torch_numopt.algorithms.conjugate_gradient

Non-linear conjugate gradient methods.

These methods combine the gradient with previous search directions to achieve faster convergence than gradient descent, without requiring explicit curvature.

Classes

ConjugateGradient(params[, lr_init, ...])	Non-linear conjugate gradient optimizer with fixed learning rate.
ConjugateGradientLS(params[, lr_init, ...])	Non-linear conjugate gradient with line search.
ConjugateGradientMixin(*args[, cg_method])	Mixin that provides the conjugate gradient direction computation.