pdft.optimizers.optimize

pdft.optimizers.optimize(opt, tensors, loss_fn, grad_fn, *, max_iter=100, tol=1e-06, record_loss=False, frozen_indices=None)

Mirror of upstream src/optimizers.jl:335-412.

Returns (final_tensors, loss_history). loss_history is empty unless record_loss=True; then it starts with the initial loss and appends one entry per iteration.

Parameters:

• opt (RiemannianGD | RiemannianAdam)

• tensors (list[Array])

• loss_fn (Callable[[list[Array]], Array])

• grad_fn (Callable[[list[Array]], list[Array]])

• max_iter (int)

• tol (float)

• record_loss (bool)

• frozen_indices (frozenset[int] | None)

Return type:

tuple[list[Array], list[float]]