train_test_split 제대로 사용하기

The function train_test_split takes your features and target and hands back train and test pieces in one tidy call.

It returns four objects in a fixed order: train features, test features, train target, test target. Unpack them carefully so each lands in the right name.

Use test_size to choose how much data to reserve. A value of 0.2 means 20% goes to the test set.

By default the rows are shuffled before splitting. That mixing prevents any hidden order in your file from biasing either set.

Pass random_state to lock the shuffle. The same number gives the same split every run, so your results are repeatable. 🔁

Without a fixed seed, every run reshuffles and your scores wobble. A locked random_state lets teammates reproduce your exact numbers.

If a class is rare, a plain random split might dump most of it into one set. Now train and test no longer reflect the same class balance.

Pass stratify equal to your target so both sets keep the same class proportions. Essential for classification with uneven classes.

The split keeps each row's features and label together. Row 7's features always travel with row 7's label, never scrambled apart.

Common splits are 80/20 or 70/30. With lots of data you can spare a smaller test slice; with little data, give the test set a bit more.

Run the split first, before scaling or filling values. Doing prep on the full set leaks test information back into training.

Your target classes are very imbalanced. Which argument helps?

Unpack four sets, set test_size, fix random_state for repeatability, and use stratify when classes are uneven. Split first, prep later. 🔁