nannyml

Why we included this project

Knowing whether a model still works after it goes live is often the trickiest part of MLOps, and NannyML targets that gap directly. It estimates performance when ground-truth labels are delayed or missing, using confidence-based methods, and combines that with univariate and multivariate drift detection so you can trace a silent failure to the specific input features that shifted. For tabular classifiers and regressors, you set a reference baseline from your test set, point the library at live data, and get charts that layer estimated metric trends, drift alerts, and ranked offenders in one view. That makes it a practical pick for data scientists and MLOps engineers who want to move past simple alerting to actually understanding why a model degraded. It stays model-agnostic and runs in plain Python notebooks.