prob-target-encoder 0.1.2

A probability-weighted target encoder.

Probability Target Encoder

A Python library for encoding categorical features using a probability-weighted target encoding technique.

This encoder extends the traditional target encoder by incorporating the probability that a target value exceeds the global target mean, and then uses the variance to increase confidence, producing encodings that capture both the average target value and the likelihood of above-average outcomes which has a higher confidence when the data isn't varied.

Features

• Handles categorical variables with numerical targets.
• Supports missing values.
• Encodes unseen categories using a fallback.
• Simple API with fit(), transform() and fit_transform() methods.
• Lightweight implementation with minimal dependencies.

Installation

pip install prob-target-encoder

Quick Start

from prob_target_encoder import ProbTargetEncoder

encoder = ProbTargetEncoder()

encoder.fit(train_column, train_target)

encoded_train = encoder.transform(train_column)
encoded_test = encoder.transform(test_column)

Example

import pandas as pd
from prob_target_encoder import ProbTargetEncoder

X = pd.Series(["A", "B", "A", "C", "B"])
y = pd.Series([10, 20, 15, 8, 25])

encoder = ProbTargetEncoder()
encoder.fit(X, y)

encoded = encoder.transform(X)

print(encoded)

How It Works

For each category, the encoder computes:

The mean target value for that category. The probability that the target exceeds the global target mean. the variance of the target value for each category. An encoding derived from these statistics.

Requirements

Python 3.9+ pandas

License

This project is licensed under the MIT License.