Disclaimer: I recently took a undergrad AI class, hence pretty new to this whole field, so if you see something that makes you think (why the hell is this guy doing X, or doing X by using Y) please let me know.
So I'm trying to classify process data to indicate whether a process bump test is worth fitting a model to or not (don't worry if that didn't make a bunch of sense), and I have a nice set of features I have extracted from the data sets.
Now I'm trying to implement a MI algorithm to aid with feature selection, and I get answers that are plain wrong (a test feature with no relationship between it and my labels comes back with MI = 1), and I'm not quite sure exactly where I'm going wrong.
Does anyone know of a resource (simpler is better) that goes into the actual implementation of a MI calculation? I feel as if I understand the theory fairly well, but I am missing something silly in my implementation, and could really use a more in-depth look on how to actually calculate MI.
Thanks
[link] [2 comments]