Types of Machine Learning
Last topic you graded emails spam or not-spam and trained a filter, and that answer column is the whole reason it worked. Delete the column and you can still learn something from those emails, just not the same thing.
- ▸The same inbox, minus one column, becomes a different kind of machine: keep the graded answers or delete them
- ▸With no answer column, no grouping can be wrong; there is nothing to grade it against
- ▸One kind gets no table at all; a bot learns only from a score it earns by acting
Start with the inbox you already built. Every email in it carries a grade you supplied, spam or not-spam, filled in on every single row. That filled-in answer column is the thing that made a filter possible: the model studied the rows, saw the answer sitting next to each, and learned to produce that same answer for mail it had never seen.
This is supervised learning, and the name is literal. Every training example came with the right answer attached, like a stack of flashcards with the solution printed on the back. The model is supervised because a human already decided, for every row, what the correct call was.
Notice what that buys and what it costs. The model can only ever hand back an answer from the set you graded, spam or not-spam, never a third thing it invented. And it only works because you had those answers to begin with. Take the answer column away and there is nothing to predict toward, the whole setup falls apart.