AI Classifier Template

Production-minded, local-first Python template for building customizable classification projects.

This repository is designed for developers who want a clean baseline that feels like an internal engineering starter kit: stable abstractions, reproducible training pipeline, clear documentation, quality gates, and straightforward extension points.

Why this template exists

Most classifier repos either overfit to one model stack or under-document architecture decisions. This template balances both sides:

• Local-first execution with no hosted dependency requirements.
• Strong extension points for project-specific classifiers.
• Professional documentation structure suited for public OSS and team handoffs.
• Reproducible artifacts and CLI workflows for daily development.
• Optional deep-learning path without forcing heavyweight dependencies in core installs.

Core Features

• Extensible classifier interface via base contracts.
• Built-in template classifier (TemplateCentroidClassifier) for fast customization.
• Optional sklearn adapter (model.kind: sklearn) for common production baselines.
• YAML-driven pipeline config for train/evaluate reproducibility.
• CLI commands for local train/predict/evaluate.
• Starter dataset and pre-trained artifact for immediate demos.
• CI-ready test/lint/type-check setup.

Repository Layout

.
|-- src/ai_classifier/      # Package source
|-- tests/                  # Unit and integration tests
|-- examples/               # Configs, sample data, starter scripts
|-- docs/                   # Architecture, API, troubleshooting, examples docs
|-- .github/workflows/      # CI and release pipelines
|-- README.md               # Public overview and quick start
|-- LEARN.md                # Deep technical handbook
|-- LICENSE.md              # Human-readable license notes
|-- LICENSE                 # Canonical MIT license text

Quick Start

1. Install locally

python -m venv .venv
. .venv/Scripts/activate
pip install -e .[dev]

Optional extras:

pip install -e .[ml]

2. Train using starter config

ai-classifier train --config examples/configs/template_config.yaml

This generates:

• artifacts/model.json
• artifacts/report.json

3. Predict locally

ai-classifier predict --model artifacts/model.json --input examples/data/predict.csv --output artifacts/predictions.csv

4. Evaluate on labeled data

ai-classifier evaluate --model artifacts/model.json --input examples/data/train.csv --label-column label --output artifacts/eval.json

Local and Customizable by Design

You can create a new classifier with minimal work:

1. Implement BaseClassifier methods (fit, predict, to_artifact, from_artifact).
2. Add your classifier module under src/ai_classifier/classifiers/.
3. Extend model factory logic in pipeline.
4. Add tests and a docs snippet in LEARN.md.

The template classifier exists as a readable baseline so teams can fork and evolve without hidden magic.

Documentation Map

• README.md: overview + quick start.
• LEARN.md: architecture and extension handbook.
• docs/ARCHITECTURE.md: module boundaries and design rationale.
• docs/API.md: public contracts and expected behavior.
• docs/TROUBLESHOOTING.md: common failures and fixes.
• docs/EXAMPLES.md: practical usage patterns.

Quality and Release

Recommended local checks:

pytest
ruff check .
mypy src
python -m build

GitHub workflows are included for CI validation and package publishing.

Public Repository Notes

• Maintainer: zaber-dev
• Suggested repository name: ai-classifier-template
• Versioning model: Semantic Versioning (MAJOR.MINOR.PATCH)

License

MIT License. See LICENSE and LICENSE.md.