Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
66 lines (49 loc) · 2.84 KB

CONTRIBUTING.md

File metadata and controls

66 lines (49 loc) · 2.84 KB

Contributing To EduScale

EduScale is a thesis/research monorepo with Android, model training, dataset, benchmarking, and documentation components. Contributions should keep research claims reproducible and avoid committing private or oversized artifacts.

Contribution Scope

Useful contributions include:

  • Android app fixes and TensorFlow Lite integration improvements
  • training, conversion, and evaluation workflow fixes
  • benchmark reproducibility improvements
  • dataset manifest and validation improvements
  • documentation updates that keep commands, paths, and results accurate

Do not commit:

  • raw datasets, rendered frames, or generated videos
  • private classroom materials, personal data, or institution-owned files
  • large checkpoints unless they are intentionally released as model artifacts
  • generated thesis exports such as PDF, DOCX, PPTX, or HTML builds
  • local logs, caches, virtual environments, or build outputs

Development Workflow

  1. Identify the subsystem being changed.
  2. Inspect nearby docs, tests, and scripts before editing.
  3. Make the smallest focused change that solves the issue.
  4. Run the narrowest relevant verification first.
  5. Update README or guide docs when commands, paths, outputs, or claims change.
  6. Run a broader verification pass before submitting the change.

Verification By Area

Area Recommended checks
Python training/evaluation relevant pytest tests, python -m models.training.evaluation --help
Dataset manifests manifest path checks, row-count checks, datasets/scripts/validate_dataset.py
Benchmarks rerun the affected benchmark command and keep summary JSON/CSV outputs when they support a published claim
Android .\gradlew.bat :app:compileDebugKotlin or .\gradlew.bat :app:assembleDebug from android/
Documentation git diff --check, command/path sanity checks

Dataset And Model Policy

EduScale tracks dataset manifests and reproducibility scripts, not the full raw dataset. Dataset sources and redistribution notes are documented in docs/DATASET_SOURCES.md.

The Android-ready TensorFlow Lite exports are tracked in android/app/src/main/assets/ and mirrored publicly on Hugging Face: https://huggingface.co/jimzzzz/EduScale.

Research Claim Policy

When updating benchmark or model-selection claims, cite the exact saved evidence file in benchmarks/results or TRAINING_REPORT.md. Avoid replacing measured results with estimates unless the text clearly marks them as estimates.

Related Docs