📏 Feature Scaling Bowls: Why One Step Size Needs Standardized Features (Condition Number)
3D loss-bowl tray comparing raw, standardized, min-max and robust scaling for gradient descent on least squares with one shared step size: condition number, contraction rate, divergence limit 2/lambda_max, with five numerical checks.
⚙ Under the hood
3D loss-bowl tray comparing raw, standardized, min-max and robust scaling for gradient descent on least squares with one shared step size: condition number, contraction rate, divergence limit 2/lambda_max, with five numerical checks.
feature scalingstandardizationgradient descentcondition numbermin-max scalerrobust scalerdata science
3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install