Motivation

We want to predict something. Our approach depends on this something.

General Approach

input: X output: Y

We divide the process into 3 parts, training, testing and the actual usage.

Training

What are we training? -> a model or simply a function, say f. We want to pass our input, X, into f and get y. If I have rainfall data of the past month or year of a certain place, I can expect the function to output the amount of rainfall for tomorrow. But would our expected output always be a number? What if I want to know whether the picture I’m giving as input is of a dog or a cat? A ‘dog’ or a ‘cat’ is not a number. We will find out how to handle this later.

If y matches Y, then we can say that the function is trained.

But where do we start with f. Well, the simplest function (except constant) we can think of is the linear function i.e.

Now we have a function to work with. What can we change in this function so that it gets trained? x is passed to the function, that leaves m and c. These are called the trainable parameters of the model.