Machine Learning 101 -- Lecture Notes

Introduction

I recently gave a little, informal lecture about the basics of machine learning (ML) to a bunch of friends. I have decided to publish the jupyter notebook I have used for the talk on this page as lecture notes for future reference, and because I think it might be useful for other people as well. Be aware that some explanations might not be as in-depth as desired, so feel free to contact me in case of any questions!

You can find the notebook here, as well as a PDF version of it.

The Contents of the Notebook

General concepts introduced:

• Supervised vs. unsupervised learning
• The importance of data preprocessing and visual inspection
• The train/test split
• Hyperparameter tuning based on 'elbow plots'
• The idea of ensemble models to quantify predictive uncertainty

Specific methods used:

• Classification with a support vector machine
• Classification with a decision tree
• Classification with a random forest
• Dimensionality reduction using PCA
• Clustering using k-means

Afterthoughts

Just some days ago, I actually used a regression tree for the very first time myself. Looking back, I think this might be a nice addition to a lecture like the one presented here. Thus, I'll likely write up a few words about regression trees in a future post on this blog.

In hindsight, I should have also said a few words about the problem of under- vs. overfitting, i.e. the famous bias-variance tradeoff. I'd definitely include a remark about this in a future lecture of mine, as this is really a fundamental concept throughout ML.

Finally, my friends asked me whether I'd be willing to also present some kind of "Neural Networks 101" similar to the one shared here. I definitely am, and as soon as I have done so, you'll find the lecture notes on this blog as well :D!