home » the workshop » Fruit Classifier
Where it started
NEW!Fruit Classifier
A single-neuron network in pure NumPy that tells limes from lemons by weight and color. Sigmoid, backprop and gradient descent, all by hand.
- access:
- public
- year:
- high school
- built with:
- Python, NumPy
Written in high school, back when OpenAI was still focused on Gym, to understand neural networks without anyone else’s library doing the thinking. It’s one neuron with two inputs, a fruit’s weight and its color (0 for green, 1 for yellow), and one output: lime or lemon.
Everything is done by hand: random weight initialization, a sigmoid forward pass, backpropagation through the sigmoid’s derivative, and gradient descent over 10,000 iterations. From the same era: a from-scratch linear regression that demonstrates gradient descent, and a reworked version of Karpathy’s min-char-rnn.