two players, a fair coin, 50/50, no edge.
after enough flips, one ends up with almost everything.
why: they bet a % of what they have, not a fixed amount. that breaks "fair."
same reason a model looks great in backtest and fails in production.
https://t.co/CmTTpoUiD9
the idea that made eigenvectors click: most directions get rotated by a transformation. a few don't, they only stretch or shrink.
those are eigenvectors. eigenvalue = amount of stretch.
PCA finds those directions in data and keeps the biggest ones.
https://t.co/YfGiDgl16U