agi-app-sklearn-pipeline

agi-app-sklearn-pipeline publishes the sklearn_pipeline_project AGILAB app as a self-contained PyPI payload. It keeps a normal scikit-learn pipeline workflow intact while making the executable app boundary, arguments, and evidence artifacts explicit.

Purpose

Use this package to run a compact classic ML proof: generate a deterministic classification dataset, train a StandardScaler + LogisticRegression pipeline, and persist the model, metrics, predictions, report, and run manifest together.

Installed Project

The distribution name is agi-app-sklearn-pipeline; the AGILAB project name is sklearn_pipeline_project. The package exposes both sklearn_pipeline and sklearn_pipeline_project through the agilab.apps entry point group, so AgiEnv(app="sklearn_pipeline_project") resolves the project without a monorepo checkout.

Install

pip install agi-app-sklearn-pipeline

The app project installs scikit-learn when AGILAB prepares its project environment. The payload package stays lightweight and only exposes the project root.

Run In AGILAB

Select sklearn_pipeline_project, then open ORCHESTRATE. Keep the defaults, run Deploy scheduler & workers, then run RUN. The worker exports model quality evidence and a manifest under sklearn_pipeline/evidence.

Expected Inputs

The default run generates a synthetic dataset with sklearn.datasets.make_classification. No external dataset, API key, notebook, cloud service, or private model is required.

Expected Outputs

The worker writes metrics.json, predictions.csv, model.joblib, sklearn_report.md, run_manifest.json, and sklearn_pipeline_summary.json.

Change One Thing

Change regularization_c from 1.0 to 0.5, then rerun the app. The manifest and metrics artifacts should make the changed behavior auditable.

Scope

This is a reproducible scikit-learn app example. It is not a generic apps-page, production model registry, or serving stack. Sklearn-specific logic stays inside the app project; shared pages should only consume app-agnostic artifact contracts such as metrics, predictions, and manifests.