Faiaz Amin Khan

I research and build systems that span the physical-digital divide. From designing patent-grade acoustic sensing mechanism at Samsung R&D to deploying AI workflows in resource contrained device, my work focuses on decoding human signals to create seamless, intelligent interactions.

Overview

A lean PyTorch MNIST model + training helper with a pytest-backed harness. Provides a ready-to-import MNISTNet and a train() helper so examples and tests run directly from the repo root.

Repo: github.com/faiazamin/lunar_mnist_net

Why?

I wanted a tiny, reproducible MNIST baseline to share patterns for structuring small research repos: clean package exports, testable entry points, and zero-friction setup for new contributors.

Problem

  • Many toy MNIST repos lack tests and import-stable packaging, causing ModuleNotFoundError when running from root.
  • Onboarding should be one command (pytest) without puzzling over PYTHONPATH tweaks.

Approach

  • Packaged the code under src/ with exports that alias cleanly to src.* from repo root.
  • Wrote a single training helper (train()) that returns a model instance for tests/examples.
  • Added pytest coverage to enforce import discipline and catch regressions.

Key Features

  • MNISTNet implementation in src/model.py.
  • train() helper in src/train.py used by tests and examples.
  • Pytest suite under test/ runnable from the repository root.
  • Guidance for fixing common import issues without manual PYTHONPATH edits.

Tech Stack

  • Python 3.8+
  • PyTorch (CPU or CUDA)
  • pytest

My Contributions

  • Authored the MNISTNet model and the training helper.
  • Structured the package so from src.* imports work from repo root.
  • Wrote pytest cases to keep the model/train loop verifiable.
  • Documented setup, troubleshooting, and contributor expectations.

Setup & Run

  • Create/activate a venv, then install: pip install torch torchvision pytest.
  • Run tests: pytest -q from the repo root.
  • Example usage: ```python from src.train import train from src.model import MNISTNet

model = train() print(hasattr(model, “forward”)) ```

Impact

  • Offers a minimal, testable MNIST baseline for quick experimentation or teaching.
  • Demonstrates repo structure patterns that avoid import pain and keep examples reproducible.

Updated: