Rigorous, reproducible, extensible, data-driven model validation for science
Django ReactJS DRF celery Redis SciUnit NeuronUnit
A science and technology go hand in hand, encouraging mutual development. Technology results from the practical application of the scientifically gained data, while collecting data is impossible without applying technology for scientific experiments.
Both software developers and neuroscientists actually face the same problems: they must understand the scope of each component of a complex project and verify that each component achieves desired input/output behavior.
So, the professor from the University of Arizona required a platform to manage his libraries for testing the scientific models he was working on.
SciDash project is a tech contribution to the validation of scientific models through experiments.
Too much time is spent on manual model verification. The way to assess the scientific model quality automatically according to certain characteristics is required.
Test Scores table is responsible for the validation of scientific models against experimental data. It helps understand how accurate the model is using the Python framework SciUnit and the extensible NeuronUnit library for neuron testing. The dashboard fetches the test repositories data from GitHub, making it easy to execute unit tests locally or in the cloud.
An area to work with models for testing, add their parameters, tags, and create instances is required.
Models dashboard has been developed from scratch with the following model features:
- source URL;
After a user has added the model URL, its data (from the SciUnit library) gets automatically fetched to the dashboard.
The process of the competitive scientific model testing against a suite of unit tests should be automated.
Tests are available for creation. The dashboard provides an opportunity to select a test name, class, as well as observation values, parameters, descriptions, and tags.
A user should have the ability to create, configure, and save a custom set of tests for a particular model.
Suite Scores is the dashboard built to incorporate and visualize the set of tests for each model. Filtered by either model or suite name, the table covers the average test score, its number, and time.
A user-friendly work area to launch tests is needed.
The Scheduling function allows a user to connect the created model to the test and schedule its verification.
The portal is actually an output table that has been created with a possibility to test different scientific models oneself and visualize the results.
The customer came to us with the specifications that our full-stack developer brought to life.
“SciDash aims to make validation of computational scientific models against experimental data easy, transparent, and continuously integrated into the model development process.”
The non-profit project is of practical use for university professors who work on reproducible execution and visualization of data-driven unit tests for assessing model quality. It is a 21st-century vision of the scientific method.