Machine learning

Big picture: a model, a cost, and a way to pick parameters that make the cost small. Classification is the same loop with a different model. Neural networks are that loop with more layers.

Reading order

1. Introduction: supervised learning, and what it means for a program to learn.
2. Model and cost function: linear regression, and the cost of a bad fit.
3. Parameter learning: gradient descent.
4. Multivariate linear regression: more than one input.
5. Computing parameters analytically: the normal equation, when you can skip the iteration.
6. Classification: predicting a category instead of a number.
7. Logistic regression: the model and the cost for that prediction.
8. Multiclass classification: more than two categories.
9. Overfitting: a model that memorized the training set, and how to hold it back.
10. Neural networks: layers of those models.