Learning Curves and kNN Vignette

Click here to watch a series of videos on building a full machine learning pipeline in sklearn

Hello Class,

I have recorded a short video tutorial on how to use learning curves to diagnose problems in your machine learning model. I talk about how to identify bias limitation, variance limitation, or hitting a limit due to irreducible error. I discuss what the best course of action is in each scenario, and show how to diagnose convergence rate using a ‘loglog’ plot and irreducible error using ‘kNN’ with 1 nearest neighbor.

Click here to watch Learning Curves and kNN.

If you want to learn more, I recommend reading starting at page 55 of Machine Learning Yearning for information on learning curves, and Chapter 13 of Elements of Statistical Learning for applications of ‘kNN’ including to irreducible error estimation.