A small MLP written from scratch in NumPy to predict pass/fail from study hours. Built to actually understand how a network works, not just to call a library.
Got bored and wanted to actually understand how a neural network works from first principles rather than calling a library function and trusting it.
Solo. Ongoing side project alongside Block 4 of ISE.
Just two files, README.md and main.py. Implements the core loop from scratch in NumPy: forward propagation (Z = Wx + b then sigmoid), loss calculation, backpropagation, and gradient descent. Trained on a small synthetic array of hours studied (1 to 5) mapped to fail/pass.
The point is to watch the model learn the pass/fail threshold on data small enough that I can sanity-check every step by hand.
Work in progress. Forward pass complete. Keeping it that way rather than rushing it to look finished.