Evaluation
Regression
What is regression?
A regression is a case where a new model performs worse than the one it replaces. Aggregate improvement almost always contains some regressions, and the question a gate has to answer is whether they are tolerable, not whether they exist.
Reporting the count and the character of regressions alongside the aggregate is more useful than the aggregate alone. Twenty small regressions spread evenly is a different situation from twenty concentrated in one category.
Concentration is the signal to look for. Regressions clustered in a single category usually mean that category is under-represented in the training data, which is a fixable problem with a clear next step.
Related terms
Start with the free tier
A magic link creates your account, your tenant and your first API key. No card until you ask for compute.