinyals, Manuela Veloso, Azalia Mirhoseini, Zoubin Ghahramani, Wes McKinney, Ion Stoica, Raluca Popa and many other speakers at the first Ray Summit, a FREE virtual conference which takes place Sep 30th and Oct 1st.

Since Max is the maintainer of Hyperopt, we also discussed the state of open source tools for hyperparameter tuning:

    Right now if I look at the open source world there are three big, open source hyperparameter tuning tools: Hyperopt, Optuna, and Tune. Optuna is essentially like a clone of Hyperopt, I would have preferred them to just contribute to Hyperopt instead. Optuna is like a rewrite of Hyperopt, with a slightly better API and a few more features.

    Tune is different because it allows you to actually integrate Hyperopt. … In Tune, you can select the TPE algorithm through Hyperopt. TPE (via Hyperopt) is one of the algorithms Tune makes available on the backend.

    For me, and I say this as a Hyperopt maintainer, Tune is the clear winner down the road. Tune is fairly well architected and it integrates with everything else, and it’s built on top of Ray so it has all the benefits stemming from that as well. … In 2020, I would certainly bet on Tune.

Our conversation opened with teaching data science. Max has just agreed to serve as a part-time faculty member at a new data science program at IUBH in Germany.  I asked him to describe the types of courses he plans to teach and how he plans to prepare his students for jobs in industry.

Subscribe to our Newsletter:
We also publish a popular newsletter where we share highlights from recent episodes, trends in AI / machine learning / data, and a collection of recommendations.

Related content:

[1] I am an advisor to Pathmind.

[Image by Hillary Wimsatt from Pixabay ]

1 Comment

Comments are closed.