ScipyOptimizer

class ScipyOptimizer(method)

Bases: Optimizer

Optimizer wrapper for scipy.optimize methods.

Supports gradient-free and gradient-based optimization algorithms from scipy, including Nelder-Mead simplex, COBYLA, and L-BFGS-B.

Initialize a scipy-based optimizer.

Parameters:

method (ScipyMethod) – The optimization algorithm to use.

Attributes Summary

n_param_sets

Get the number of parameter sets used by this optimizer.

Methods Summary

get_config()	Get optimizer configuration for checkpoint reconstruction.
load_state(checkpoint_dir)	Load the optimizer's internal state from a checkpoint directory.
optimize(cost_fn[, initial_params, callback_fn])	Run the scipy optimization algorithm.
reset()	Reset the optimizer's internal state.
save_state(checkpoint_dir)	Save the optimizer's internal state to a checkpoint directory.

Attributes Documentation

n_param_sets

Get the number of parameter sets used by this optimizer.

Returns:

Always returns 1, as scipy optimizers use single-point optimization.

Return type:

int

Methods Documentation

get_config()

Get optimizer configuration for checkpoint reconstruction.

Raises:

NotImplementedError – ScipyOptimizer does not support checkpointing.

Return type:

dict[str, Any]

classmethod load_state(checkpoint_dir)

Load the optimizer’s internal state from a checkpoint directory.

Scipy optimizers do not support loading state as they cannot save state.

Parameters:

checkpoint_dir (Path | str) – Directory path where the optimizer state would be loaded from.

Raises:

NotImplementedError – Always raised, as scipy optimizers cannot load state.

Return type:

ScipyOptimizer

optimize(cost_fn, initial_params=None, callback_fn=None, **kwargs)

Run the scipy optimization algorithm.

Parameters:

• cost_fn (Callable[[ndarray[tuple[Any, ...], dtype[double]]], float]) – Function to minimize. Should accept a 1D array of parameters and return a scalar cost value.

• initial_params (ndarray[tuple[Any, ...], dtype[double]] | None) – Initial parameter values as a 1D or 2D array. If 2D with shape (1, n_params), it will be squeezed to 1D.

• callback_fn (Callable[[OptimizeResult], Any] | None) – Function called after each iteration with an OptimizeResult object. Defaults to None.

• **kwargs –

  Additional keyword arguments:

  • max_iterations (int, optional): Total desired number of iterations. Defaults to None (no limit for some methods).

  • jac (Callable): Gradient function (only used for L-BFGS-B).

Returns:

Optimization result with final parameters and cost value.

Return type:

OptimizeResult

reset()

Reset the optimizer’s internal state.

ScipyOptimizer does not maintain internal state between optimization runs, so this method is a no-op.

Return type:

None

save_state(checkpoint_dir)

Save the optimizer’s internal state to a checkpoint directory.

Scipy optimizers do not support saving state mid-minimization as scipy.optimize does not provide access to the internal optimizer state.

Parameters:

checkpoint_dir (Path | str) – Directory path where the optimizer state would be saved.

Raises:

NotImplementedError – Always raised, as scipy optimizers cannot save state.

Return type:

None