Summary - Scaling

In this group you learned StandardScaler, MinMaxScaler, and Pipeline.

Summary: scaling

In this group:

Features in different units break distance- and gradient-based models; trees aren't affected.
StandardScaler (z-score), MinMaxScaler ([0, 1]), RobustScaler (median/IQR).
The scaler is fit only on training data; inverse_transform converts back.
A Pipeline combines preprocessing and the model, prevents leakage and is tuned with step__parameter.

Exercise Choosing a scaler

For x, which contains an outlier, print the MinMaxScaler and RobustScaler results with ravel().round(2) on separate lines.

Show example solution

import numpy as np
from sklearn.preprocessing import MinMaxScaler, RobustScaler

x = np.array([[10.0], [12.0], [11.0], [13.0], [90.0]])

print(MinMaxScaler().fit_transform(x).ravel().round(2))
print(RobustScaler().fit_transform(x).ravel().round(2))

Test yourself

Which causes data leakage?
- Fitting the scaler on X_train
- Training a tree model without scaling
- Fitting the scaler on all data and splitting afterwards
- Using a pipeline
Show answer

Fitting the scaler on all data and splitting afterwards — The test set's statistics mix into the scale; the score comes out optimistic.

Create a pipeline that names the steps automatically.

pipe = ____(StandardScaler(), LogisticRegression())

Show answer

pipe = make_pipeline(StandardScaler(), LogisticRegression())

Match the formulas.

StandardScaler

(x - mean) / std
MinMaxScaler

(x - min) / (max - min)
RobustScaler

(x - median) / IQR
inverse_transform

Back to original unit

Next lesson: What is a feature?

We Are Enhancing Your Experience

Cookie Preferences

Strictly Necessary Cookies

Analytical Cookies