Multi-start metrics

If a multi-start run is performed, the Multi-Start Metrics window shows the metrics from each separate optimization run. For example, if eight start points are used, the optimization is performed separately eight times (with eight different starting points) and the objective function value and other details from each run are returned.

The highlighted row shows the solution which is returned to the user interface and used for reporting purposes. This solution is identified based on its convergence behavior, minimal (or preferably zero) penalty cost, objective function value and minimal (or preferably zero) difference between economic (simulated) and optimizer objective function values.

Partial completion

When the Partial Completion option in the Run Settings window is enabled, the optimization may return fewer start points than the specified number. When this happens, an alert appears at the bottom of the Multi-Start Metrics window. A link in the alert allows you to open the Run Settings window to check your configuration. Any runs that were not completed are not shown in the Multi-Start Metrics window.

Understanding multi-start metrics

Running the model from multiple start points is a useful way to ensure that the solution is not a bad local optimum and that the model convergence is stable. Well-performing models will typically return solutions whose objective function values are very similar and that exhibit consistent convergence behavior. Use both the multi-start metrics table and multi-start metrics chart to ensure that optimization behavior is consistent. Look for:

• Economic Objective Function and Optimizer Objective Function: the solution values are similar, and the final solution value (the highlighted row) is from a consistent area of solution (a plateau in the solution results).

• Optimizer Iterations: each run takes a similar number of iterations to converge. No runs reach the maximum iteration count.

• Solution Time: each run takes a similar time to converge. No runs reach the maximum solution time.

Multi-start metrics chart

The Multi-Start Metrics Chart window shows charts of the following metrics from the latest multi-start optimization:

• Economic Objective Function

• Optimizer Iterations

• Optimizer MIP Nodes

• Optimizer Objective Function

• Solution Time

See the table in Solution Metrics for a definition of these metrics.

You can choose the metric to display from the Metric drop-down menu. The value of the metric is on the Y axis of the graph, while the X axis is the Metric Order. Results from each multi-start run are displayed in descending order. Hover the mouse pointer on each data point to display a tooltip with the run number and the exact value of the metric.

These charts are useful to detect large changes in the metrics, and to find out the multi-start run in which they occurred. This allows you to evaluate the optimal number of multi-start points you need.

Zooming and panning. Below the main chart you can see a smaller, stretched-out version of the graph. Click and drag the handles at either end of this smaller graph to zoom the main graph along the X axis. When the main graph is zoomed in, you can click and drag anywhere on it to pan along the X axis.

Displaying the legend. Click the icon to display or hide the graph legend.

Copying the graph. Click the icon to copy an image of the graph to the Clipboard.

Warning: The Multi-Start Metrics are always sorted largest to smallest. Therefore, the order of the points between two different charts will not necessarily be the same.