Interpreting Results

It may be useful to interpret the results of an Optimization.

This page discusses:

Result File

You can find result files.

If you want to study the computation results after an optimization has been carried out, check the Save Optimization Data box. If need be, modify the default path. The file generated contains (for each iteration), the value of the objective function as well as the free parameter values (and the constraints values if any.)

The optimization curves are generated from the data provided by this result file.

Optimization Curves

You can optimize curves.

Click Show Curves.. to display the optimization curves. The abscissa represents the iteration numbers and the ordinate represents the objective function value, or the free parameter values. You must click the relevant curve key displayed on the right of the curves to obtain the proper values on the ordinate axis.

The optimization curves are generated from the result file. To display the optimization curves, you must have checked the Save optimization data box.

As far as the evolution of the "Optimized Parameter" is concerned, 2 curves are drawn:

  • Its actual value, and
  • the "Optimization log" which reproduces the evolution of its best value.

Gradient Algorithm

You can find below the gradient algorithm.



Simulated Annealing Algorithm

You can find below the simulated annealing algorithm.



Warnings and Errors

A Knowledge Report appears listing the errors encountered while running the optimization. Clicking a line in the top frame displays the explanation in the Message section.

The optimization process may issue several warnings. One of the most common is the "Unable to restore best solution" warning.

The openness of Design Optimization enables you to insert any kind of objective functions. Some of them might be either non deterministic or dependent on previous actions (case of a model with rules impacting the result of the objective function.)

The consequence for optimization is the disability to restore a previously found configuration. A warning is issued specifying the difference between the restored solution and the previously best configuration. A small difference is usually due to rounding errors: In this case the optimization is valid. Accept the results. A large difference might indicate that there is a real problem.

Causes of errors

This section gives you more information about causes of errors.

Problems

This section gives you information about one problem you can meet.



f(x)=a x2+1

if (x>1) a=1

else a = -1

During the optimization process if x takes a value between 1 and 2, the definition of the problem changes.

Solutions

This section gives you information about its solution.

Check the model

Problems

This section gives you information about another problem you can meet.

Models with sketches that can take unreasonable configurations from which they cannot escape.



Solutions

This section gives you information about its solution.

Always use constraint sketches and use ranges in the optimization.

Problems

This section gives you information about the last problem you can meet.

Analysis models with adapted types of meshing elements that are not adapted (T4 instead of T10.)

Solutions

This section gives you information about its solution.

Use polynomenial interpolation (T10) instead of linear (T4).