(Day 90) Probability - Univariate Models and colab 0 from XCS224W - ML with Graphs

Ivan Ivanov · March 31, 2024

Hello :) Today is Day 90!

A quick summary of today:

Probabilistic Machine Learning: An Introduction

Chapter 2: Probability: Univariate Models

image image image image image image image image image image

Today I got added to the slack channel for XCS224W: ML with Graphs by Stanford, and I went over setting up the environment and over the 0th colab - some basics about graphs. I am not allowed to share any of the code from the colabs, but I found this similar tutorial that is public and I can talk over how I improved upon it. (I need to go over the animation part of this tutorial and showing the nodes in 3d space)

To avoid any risk of a lawyer sending me an email, I will just say how I improved upon the above tutorial.

After training a GNN on KarateClub data from pytorch_geometric, the result looked like

image

And there was a conclusion under it saying how even a simple model, can learn to separate and classify the nodes. Given that final visualization, I was not sure how one can come up with that conclusion, so I looked around to learn a bit more about the dataset, and how graphs work, and the final new visualization I got is:

image

And now we can easily make the conclusion that our simple GNN classifies almost all nodes correctly and separates them well enough.

That is all for today!

See you tomorrow :)