Optimization · CMNA Laboratory
Define an objective, its gradient, and a starting point in real R. Then watch the descent path move across the surface your function creates.
Gradient descent repeats x_next = x - h * grad f(x). The laboratory keeps that rule visible while letting the objective be arbitrary R. Step size is where things get interesting: too small crawls; too large can overshoot or explode.
Bring your own landscape
Then play every descent step across a surface sampled by the same R objective.
Preparing R…f, fp, and x0| n | x₁ | x₂ | f(x) | ||∇f|| |
|---|