feat(ml): build Python ML pipeline against FeatureVector and ModelResult contract (#364)

[s6pa1rta3n-lab]

Summary

Implements the Python ML pipeline against the existing FeatureVector and ModelResult contract, closing #364.

Key Components Implemented

• Contract Layer (ml/contract.py): Dynamically extracts NUMERIC_FEATURE_SPECS directly from packages/chenai-mlflow/src/index.ts to preserve a single source of truth across languages. Implements strict validation and missing-value imputation adhering to FeatureVector and ModelResult schemas.
• Horizon Ingest (ml/ingest.py): StellarHorizonClient consuming Stellar Horizon REST API for accounts, operations, payments, and transactions.
• Feature Builder (ml/features.py): Transforms raw Horizon ledger history into contract-compliant FeatureVector payloads.
• Scoring Models (ml/models.py):
  • CreditScoreModel: Gradient Boosting classifier producing calibrated risk scores in [0, 1] with risk labels (low, medium, high) and confidence values.
  • FraudDetectModel: Isolation Forest anomaly detector producing calibrated anomaly risk scores in [0, 1].
  • Computes 32-byte SHA-256 artifact digests matching Soroban BytesN<32> in contracts/model-attestation.
• ModelResult Emitter (ml/emitter.py): Scores vectors, tracks imputedFields, and formats contract-compliant ModelResult payloads and Soroban model-attestation records.
• Sinks (ml/sinks.py): Supports FileDropSink (.jsonl or atomic per-account JSON), SqliteSink (indexed SQLite database), and StdoutSink.
• Pipeline Runner (ml/pipeline.py): CLI and programmatic interface for single-subject Horizon scoring and batch file processing.
• Reproducible Training (ml/train.py): Trains models against synthetic dataset ml/data/sample_dataset.json, emits tamper-evident signed evaluation reports (ml/evaluation_report.py), and writes ml/artifacts/manifest.json.
• Bidirectional Verification (ml/tests/): 49 unit tests covering all modules, plus cross-language tests validating Python outputs in @chenaikit/chenai-mlflow's Node harness and vice versa.
• CI Workflow (.github/workflows/backend.yml): Adds automated Python ML testing job to GitHub Actions CI.

Testing done

• pytest ml/tests: 49 passed.
• pnpm --filter @chenaikit/chenai-mlflow run build and pnpm --filter @chenaikit/chenai-mlflow run test: 53 passed.
• pnpm run test:integration: 11 passed.
• Tested CLI single mode: python3 ml/pipeline.py --mode single --subject-id GACC123 --sink stdout
• Tested CLI batch mode: python3 ml/pipeline.py --mode batch --input-file ml/data/sample_vectors.jsonl --sink sqlite --output ml/output/scores.db
• Ran Node PR checklist validator: checkPullRequest returned ok: true.

Checklist

• Branch named fix-issue-364
• Follows conventional commit format
• Implements strict contract validation against FeatureVector and ModelResult (schema version 1.0.0)
• All unit and integration tests pass locally
• Documentation updated in ml/README.md
• CI workflow updated in .github/workflows/backend.yml

Payout Routing

• EVM (Base/Arbitrum/Polygon/ETH): 0xF46C9F6d70C50BF81ef3588AB523a90a594a2F89
• Stellar: GCL6OXAMLD75BMTINA6EMRUDWK5THQUSHMYNLSNBCJAPZJHNYJTUNIBC

[vercel]

@s6pa1rta3n-lab is attempting to deploy a commit to the Isaac Gellu's projects Team on Vercel.

A member of the Team first needs to authorize it.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}