혼합 자료형을 위한 ColumnTransformer

Real tables mix numbers and categories, and each kind needs different prep. One scaler for everything simply will not work. 🧩

A ColumnTransformer applies different transformers to different column groups, all in one step you can drop inside a pipeline.

First decide which columns are numeric and which are categorical. You will hand each group its own transformer by name.

Numeric columns usually want a StandardScaler so large and small values share a fair footing before modeling.

Categorical columns need numbers, so reach for OneHotEncoder to turn each category into its own 0/1 column.

Each entry is a tuple of name, transformer, and the columns it touches. The ColumnTransformer runs them side by side.

Place the transformer as the first step and a model as the last. Now mixed-type prep and modeling live in one pipeline.

Columns you did not list are dropped by default. Set remainder to passthrough if you want to keep the rest untouched.

Test data may contain a category training never saw. Set handle_unknown to ignore so the encoder stays calm instead of crashing.

Tired of typing column names? make_column_selector grabs columns by dtype so your prep adapts as the table changes.

Calling fit once trains the scaler and the encoder together on the right columns. Each path learns only from its own columns.

You have numeric and categorical columns needing different prep. What handles that?

You can now route numeric and categorical columns to their own transformers with a ColumnTransformer inside one pipeline. Next, tuning. 🔧