From-Scratch Neural Network

From-scratch NumPy implementation of a feedforward neural network, applied to Titanic survival prediction.
This project demonstrates backpropagation coded by hand (no ML frameworks), mini-batch gradient descent with Adam, early stopping, and reproducibility checks.

Highlights:

• Emphasizes the math and statistics behind neural nets.
• From-scratch NN (NumPy) with tested backprop (gradient check).
• No data leakage: split, fit scaler on train, transform val&test.
• Mini-batch, shuffling, early stopping, Adam optimizer.
• Fast demo notebook runs in < 2 minutes on CPU (subset).

Quickstart (fast)

python -m final_project.cli fast-titanic

If Titanic Dataset.csv is not present, the script generates a small synthetic dataset so the pipeline still runs.