MNIST Digit Classification (AI 100 Midterm Project)

This repo trains a simple Convolutional Neural Network (CNN) on the MNIST handwritten digit dataset (10 classes: 0-9).

Quick start

1) Create environment + install deps

python -m venv .venv
source .venv/bin/activate  # (Windows: .venv\Scripts\activate)
pip install -r requirements.txt

2) Train

python train.py --epochs 5 --batch-size 128 --lr 0.001

3) Evaluate + artifacts

The script saves:

• runs/<timestamp>/metrics.json
• runs/<timestamp>/confusion_matrix.png
• runs/<timestamp>/training_curves.png
• runs/<timestamp>/sample_predictions.png
• runs/<timestamp>/best_model.pt

Notes

• First run downloads MNIST automatically via torchvision.
• Works on CPU. If you have a GPU + CUDA, PyTorch will use it automatically.

---

Group Members

• Nawaf Alqahtani
• Nasser Al Malki
• Mohammed Ajwah
• Yazzed Alsheri
• Maria Almansour

Group Contributions

Nawaf Alqahtani – Project coordination, repository setup, final integration, model tuning, and overall review.

Nasser Al Malki – CNN model architecture design and explanation of layer choices (convolution, pooling, dropout).

Mohammed Ajwah – Training pipeline implementation, dataset splitting, and hyperparameter experimentation.

Yazzed Alsheri – Model evaluation, confusion matrix generation, and accuracy analysis.

Maria Almansour – Report writing, formatting, references, and presentation of results.