This node provides an efficient hybrid optimization strategy that comes with only one major setting to be tuned, the maximum number of design evaluations. Depending on the type and number of input parameters and the defined optimization criteria, the optimizer automatically selects the most suitable optimization algorithms with their most appropriate settings to solve the optimization problem. The ability to dynamically switch between optimization algorithms and to run multiple algorithms simultaneously makes OCO one of the most reliable and efficient optimization strategies. OCO is a surrogate assisted optimization strategy, using capabilities of the Metamodel of Optimal Prognosis (MOP) for function approximation to significantly speed up the optimization process.
Node Settings
To access the options shown in the following table, double-click the One-Click Optimization system on the Scenery pane and in the Settings tab, click .
| Name | Default Value | Description |
|---|---|---|
| Maximum number of design evaluations | 200 | Maximum number of design evaluations to compute before stopping. |
| Advanced settings | ||
| Use MOP | On | Accelerate convergence and save the number of evaluations using the MOP. Deactivate when dealing with computational inexpensive evaluations. |
| Stop at the first feasible design | False | Stops the optimizer when at least one feasible design has been evaluated. |
| Stop after the given number of design evaluations without improvement | 40 | Stops the optimization when no improvement of the best individual is seen after this number of design evaluations and the minimum number of design evaluations has been generated. |
| Fixed number of design evaluations per iteration | Auto | Defines the number of design evaluations generated per
iteration. When set to Auto (or no
value), the number of design evaluations can vary from one
iteration to another one. The maximum allowable value is
999. |
| Expert settings: Initial designs | ||
| Use start designs only | False | When selected, the first iteration uses only start designs as initial design evaluations. Number of initial design evaluations is deactivated. |
| Number of initial design evaluations | Integer | Defines the number of design evaluations generated for the first iteration. The number of initial design evaluations depends on the number of input parameters and the maximum number of design evaluations. OCO automatically adjusts the number of initial design evaluations when one of these two properties is modified. |
| Expert settings: Convergence test | ||
| Minimum number of design evaluations before the convergence test | 100 | Defines the minimum number of designs to evaluate before stopping the algorithm due to stagnation. |
| Expert settings: MOP | ||
| Activate dimension reduction | On | Filters the most relevant input variables to speed-up the convergence of the optimization process. |
| Dynamic selection | On |
When selected, OCO activates/deactivates the most suitable metamodels according to the number of submitted designs, based on the default pre-selection of metamodels. You can modify the pre-selection by enabling/disabling any metamodel. When cleared, all selected metamodels are used independently of the number of submitted designs. |
| Isotropic Kriging | On | Selects Isotropic Kriging as tested metamodel to create a MOP. |
| Anisotropic Kriging | Off | Selects Anisotropic Kriging as tested metamodel to create a MOP. |
| Moving least squares | On | Selects Moving least squares as tested metamodel to create a MOP. |
| Polynomials | On | Selects Polynomials as tested metamodel to create a MOP. |
| Deep Feed Forward Network | Off | Selects Deep Feed Forward Network as tested metamodel to create a MOP. |
| Deep Infinite Mixture Gaussian Process | Off | Selects Deep Infinite Mixture Gaussian Process as tested metamodel to create a MOP. |
| Genetic Aggregation Response Surface | Off | Selects Genetic Aggregation Response Surface as tested metamodel to create a MOP. |
| Support Vector Regression | Off | Selects Support Vector Regression as tested metamodel to create a MOP. |
| Expert settings: Optimization methods competition | ||
| Maximum number of concurrently competing methods | 2 | Maximum number of allowable optimizations in parallel. |
| Non-linear Programming by Quadratic Lagrangian (NLPQL) | On | Defines NLPQL as the allowable optimization method. |
| Downhill Simplex Method | On | Defines Downhill Simplex Method as the allowable optimization method. |
| Mixed-Integer Sequential Quadratic Programming (MISQP) | On | Defines MISQP as the allowable optimization method. |
| Adaptive Response Surface Method (ARSM) | On | Defines ARSM as the allowable optimization method. |
| Adaptive Metamodel of Optimal Prognosis (AMOP) | Off | Defines AMOP as the allowable optimization method. |
| Evolutionary Algorithm (EA) | On | Defines EA as the allowable optimization method. |
| Particule Swarm Optimization (PSO) | On | Defines PSO as the allowable optimization method. |
| Stochastic Design Improvement (SDI) | On | Defines SDI as the allowable optimization method. |
| Covariance Matrix Adaptation (CMA) | On | Defines CMA as the allowable optimization method. |
Additional Options
This algorithm allows Additional Options.
Limitations
OCO can not support boolean or string parameters and parameters with nominal discrete values.
Logging
This node creates a log file named OCO_logfile.txt. The log file has two levels, default and verbose. The default log captures initial settings at the start of the algorithm, internal activation and deactivation events, current number of evaluations, and so on. The verbose log captures all of the information of the default log and also logs which optimization algorithms were chosen for use.
To change the log level to verbose, create an OCO_VERBOSE_LOG
environment variable.