Model-Selection on Ryan Orban

Scikit-Learn: Model Validation and Testing (PyCon 2013 Notebook)

me@ryanorban.com (Ryan Orban) — Tue, 11 Mar 2014 00:00:00 +0000

Summary

This is notebook 9 from Jake VanderPlas’s scikit-learn tutorial series for PyCon 2013 — one of the most widely-followed Python machine learning tutorials of its era. The notebooks were designed to be run interactively, making abstract concepts like cross-validation visible through immediate code execution. IPython Notebook (later Jupyter) was the delivery vehicle, and this tutorial helped establish the notebook as the standard format for data science education.

Why You Should Be Spot-Checking Algorithms on Your Machine Learning Problems

me@ryanorban.com (Ryan Orban) — Fri, 07 Feb 2014 00:00:00 +0000

Summary

Jason Brownlee’s post on Machine Learning Mastery makes the case for spot-checking as the starting point of any new ML project. The default mistake: pick an algorithm you like or know well, spend time tuning it, and declare it your model. The spot-checking approach: try 10-15 algorithms from different families (linear models, tree-based models, kernel methods, instance-based learners, ensemble methods) with default parameters and compare their untuned performance first.