pdft.training.train_basis

pdft.training.train_basis(basis, *, target, loss, optimizer, steps, seed=0, device='cpu')

Train basis to minimize loss(basis.tensors, target) over steps.

Works for any basis registered as a JAX pytree whose leaves begin with the forward-circuit tensor list followed by the inverse-circuit tensor list (current convention for all four bases: QFTBasis, EntangledQFTBasis, TEBDBasis, MERABasis).

Parameters:

• target (Array)

• loss (AbstractLoss)

• optimizer (RiemannianGD | RiemannianAdam)

• steps (int)

• seed (int)

• device (str)

Return type:

TrainingResult