Classification - Hollerith

You can use Hollerith to predict which category a row belongs to. The classifier can be used for both binary and multi-class classification on tables.

Getting started

The following is a full example of using the Hollerith classifier. This is a simple binary classification problem for a hotel booking dataset, where we predict if a hotel booking will be cancelled or kept based on details at time of reservation.

from hollerith import Hollerith, read_csv

df = read_csv("hotel-cancellations.csv")
train = df[df["__hollerith_split"] == "train"].drop(columns="__hollerith_split")
score = df[df["__hollerith_split"] == "test"].drop(
    columns=["__hollerith_split", "booking"]
)

clf = Hollerith()
clf.fit(train, target="booking", task="classification")

preds = clf.predict(score)
print(preds[:5])
# ['kept', 'cancelled', 'kept', 'kept', 'cancelled']

Multi-class

You can use the same call to also perform multi-class classification, as shown below.

tiers = Hollerith().fit(accounts, target="tier")
tiers.classes_
# ['enterprise', 'free', 'pro']

Class probabilities

proba = clf.predict_proba(score)
proba.shape
# (3990, 2)
clf.classes_
# ['cancelled', 'kept']

cancel_score = proba[:, clf.classes_.index("cancelled")]

Evaluating accuracy

Learn how Hollerith scores itself

Regression

Learn about Hollerith’s regression capabilities