In this tutorial, I explain the process of building models to fit a dataset using various degrees of polynomials.
I then compare the predictions made by these various models and investigate why a more capable model (one with a higher degree polynomial) overfitted and explain the various remedies to tackle it by making use of more training data points and performing regularization.
The tutorial provides the various derivations & mathematical justifications behind the usage of the error function (sum squared) but then argues that this view of regression is very limiting and if there is an alternative view to modeling that explains these formulations better.
#
PRML - (available for free, thanks to Dr. Bishop) - https://www.microsoft.com/en-us/research/uploads/prod/2006/01/Bishop-Pattern-Recognition-and-Machine-Learning-2006.pdf
The (python) source code used to reproduce the experiment from the book can be found here in the form of colab/jupyter notebook -
https://colab.research.google.com/drive/1_wUKLaMr3pYSHys2Y8foTw_mPvVQbXUr?usp=sharing
or, for read-only view see this
https://ksachdeva.github.io/prml/chapter-1/1_1_example_polynomial_curve_fitting.html
#linearregression
#ridgeregression
#overfitting