Introducing OpenAI Lab for Reinforcement Learning

Due to the popularity of our last reinforcement learning event with Keng and Laura, we're running a sequel!

Our speakers for this event, Wah Loon Keng and Laura Graesser, have created OpenAI Lab to do Reinforcement Learning (RL) like science.

The Lab provides an easy to use interface to OpenAI Gym and Keras, combined with an automated experimental and analytics framework. OpenAI Gym is a training ground for developing reinforcement learning (RL) algorithms; Keras is a high-level neural networks library that uses Tensorflow or Theano.

While these are powerful tools, they take a lot to get running. In many implementations which solve OpenAI gym environments, we often have to rewrite the same basic components. To address this, OpenAI Lab does three things:

1. Handles the basic RL environment and algorithm setups.
2. Provides a standard, extensible platform with reusable components for developing deep reinforcement learning algorithms.
3. Provides a rigorous experimentation system with logs, plots and analytics for testing new RL algorithms. Experimental settings are logged in a standardized format so that solutions can be reproduced by anyone using the Lab.

With the Lab, we can focus on researching the essential elements of reinforcement learning such as the algorithm, policy, memory (experience replay), and parameter tuning to solve the OpenAI environments. We can also test our hypotheses more reliably.

This talk is an introduction to OpenAI Lab. We’ll cover the following topics:
 Introduction to Reinforcement Learning
 Brief run through of OpenAI Gym
 Introduction to the OpenAI Lab

• Running an experiment, to breed solution RL agents
• Parameter tuning - 1 variable
• Searching the hyperparameter space of multi-variables
• Analyzing your results

We recommend that you bring a laptop if you would like to follow along. Before the talk, complete the instructions at http://kengz.me/openai_lab/ to install the required libraries.

Speakers:

Wah Loon Keng is a software engineer at Eligible API. He's a mathematician at heart, who specializes in graphs, complexity, computation, and their roles in artificial intelligence and computational cognitive science. Keng also actively contributes to open source projects such as TFLearn and spaCy.

Laura Graesser is studying for an MS in computer science at NYU, focusing on machine learning. Laura is particularly interested in neural networks and their application to computer vision problems, cross-fertilization between computer vision and NLP, the representations perspective (machine learning as data transformation and representation), and the manifold hypothesis.

Hosted by Squarespace (https://www.squarespace.com/).

---

Join the new NYC Data Wranglers page on Facebook at: https://www.facebook.com/groups/NYCData/