O contrato de fit e predict

In scikit-learn, every model is an estimator: one object you create, teach, and then ask for answers. Same shape, every time. 🤖

The whole library rests on two methods: fit to learn from data, and predict to use what it learned. Master these and you can use almost any model.

Calling fit shows the model your examples so it can find patterns. Nothing is predicted yet, the model is simply studying the data.

Once trained, predict takes fresh inputs and returns the model's best guesses. This is where the learning finally pays off.

This fit-then-predict pattern is a contract every estimator honors. Swap a tree for a linear model and your code barely changes.

You always build the estimator first, often with settings, before any data touches it. That blank model is ready to learn.

You must fit before you predict. Asking an untrained model for answers raises an error, since it has learned nothing yet.

The fit call also returns the model itself, so you can chain steps in one line when you want compact, readable code.

After fitting, the model stores what it learned in attributes ending with an underscore, like coef_. They appear only once training is done.

Because the API is shared, you can try several models by changing one line. The fit and predict calls stay identical.

One predictable interface means less to memorize and faster experiments. You focus on the problem, not on each library's quirks.

Let's lock in the core contract every estimator follows.

Every estimator follows one contract: create it, call fit to learn, then predict to answer. One pattern unlocks the whole library. 🎯