More Experiments and Analysis – A Look at the New Cloud Features

A host of new features were added to the AnyLogic Cloud recently. Among them, Monte Carlo 2nd Order, Variation, and Variation with Replication experiments. There are also enhancements to help you communicate your models more widely and clearly. Outputs now include Box Plot, 2D histogram, and 3D surface graphs. Let’s take a closer look at these new features.

In Public models find the Consumer Credit model, use search if you don’t see it immediately.

The Consumer Credit model describes the life-cycle of consumer credit applications, from both a branch office and online. The verification process involves three stages: scoring, personal review, and credit rating inspection. By using the results of multiple model runs, we can determine approval rates and find an optimal number of bank employees. In this model we’ll look at the new experiments and graphs.

Open the model and select Staff capacity variation from the sidebar (on a mobile device, you may need to click the sandwich icon in the top left to expand the sidebar). This experiment provides an example of Variation with Replication. It is worth noting that replication is not needed when you are certain your model is not significantly affected by randomness. Choose Variation, and not Variation with Replication, if this is the case when you are modeling. This time, we will use replication because we are unsure of the effect of variables such as the customer arrival rate.

We can vary the inputs and change the number of bank clerks and analysts by setting the range and step change. An experiment is run for every combination of both ranges, this provides the variation. Each combination can also be replicated, meaning the experiment runs several times with different seeds. A seed is a number which is used to determine randomly generated values such as the arrival rate of customers or the processing time of forms for an employee. By running each combination a number of times, you can test the robustness of the model to get greater confidence in the setup.

Set the parameters and the number of replications you would like and then hit Run in the top menu bar. Scrolling down will show you the progress of the experiment and then the output, where you will find more of the new features.

Both the new box plot and 3D surface graphs appear at the bottom of the experiment outputs. The box plots indicate the range of time applications were in the system. The data is split into quartiles, with 25% of the applications in each. The line within the box marks the median, 50% of the applications took less than this time and fall between it and the lower line outside of the box (the lower whisker). Outliers may be marked as points.

The 3D surface graphs allow us to see the interaction of two variables. For example, this scenario shows the number of abandoned applications is lowest when there are four bank clerks and five analysts (see the blue 3D surface graph in the lower left).

Some experiments can have input ranges that do not have clear steps. We can use Monte Carlo 2nd Order experiments to solve these problems. 2nd Order Monte Carlo allows us to randomly set both internal and input parameters according to a probability distribution. Not just randomizing internal parameters of a model, as with a Monte Carlo 1st Order experiment.

To try Monte Carlo 2nd Order, select the Offline ratio variation sensitivity experiment in the sidebar. This experiment allows us to randomly set the ratio of applications coming from people in a bank branch office or online. The ratio of applications is bound by the two values entered in the App ratio range. A random distribution of values is created, one application ratio figure for each iteration. These iterations run for the number of replications specified, each replication having a different seed value that affects the parameters inside the model, for example, the customer arrival rate. Try some different ranges for the application ratio and see how the credit approval rates are affected.

Moving to the Offline applications prevalence MC1 experiment, we can find another of the new features, 2D histograms. This experiment is set to run with a high frequency of offline applications. The 2D histograms appear at the bottom of the outputs.

Our last demonstration with this model is a visualization of it in action. Select Experiment with online applications prevalence from the sidebar.

In the top menu bar click the second button from the left, the Play button. An animation window will appear containing a dynamic graphical representation. By contrast, the Run button simply executes the model and produces the outputs.

While the animation window is open you will see that interactive model elements are now highlighted on mouse-over: the sliders change color as you hover over them. You can see the effect of changing a variable by moving its slider (the ‘Statistics’ tab provides even more detail).

This is not the only new feature that improves model interactivity. Inspection windows for model elements now show basic statistics. For system dynamics elements these windows can also show graphs to illustrate changes over time – you can test this feature with the Insurgency Dynamics model.

This brings the showcase to an end for now. Join us on LinkedIn, Facebook, Google+, or Twitter and get the latest developments. You can also use these services to sign in to the AnyLogic Cloud, where you will then be able to copy and share models, as well as edit the experiment output display. Check out the new cloud features now and see how you can use them!