Skip to main content

Usage

Import​

from quickopt.bayesopt_tpe import bayesopt_tpe

bayesopt_tpe​

Uses Bayesian Optimization with Tree-structured Parzen Estimator (TPE) to minimize a double function.

Signature​

bayesopt_tpe(funct, space_min, space_max, iterations=100, samples=10, acquisition_samples=100, verbose=1, threshold=0.5, group_cap=9999, bandwidth=[], prior_weight=1, split=-107.2931, seed=[])

Parameters​

  • funct - function: (input: List[float]) -> float : The function to be minimized. It should take a list of doubles as input and return a double.

  • space_min - List[float] : Lower bound vector for the search space. Defines the minimum values for the parameters.

  • space_max - List[float] : Upper bound vector for the search space. Defines the maximum values for the parameters.

  • iterations - int, optional : The number of iterations to run the algorithm. Default is 100.

  • samples - int, optional : The number of samples to draw initially. Default is 10.

  • acquisition_samples - int, optional : The number of samples to draw for acquisition function maximization. Default is 100.

  • verbose - int, optional : The verbosity level. 0 for final output, 1 for output at each iteration. Default is 1.

  • threshold - double, optional : Threshold value for the splitting function. Default is 0.5.

  • group_cap - int, optional : Group cap for the good set. Default is 9999.

  • bandwidth - List[float], optional : Optional bandwidth vector for the KDE. If empty, dynamic bandwidth calculation is used.

  • prior_weight - int, optional : Prior weight parameter for Non-Informative Prior. Default is 1.

  • split - double, optional : Optional splitting threshold. Overrides dynamic threshold calculation.

  • seed - List[Tuple[List[float], float]], optional : Seed data for the algorithm.

Output​

  • best_params - List[float]: The best set of parameters found by the Bayesian Optimization process.

Notes​

  • Maximization can be achieved by returning the negative of objective function values.
  • The threshold value must be positive and at most the square root of the number of samples.
  • The split value must be within [0, 1] if provided; otherwise, dynamic calculation is used.
  • The space_min and space_max vectors must be of the same size, with each element of space_min less than the corresponding element of space_max.
  • The bandwidth vector, if provided, must be of the same size as the search space; otherwise, dynamic bandwidth calculation is used.
  • Ensure the number of iterations, samples, and acquisition_samples are positive integers.
  • The seed parameter must be a list of tuples, in which each tuple contains a list of parameters and the corresponding function value.
  • Adjust the verbosity level with the verbose parameter to control the amount of output during the iterations.