DigitScout

MNIST handwritten digit classifier using Multi-Layer Perceptron (MLP).

Preprocessing: normalize [0,1], flatten to 784 features, with a 80/20 train/val split.

Baseline: Logistic Regression achieves 92% accuracy in ~5s.

Run

pip install -r requirements.txt
python explore.py    # visualize data
python baseline.py   # train baseline model

Author: Alex Waisman
License: MIT