Logo Utrecht University

Methodology & Statistics Data Science Lab

Uncategorized

05/10/2017 – Mutual Information

After discussing the source and uses of Mutual Information, Erik-Jan introduced the concept of Maximal Information Criterion (MIC): a relatively new measure of dependence designed to detect both linear and nonlinear dependencies between two variables.

We held a competition to find a simple (maximum 2 operations) functional relationship that the Maximal Information Coefficient could not find given 300 data points. This competition was won by Daniel using the function f(x) = 9 * (x < 9), which yielded an MIC of 0.

mi.jpg

Photo: Everyone hard at work finding functions of data which are hard to detect.

Discussing the results of our competition gave us new insights into how the MIC determines whether a pattern can be considered a relevant dependence, and how the choice of N = 300 limits its ability to detect complex functional relations!

If you want to view the presentation at your own leisure, click here (PDF).

 


Announcement:

In this meeting, Erik-Jan will present on mutual information, leading up to a non-parametric measure of association that performs well in many different situations & functional forms. Here is a nice associated image to pique your interest:

Capture.PNG

Interactive component: We will have a competition to try to break down on this measure! The winner will receive a prize. Bring your laptop (with R)!

Preparation:  none 🙂

You must be logged in to post a comment.