Finite Diff

Finite Diff#

Finite-Difference Optimizer (Central Difference)#

Derivative-free gradient estimation by central differences. Suitable for low- to medium-dimensional parameter vectors and backends without analytic gradients. Cost: 2 evaluations per parameter per step.

Interface:

  • initialize(params) -> state

  • step_params(model, params, context) -> (new_params, state)

class qmlhc.optim.numpy_optim.finite_diff.HCFiniteDiffOptimizer(lr=0.01, eps=0.001, clip=None)[source]#

Bases: object

Central-difference gradient descent with optional clipping.

initialize(params)[source]#

Optionally initialize optimizer state (none needed).

Return type:

Dict[str, Any]

step_params(model, params, context)[source]#
Return type:

Tuple[Dict[str, Any], Dict[str, Any]]

Contents