KFCProcedure

kfc-procedure is a Python package for clusterwise predictive modeling and COBRA-based ensemble aggregation.

It is designed for machine learning workflows where a dataset may contain heterogeneous subgroups. Instead of relying on one global model, KFCProcedure clusters observations, trains local models inside clusters, and combines their predictions into a final output.

flowchart LR
    A["Input Data"] --> K["K-step<br/>Cluster observations"]
    K --> F["F-step<br/>Train local models"]
    F --> C["C-step<br/>Combine predictions"]
    C --> Y["Final Prediction"]

Package name and import name

Install with pip install kfc-procedure, but import with import kfc_procedure.

Documentation sections

Audience	Section	Purpose
Normal users	User Documentation	Installation, quick start, examples, FAQ
ML/technical readers	Technical Documentation	Methods, algorithms, math, complexity, diagrams
Developers	Developer Documentation	Project structure, registries, extension, testing
API users	API Reference	mkdocstrings-generated class/module reference

Public API

from kfc_procedure import KFCProcedure, KFCRegressor, KFCClassifier
from kfc_procedure.cobra import GradientCOBRA, MixCOBRARegressor, CombinedClassifier, SuperLearner

Quick installation

pip install kfc-procedure

With COBRA extras:

pip install "kfc-procedure[cobra]"

Minimal regression example

from sklearn.datasets import make_regression
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error
from kfc_procedure import KFCRegressor

X, y = make_regression(n_samples=300, n_features=8, noise=0.2, random_state=42)
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42)

model = KFCRegressor(
    divergences=["euclidean"],
    local_model="linear_regression",
    combiner="mean",
    n_clusters=3,
    random_state=42,
)
model.fit(X_train, y_train)
print(mean_squared_error(y_test, model.predict(X_test)))