Skip to content
LearnPY
scikit-learn

Prepare features without leakage

Handle missing values, scale numeric columns, encode categories, and combine those steps with ColumnTransformer.

14 min readbeginnerscikit-learn, numpy, pandas

By the end

  • Impute missing numeric values with training statistics.
  • Scale features after splitting the data.
  • One-hot encode categorical columns.
  • Build a ColumnTransformer for mixed data.
  • Keep preprocessing inside a fitted pipeline.

Missing values need a rule

Most scikit-learn estimators do not accept NaN. SimpleImputer replaces missing entries with a learned statistic such as the mean or median. Fit the imputer on training data, then use the same learned values on test data.

Python

Learn imputation values on the training split

Ready
Loading editor…

Output

Press Run to execute this cell. The first run downloads Python; later runs reuse the session.

Scale on the training distribution

StandardScaler subtracts the training mean and divides by the training standard deviation. The transformed training columns have mean 0 and standard deviation 1; the test columns use those same training values, so they will not necessarily be centered at 0 themselves.

Python

The test set is transformed, not refitted

Ready
Loading editor…

Output

Press Run to execute this cell. The first run downloads Python; later runs reuse the session.

Scaling matters for distance-based models such as K-nearest neighbors, for regularized linear models, and for PCA. Tree-based models usually do not need it.

Turn categories into columns

OneHotEncoder creates one binary column per known category. handle_unknown="ignore" keeps prediction working when a new category appears, and sparse_output=False returns a dense array that is easy to inspect.

Python

One binary column per category

Ready
Loading editor…

Output

Press Run to execute this cell. The first run downloads Python; later runs reuse the session.

Apply different rules to different columns

Real tables mix numeric and categorical columns. ColumnTransformer applies one preprocessing pipeline to the numeric columns and another to the categorical columns, then concatenates the results.

Python

A mixed-data preprocessing recipe

Ready
Loading editor…

Output

Press Run to execute this cell. The first run downloads Python; later runs reuse the session.

  • Numeric columns are imputed with the training median, then standardized.
  • The city column becomes one binary column per known city.
  • The fitted preprocessor is now ready to sit in front of a model.

Make preprocessing part of the model

The safest place for preprocessing is inside the same Pipeline as the estimator. Then cross-validation, fitting, and prediction all reuse the same learned rules without manual bookkeeping.

Python

Cross-validation would refit this whole recipe

Ready
Loading editor…

Output

Press Run to execute this cell. The first run downloads Python; later runs reuse the session.

Practice

Make it stick

Run the tests as often as you like. Hints and the solution are available when you need them.

Exercise

Build a ColumnTransformer named preprocessor. Apply median imputation then StandardScaler to age and spend, and OneHotEncoder(handle_unknown="ignore", sparse_output=False) to city. Fit it on X_train and transform both X_train and X_test into X_train_ready and X_test_ready.

Your solution

Ready
Loading editor…

Results

Run the tests to check your solution. The first run downloads Python; later runs reuse the session.

Quiz

Check your understanding

01Where should SimpleImputer learn its fill values?
02What does handle_unknown="ignore" do for unseen categories?
03What is ColumnTransformer for?

Lesson progress