clinikit

Prepared by Berat Kaan SEVEN

A lightweight, sklearn-compatible Python toolkit for tabular machine learning. clinikit bundles 14 hybrid classifiers, 5 experiment protocols, calibration utilities, label-noise diagnostics, fairness audits, and structured HTML reports behind a single drop-in package.

Research and development use only. This is an integration toolkit, not a regulated product and not a research paper of original methods. See CITATIONS.md for source-method references.

Why clinikit

clinikit is a complement to existing libraries, not a competitor.

Installation

pip install clinikit

Optional dependency groups:

pip install "clinikit[diagnostics]"   # Cleanlab-based label-noise tools
pip install "clinikit[explain]"       # SHAP and LIME wrappers
pip install "clinikit[automl]"        # TabPFN, FLAML, AutoGluon wrappers
pip install "clinikit[synthetic]"     # CTGAN / TVAE wrappers
pip install "clinikit[conformal]"     # MAPIE conformal prediction
pip install "clinikit[all]"           # Everything

Supported Python versions: 3.10, 3.11, 3.12, 3.13.

Quickstart

from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split

from clinikit.datasets import load_pima
from clinikit.metrics import sensitivity, specificity
from clinikit.models import RuleAugmentedClassifier

X, y = load_pima(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 = RuleAugmentedClassifier(base_estimator=LogisticRegression(max_iter=1000))
model.fit(X_train, y_train)
y_pred = model.predict(X_test)

print("Sensitivity:", sensitivity(y_test, y_pred))
print("Specificity:", specificity(y_test, y_pred))

For a complete walkthrough, see examples/quickstart.ipynb or open it in Colab.

What is in the box

14 hybrid classifiers (clinikit.models)

All sklearn-compatible, all pass sklearn.utils.estimator_checks.check_estimator.

• RuleAugmentedClassifier
• BoundaryRefineClassifier
• SubgroupThresholdClassifier
• ErrorAwareCalibrator
• MonotonicBooster
• HardSampleWeightedEnsemble
• ClassConditionalImputer
• CrossDistributionDistiller
• SelectiveClassifier
• InstanceAdaptiveThreshold
• DialecticalEnsemble
• LatentSubtypeRouter
• IterativeLabelRefiner
• DualViewCoTrainer

Supporting modules

• preprocessing — imputers, scalers, outlier flags, missing indicators
• metrics — sensitivity, specificity, NPV, PPV, F2, MCC, Brier, ECE
• curves — ROC, PR, calibration, Decision Curve Analysis
• protocols — 5 experiment protocols (Defensible, MaxScore, OriginalOnly, Deployment, Audit)
• leaderboard — experiment tracking CSV with 38 columns
• report — HTML structured report generator (Jinja2 templates)
• audit — leakage detection, subgroup fairness, documentation checks
• governance — audit-trail manifest templates (documentation only)
• reproducibility — manifest files (data hash + config + library versions)
• datasets — UCI benchmarks (PIMA, Wisconsin, UCI Heart, Frankfurt)
• cli — Typer-based CLI: train, benchmark, audit, validate, report
• thresholds, calibration, statistics, diagnostics, cost_sensitive, monitor, modelcard, cross_val, explainability, automl, external_val, time_split, active_learning, synthetic

Command-line interface

clinikit train      --config config.yaml
clinikit benchmark  --dataset pima --models all
clinikit audit      --data data.csv --report audit.html
clinikit validate   --model model.joblib --data data.csv
clinikit report     --leaderboard runs.csv --out report.html

Project notes

clinikit is an integration toolkit. The methods it bundles are adaptations of techniques published in the academic literature; see CITATIONS.md for source-method references. It is not a research paper of original methods, and it is not a regulated product. Research and development use only.

Contributing

Contributions are welcome. Please read CONTRIBUTING.md for the development workflow, coding standards, and pull-request process. By participating, you agree to abide by the Code of Conduct.

Citation

If you use clinikit in academic work, please cite it via the CITATION.cff file, or use:

@software{clinikit,
  author  = {SEVEN, Berat Kaan},
  title   = {clinikit: a tabular machine-learning toolkit},
  year    = {2026},
  url     = {https://github.com/clinikit/clinikit},
  version = {0.1.0}
}

License

Distributed under the MIT License. See LICENSE for the full text.