Centering and scaling

Supervised Learning with scikit-learn

George Boorman

Core Curriculum Manager

Why scale our data?

print(music_df[["duration_ms", "loudness", "speechiness"]].describe())

         duration_ms     loudness       speechiness
count    1.000000e+03    1000.000000    1000.000000
mean     2.176493e+05    -8.284354      0.078642
std      1.137703e+05    5.065447       0.088291
min      -1.000000e+00   -38.718000     0.023400
25%      1.831070e+05    -9.658500      0.033700
50%      2.176493e+05    -7.033500      0.045000
75%      2.564468e+05    -5.034000      0.078642
max      1.617333e+06    -0.883000      0.710000

Why scale our data?

Many models use some form of distance to inform them
Features on larger scales can disproportionately influence the model
Example: KNN uses distance explicitly when making predictions
We want features to be on a similar scale
Normalizing or standardizing (scaling and centering)

How to scale our data

Subtract the mean and divide by variance
- All features are centered around zero and have a variance of one
- This is called standardization
Can also subtract the minimum and divide by the range
- Minimum zero and maximum one
Can also normalize so the data ranges from -1 to +1
See scikit-learn docs for further details

Scaling in scikit-learn

from sklearn.preprocessing import StandardScaler

X = music_df.drop("genre", axis=1).values
y = music_df["genre"].values

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, 
                                                    random_state=42)

scaler = StandardScaler()

X_train_scaled = scaler.fit_transform(X_train)

X_test_scaled = scaler.transform(X_test)

print(np.mean(X), np.std(X))
print(np.mean(X_train_scaled), np.std(X_train_scaled))

19801.42536120538, 71343.52910125865
2.260817795600319e-17, 1.0

Scaling in a pipeline

steps = [('scaler', StandardScaler()),
         ('knn', KNeighborsClassifier(n_neighbors=6))]
pipeline = Pipeline(steps)

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, 
                                                    random_state=21)

knn_scaled = pipeline.fit(X_train, y_train)

y_pred = knn_scaled.predict(X_test)

print(knn_scaled.score(X_test, y_test))

0.81

Comparing performance using unscaled data

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, 
                                                    random_state=21)
knn_unscaled = KNeighborsClassifier(n_neighbors=6).fit(X_train, y_train)
print(knn_unscaled.score(X_test, y_test))

0.53

CV and scaling in a pipeline

from sklearn.model_selection import GridSearchCV
steps = [('scaler', StandardScaler()),
         ('knn', KNeighborsClassifier())]
pipeline = Pipeline(steps)

parameters = {"knn__n_neighbors": np.arange(1, 50)}

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, 
                                                    random_state=21)

cv = GridSearchCV(pipeline, param_grid=parameters)

cv.fit(X_train, y_train)

y_pred = cv.predict(X_test)

Checking model parameters

print(cv.best_score_)

0.8199999999999999

print(cv.best_params_)

{'knn__n_neighbors': 12}

Let's practice!

Supervised Learning with scikit-learn