examples

Bicyclus minimal working example

This MWE will show the basics on how Bicyclus can be used. We consider a very simple fuel cycle with a uranium source, an enrichment facility, and two repositories for enriched and depleted uranium, respectively.

The goal in this scenario is to reconstruct the enrichment grade of the feed uranium (the feed assay) through measurement of the total depleted uranium mass. For simplicity, all other parameters are kept constant.

File overview

The different files used are

• run.py: Driver file, where the specific behaviour is defined (what to extract from each simulation, how to calculate likelihoods, ...)
• cyclus_input.json: Base Cyclus input file. This could also be an .xml or .py file.
• start_reconstruction.sh: Sample script to start an inference process.
• parameters/
  • true_parameters.json: The 'true' parameters, i.e., the ones used to generate the ground truth.
  • sampled_parameters.json: The prior distributions of the sampled parameters.
• plots/*.png: The visualised reconstruction results. Below, it is explained how these can be obtained.

How-to

This MWE can be tested by running $ ./start_reconstruction.sh. In this example, we use four chains of 200 samples each, run on 4 cores in parallel, to perform the inference (which takes some time, say, 30 to 90 minutes).

To combine the four chains (i.e., four independent reconstructions) into one final result and to visualise said result, run (from within this directory) $ python3 ../bicyclus/visualize/merge.py NAME_OF_OUTPUTFILE.cdf ¹. We get the following density plot: Note how the prior density (uniform on 0.0065 to 0.01) got restricted to a much narrower distribution located between approximately 0.0067 to 0.0076. As we have fixed the seed, you should be able to reproduce this distribution exactly ².

Footnotes

1. Please note that you may need to install bicyclus using the [plotting] option to ensure that all dependencies are available. ↩

2. Please note that we have not yet been able to test the software on different OS or different machines. We think this does not influence reproducibility but we cannot guarantee it. ↩