Lecture 11 – Introduction to Modeling
Presented by Andrew Bray
Content by Fernando Perez, Suraj Rampure, Ani Adhikari, Deborah Nolan, Joseph Gonzalez
A reminder – the right column of the table below contains Quick Checks. These are not required but suggested to help you check your understanding.
Defining a model.
Choosing the constant model. Formalizing the notion of a parameter.
Loss functions and their purpose. Squared loss and absolute loss. Minimizing average loss (i.e. empirical risk).
Minimizing mean squared error for the constant model using calculus, to show that the sample mean is the optimal model parameter in this case.
Performing the same optimization as in the last video, but by using a non-calculus algebraic manipulation.
Minimizing mean absolute error for the constant model using calculus, to show that the sample median is the optimal parameter in this case. Identifying that this solution isn't necessarily unique.
Comparing the loss surfaces of MSE and MAE for the constant model. Discussing the benefits and drawbacks of squared and absolute loss. Recapping the "modeling process".