Defusing Deterioration Model Outputs

A quick look at some of the tools in Cassandra for de-bugging complex model outputs and finding anomalies in your model's assigned treatments

Juno Cassandra Desktop comes with a comprehensive set of tools for post-processing model forecasts and treatment data. Shown below is a screenshot of the tools under the Post-Processing menu:

As you can see from the above image, Cassandra’s post-processing toolset contains a comprehensive set of tools to quickly view your deterioration model outputs. In time, I hope to discuss each of these tools in detail, but in this post I want to focus on the tools that we have included specifically to “defuse” the modelling outputs.

I use the words “defuse” because I have often felt that, under the stress-and-strain of a tight deadline, assessing whether the treatments assigned by the model are reasonable and without glaring anomalies can feel like defusing a time-bomb!

We have therefore put careful thought into the design of our post-processing tools in Cassandra, specifically to help the modelling analyst quickly determine when a model has drifted off target.

Below is a short checklist of things that we feel a modeller should check before finalising a Forward Works Programme. Cassandra has specific tools to answer each of these questions with just a few button clicks. Feel free to get in touch with suggestions based on your own experience.

Looking at each of the questions below, the modeller can then assess whether the answers are reasonable or whether some adjustment is needed:

1. How many elements are there that are not being treated at all over the modelling period?

2. How many elements are there with more than X treatments over the modelling period - and which elements are these?

   

3. How many elements are there with less than X treatments over the modelling period - and which elements are these?

4. What percentage of the available Budget is utilised by the model in each year and for each budget category.

Another tool we found really useful to debug a model is the Pre-Treatment Statistics. This tool shows the statistics for a chosen condition parameter in the period just before a treatment is applied, grouped by treatment type or category. Using these statistics, the modeller can assess whether treatments triggered by the model are indeed “the right treatment at the right time”.

Finally - where are we with the release of Juno Cassandra? We are at the moment still working on some tutorials and documentation. As soon as we are ready with that, we will provide more information and release an upgrade for the intrepid modellers among the readers of this blog that want to roll up their sleeves and try working with Cassandra.

Have a great weekend!