MoffranCommunity·50 Agent Skills Included

calibrated_explanations

Calibrated model explanations with uncertainty intervals and counterfactuals

Generates trustworthy explanations for scikit-learn classification and regression models with calibrated uncertainty intervals. Fixes the miscalibration problem in tools like LIME and SHAP by calibrating the model first, then explaining the calibrated output. Produces factual rules, counterfactual alternatives, fairness-aware group calibration, and plain-language narratives for audits and compliance.

78|15

↗

npx skills add Moffran/calibrated_explanations --all -g -y

Available:

Instructs the agent to follow the CE-first lifecycle (fit, calibrate, then explain) using the public WrapCalibratedExplainer API, with platform-specific validation commands and feedback rules.

All Skills in This Repository (50)

Pure Emerald Level Indicators

📦 In Repo

Moffran

ce-standards-gap-analyzer

Verify STD decisions are implemented in code and documentation.

ce-regulatory-compliance

Map calibrated_explanations capabilities to EU regulatory obligations for audit-ready compliance documentation.

ce-release-planner

Generate structured vX.Y.Z implementation plans from RELEASE_PLAN_v1.md with ADR and STD references.

ce-fallback-test

Verify fallback visibility with pytest.warns and caplog in pytest tests.

ce-adr-consult

Identify applicable ADRs and extract binding constraints from docs/improvement/adrs/.

ce-notebook-audit

Scan notebooks for private-member usage, policy violations, and rendering errors.

ce-serializer-impl

Serialize calibrator state with ADR-031 primitives and sha256 checksums.

ce-classification

Calibrate binary and multiclass classification predictions with Venn-Abers calibration.

ce-onboard

Guide new CE users through core invariants and session routing at startup.

ce-release-task

Identify, implement, and verify the next release task from plan files.

ce-test-creator

Generate targeted tests for documented coverage gaps using the ADR-030 workflow.

ce-data-preparation

Validate and preprocess mixed-type data with categorical encoding and mapping export.

Community

Advanced

Frequently Asked Questions

FAQPage Schema

How to install calibrated_explanations?▼

Run `npx skills add Moffran/calibrated_explanations --all -g -y` in your terminal to install all skills in this suite globally.

How is this different from SHAP or LIME?▼

It calibrates the model before explaining it, so every explanation includes a calibrated uncertainty interval instead of reflecting an overconfident model. It also provides built-in counterfactual alternatives and deterministic output.

What models does it work with?▼

Any scikit-learn-compatible classifier or regressor. It supports binary and multiclass classification, regression with conformal intervals, and probabilistic regression through one consistent API.

Can it help with EU AI Act or GDPR compliance?▼

Yes. It maps its capabilities to specific articles, providing per-instance explanations, uncertainty quantification, human-oversight reject policies, and machine-readable audit records.

Does it support fairness analysis?▼

Yes. Mondrian conditional calibration produces separate uncertainty intervals per subgroup, revealing where the model is less reliable for specific groups.