Scikit-learn

As previously mentioned, user needs to call transform method to ensure that the data contains only numerical values before passing it to scikit-learn functions.

Pairwise distances

from sklearn.metrics import pairwise_distances
from gower_metric import Gower

gower = Gower(feature_types=feature_types).fit(df)
df_num = gower.transform(df)

matrix_scikit = pairwise_distances(df_num, metric=gower, n_jobs=-1, ensure_all_finite=False)

Note

When using pairwise_distances from scikit-learn, make sure to set ensure_all_finite=False to avoid errors due to custom handling of NaN values.

KNeighborsClassifier

from sklearn.metrics import accuracy_score, classification_report
from sklearn.model_selection import train_test_split
from sklearn.neighbors import KNeighborsClassifier
from gower_metric import Gower

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

gower = Gower(feature_types=feature_types).fit(X_train)
X_train = gower.transform(X_train)
X_test = gower.transform(X_test)

knn = KNeighborsClassifier(n_neighbors=5, metric=gower, n_jobs=-1)
knn.fit(X_train, y_train)

y_pred = knn.predict(X_test)

# you can now evaluate and print the results
accuracy = accuracy_score(y_test, y_pred)
print(f"Accuracy: {accuracy:.2f}")

print("Classification Report:")
print(classification_report(y_test, y_pred))

HDBSCAN

from sklearn.cluster import HDBSCAN
from sklearn.metrics import silhouette_score
from gower_metric import Gower

gower = Gower(feature_types=feature_types).fit(df)
df_transformed = gower.transform(df)

clusterer = HDBSCAN(metric=gower, min_cluster_size=10, min_samples=5)
clusterer.fit(df_transformed)

labels = clusterer.labels_
silhouette_avg = silhouette_score(df_transformed, labels)
print(f"Silhouette Score: {silhouette_avg:.2f}")