Main Content

resume

Resume a Bayesian optimization

Description

example

newresults = resume(results,Name,Value) resumes the optimization that produced results with additional options specified by one or more name-value arguments.

Examples

collapse all

This example shows how to resume a Bayesian optimization. The optimization is for a deterministic function known as Rosenbrock's function, which is a well-known test case for nonlinear optimization. The function has a global minimum value of 0 at the point [1,1].

Create two real variables bounded by -5 and 5.

x1 = optimizableVariable("x1",[-5,5]);
x2 = optimizableVariable("x2",[-5,5]);
vars = [x1,x2];

Define an objective function named rosenbrocks that corresponds to Rosenbrock's function.

type rosenbrocks.m % Display contents of rosenbrocks.m file
function f = rosenbrocks(x)

f = 100*(x.x2 - x.x1^2)^2 + (1 - x.x1)^2;

Note: If you click the button located in the upper-right section of this page and open this example in MATLAB®, then MATLAB opens the example folder. This folder includes the objective function file.

Create a function handle for the objective function.

fun = @rosenbrocks;

For reproducibility, set the random seed, and set the acquisition function to "expected-improvement-plus" in the optimization.

rng("default")
results = bayesopt(fun,vars,"Verbose",0, ...
    "AcquisitionFunctionName","expected-improvement-plus");

View the best point found and the best modeled objective.

results.XAtMinObjective
ans=1×2 table
      x1        x2  
    ______    ______

    1.7902    3.2287

results.MinEstimatedObjective
ans = -9.1194

The best point is not very close to the optimum, and the function model is inaccurate. Resume the optimization for 30 more points (a total of 60 points), this time telling the optimizer that the objective function is deterministic.

newresults = resume(results,"IsObjectiveDeterministic",true, ...
    "MaxObjectiveEvaluations",30);

newresults.XAtMinObjective
ans=1×2 table
      x1        x2  
    ______    ______

    1.0757    1.1572

newresults.MinEstimatedObjective
ans = -0.0057

This time, the best point is closer to the true optimum, and the objective function model is closer to the true function.

Input Arguments

collapse all

Bayesian optimization results, specified as a BayesianOptimization object.

Name-Value Arguments

Specify optional pairs of arguments as Name1=Value1,...,NameN=ValueN, where Name is the argument name and Value is the corresponding value. Name-value arguments must appear after other arguments, but the order of the pairs does not matter.

Before R2021a, use commas to separate each name and value, and enclose Name in quotes.

Example: resume(results,'MaxObjectiveEvaluations',60)

You can use any name-value argument accepted by bayesopt except for those beginning with Initial. See the bayesopt Input Arguments.

Note

The MaxObjectiveEvaluations and MaxTime name-value arguments mean additional time or evaluations, above the numbers stored in results. So, for example, the default number of evaluations is 30 in addition to the original specification.

Additionally, you can use the following name-value argument.

Modify variable, specified as an OptimizableVariable object.

You can change only the following properties of a variable in an optimization.

  • Range of real or integer variables. For example,

    xvar = optimizableVariable('x',[-10,10]);
    % Modify the range:
    xvar.Range = [1,5];
  • Type between 'integer' and 'real'. For example,

    xvar.Type = 'integer';
  • Transform of real or integer variables between 'log' and 'none'. For example,

    xvar.Transform = 'log';

Output Arguments

collapse all

Optimization results, returned as a BayesianOptimization object.

Version History

Introduced in R2016b