Quick Start

KFC Procedure provides estimators for classification, regression, and ensemble learning.

This page introduces the main public estimators:

• CombinedClassifier
• MixCOBRARegressor
• GradientCOBRA
• KFCProcedure
• KFCClassifier
• KFCRegressor

---

CombinedClassifier

Use CombinedClassifier to combine predictions from multiple classification models.

from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split

from kfc_procedure.cobra import CombinedClassifier

X, y = load_iris(return_X_y=True)

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

model = CombinedClassifier(
    random_state=42,
)

model.fit(X_train, y_train)

predictions = model.predict(X_test)

print(predictions[:5])

---

MixCOBRARegressor

MixCOBRARegressor combines information from the feature space and prediction space to produce regression predictions.

from sklearn.datasets import load_diabetes
from sklearn.model_selection import train_test_split

from kfc_procedure.cobra import MixCOBRARegressor

X, y = load_diabetes(return_X_y=True)

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

model = MixCOBRARegressor(
    random_state=42,
)

model.fit(X_train, y_train)

predictions = model.predict(X_test)

print(predictions[:5])

---

GradientCOBRA

GradientCOBRA is a regression ensemble that combines predictions using distance, kernel, and aggregation components.

from sklearn.datasets import load_diabetes
from sklearn.model_selection import train_test_split

from kfc_procedure.cobra import GradientCOBRA

X, y = load_diabetes(return_X_y=True)

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

model = GradientCOBRA(
    random_state=42,
)

model.fit(X_train, y_train)

predictions = model.predict(X_test)

print(predictions[:5])

---

KFCProcedure

KFCProcedure is the general three-stage KFC estimator:

K-step → F-step → C-step

You explicitly define whether the task is regression or classification.

from sklearn.datasets import load_diabetes
from sklearn.model_selection import train_test_split

from kfc_procedure import KFCProcedure

X, y = load_diabetes(return_X_y=True)

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

model = KFCProcedure(
    divergences=["euclidean"],
    local_model="linear_regression",
    combiner="mean",
    task="regression",
    n_clusters=3,
    random_state=42,
)

model.fit(X_train, y_train)

predictions = model.predict(X_test)

print(predictions[:5])

---

KFCClassifier

KFCClassifier is the classification-specific KFC estimator.

The task is automatically configured as classification.

from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split

from kfc_procedure import KFCClassifier

X, y = load_iris(return_X_y=True)

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

model = KFCClassifier(
    divergences=["euclidean"],
    local_model="logistic_regression",
    combiner="majority_vote",
    n_clusters=3,
    random_state=42,
)

model.fit(X_train, y_train)

predictions = model.predict(X_test)

print(predictions[:5])

---

KFCRegressor

KFCRegressor is the regression-specific KFC estimator.

The task is automatically configured as regression.

from sklearn.datasets import load_diabetes
from sklearn.model_selection import train_test_split

from kfc_procedure import KFCRegressor

X, y = load_diabetes(return_X_y=True)

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

predictions = model.predict(X_test)

print(predictions[:5])

---

Which estimator should I use?

Estimator	Task	Purpose
CombinedClassifier	Classification	Combine multiple classifiers
MixCOBRARegressor	Regression	Mix feature-space and prediction-space information
GradientCOBRA	Regression	Kernel-based COBRA ensemble
KFCProcedure	Both	General configurable KFC pipeline
KFCClassifier	Classification	KFC classification pipeline
KFCRegressor	Regression	KFC regression pipeline