Learning Strategies through Reinforcement Learning

Learning Strategies through Reinforcement Learning

By Max Candocia

|

August 31, 2018

This past April I gave a talk on reinforcement learning while visiting UIUC before the Illinois Marathon the following day. The lecture was given to the STAT 385 (Statistics Programming Methods) class taught by James Balamuta, who invites a few guest lecturers to the class per semester to talk about statistics and data science-related topics.

The focus of this talk was reinforcement learning, using the two games I simulated in Python: Machi Koro and Splendor. In contrast to the content I post on the site, the talk was geared more towards the process of working on a project and structuring files/workflow. Since this was an exploratory project, this does does not represent how I would structure data/files in a more serious environment.

Below the embedded video, there are links to the original articles on these topics that I wrote. To view the video on YouTube directly, go to https://www.youtube.com/watch?v=YZtME6NDCAM.

Original Articles

* Note that this article was made a week after the talk was given. The conclusions are largely the same, although the general content is quite different from the video's.


Tags: 

Recommended Articles

"Error Bars" on Tiled Heatmaps

Heatmaps are a useful way of plotting 2-dimensional data, such as cross-tabulations. Adding "error bars" can seem non-intuitive, but expressing them in your visualization is possible with a small trick.

Modifying Naive Bayes for Text Classification

Articles and other bodies of text can be classified with Naive Bayes efficiently by limiting the impact of features on the model.