A single neuron sees 2D points one at a time, outputs y = w·x, and learns by Oja’s rule: w ← w + η·y·(x − y·w).
Tap to rotate the data cloud toward your finger; the neuron re-learns. The same ?seed= always gives the same run.
LITOja’s rule (1982) converges to the first principal component. The page throws if |w| runs away (it must stay below 3).
AMBERThe angle error is measured against the cloud’s generating axis, not its sample covariance.