DigiShape project Calculating by the data completed

The Calculate by Data project of Deltares and Rijkswaterstaat has been successfully completed! Last year, we worked in DigiShape on a project that explored how consulting firms and knowledge institutions can analyse large datasets without having to make the data externally accessible. One way to do this is to turn it around: you bring an algorithm to the data and only share the results with the questioner.

In a test environment, we have successfully enabled this computation with the data, yet we have chosen not to put this new method into production yet. Fedor Baart of Deltares: “The main reason for this is that it is currently easier to trust a person than an algorithm. We already have many procedures to trust people ‘coming from outside’ (agreements, guest workers), but how do you trust an algorithm on your server? It appears to be too early for that now, but we have established some insights and a reference architecture that we want to share with the community.”

Aspects involved in trusting algorithms:

1. Understandability: Is the algorithm understandable? Can we explain the decisions it makes to users? For this, we use a code review process in github. This looks at documentation, structure and testing.
2. Readability: Is the code behind the algorithm well documented and easy to read? For this, we use code standards (in this project black, isort and code quality checks such as sonarcube).
3. Explainability: Can we understand the reasoning behind the algorithm’s outcomes? This is an active topic of research in explainable AI.

Reference architecture

As a result of this project, we can share the reference architecture above. This can be used to enable, with simple existing open source components, computation at the data.

Fedor: “Although it may have been too early for this idea, it is essential to carry out these kinds of experiments at an early stage. It’s the only way to find out if something works, and even if it doesn’t succeed immediately, it gives us valuable insights that help us in future attempts. And: we can always try it again later.”