Created as a hack-week project at Zymergen, Sundry is a unique way to compare the performance of multiple statistical models by focusing on the input data that results in differences in model outputs.
Sundry was inspired and based on Uber's Manifold project and was built prior to Manifold's source code being released. Instead of attempting to look at metrics generated by the models, Sundry splits up the input data into clusters and compares the performance of multiple models across these clusters.
Comparing the input data clusters, users can see how features differed between clusters, providing a quick way to discover underlying reasons for differences in model performance.
Sundry allows users to quickly upload input data and model output with no code or additional infrastructure.