Code for Response coding for binary/multi classification tasks using Python.

response_coding.py

import numpy as np
import pandas as pd

# Column names of categorical features
categorical_features = (
    "school_state",
    "teacher_prefix",
    "project_grade_category",
    "clean_categories",
    "clean_subcategories",
)

# Name of target variable column
target = "project_is_approved"

response_coding = {}
classes = sorted([0, 1])

for category in categorical_features:
    cat = {}
    for sub_category in pd.unique(X_train[category]):
        count_dict = pd.value_counts(
            X_train[X_train[category] == sub_category][target]
        ).to_dict()
        
        count = [count_dict.get(cls, 0) for cls in classes]
        prob = tuple(map(lambda ele: ele / sum(count), count))
        cat.update({sub_category: prob})
        
    response_coding.update({category: cat})


def get_response_coding(df: pd.DataFrame) -> np.array:
    """
    Returns response of type np.array of size (rows, no. of categorical features * n_classes)
    """
    n_classes = len(classes)
    n_rows = df.shape[0]
    n_cols = len(categorical_features) * n_classes
    default = tuple([1 / n_classes] * n_classes)

    response = np.zeros((n_rows, n_cols), dtype=float)

    for category in categorical_features:
        start = categorical_features.index(category) * n_classes
        cols = (start, start + n_classes)

        for sub_category in pd.unique(X_train[category]):
            rows = (df[category] == sub_category).values
            response[rows, cols[0] : cols[1]] = list(
                response_coding[category].get(sub_category, default)
            )

    return response


train_response = get_response_coding(X_train)
test_response = get_response_coding(X_test)

Change categorical_features and target variables to fit your dataset.

train and test dataframes are named as X_train and X_test respectively.