methods: Optimization Algorithms

This module provides various algorithms to optimize the sensor placements or paths. Use the get_methods method to retrieve an optimization method class by its string name.

• ContinuousSGP: This method directly optimizes the inducing points of the AugmentedSGPR model to maximize the Evidence Lower Bound (ELBO). This is the main SGP-based optimization approach proposed in the papers associated with this library.

• GreedySGP and GreedyObjective: These implement greedy algorithms for sensor placement. GreedySGP iteratively selects inducing points to maximize the SGP's ELBO, while GreedyObjective uses a more general objective function like Mutual Information.

• BayesianOpt: This method uses Bayesian Optimization, a powerful black-box optimization algorithm, to find the best sensor locations by maximizing a general objective function.

• CMA: This method uses Covariance Matrix Adaptation Evolution Strategy (CMA-ES), a powerful black-box optimization algorithm, to find the best sensor locations by maximizing a general objective function.

• DifferentiableObjective: This method leverages TensorFlow's automatic differentiation to directly optimize the objective function with respect to the sensor locations. This can be more efficient than black-box methods for smooth objective functions. However, the method is also more prone to getting stuck in local minima.

sgptools.methods.get_method(method)

Retrieve an optimization method class by name.

Parameters:

Name	Type	Description	Default
method	str	Name of the optimization method. Must be one of the keys in :data:METHODS, e.g. 'ContinuousSGP', 'CMA', 'BayesianOpt', 'GreedyObjective', etc.	required

Returns:

Type	Description
Type[Method]	Type[Method]: The corresponding method class.

Raises:

Type	Description
KeyError	If method is not a valid key in :data:METHODS.

Usage

ContinuousSGPClass = get_method('ContinuousSGP')
csgp = ContinuousSGPClass(
    num_sensing=10,
    X_objective=X_train,
    kernel=kernel_opt,
    noise_variance=noise_var_opt,
)

Source code in sgptools/methods.py

def get_method(method: str) -> Type[Method]:
    """Retrieve an optimization method class by name.

    Args:
        method (str):
            Name of the optimization method. Must be one of the keys in
            :data:`METHODS`, e.g. `'ContinuousSGP'`, `'CMA'`, `'BayesianOpt'`,
            `'GreedyObjective'`, etc.

    Returns:
        Type[Method]:
            The corresponding method class.

    Raises:
        KeyError:
            If `method` is not a valid key in :data:`METHODS`.

    Usage:
        ```python
        ContinuousSGPClass = get_method('ContinuousSGP')
        csgp = ContinuousSGPClass(
            num_sensing=10,
            X_objective=X_train,
            kernel=kernel_opt,
            noise_variance=noise_var_opt,
        )
        ```
    """
    if method not in METHODS:
        raise KeyError(f"Method '{method}' not found. Available methods: {', '.join(METHODS.keys())}")
    return METHODS[method]