all projects
from scratch · NumPy

Custom MLP Library

A minimal neural-network framework written from scratch in NumPy, with multiple activation functions and MSE loss, supporting classification and regression.

Result Forward + backprop implemented by hand to understand the gradients end-to-end Stars ★ 1 Tags NumPy · Backprop · MLP · Python
View on GitHub

A neural network with no autograd and no framework — just NumPy and the chain rule.

What it does

  • A configurable multilayer perceptron supporting classification and regression.
  • Multiple activation functions and MSE loss, with forward and backward passes implemented by hand.
  • Built to make the gradients concrete: every weight update is traceable to the math, which is the best way to really understand training.