Data Drift
-
Types of Data Drift: Covariate, Label, and Concept
The four dataset-shift types, the shapes each takes over time, the pipeline drifts nobody names, and which monitor actually catches each one.
-
Concept Drift vs Data Drift: Key Differences
Compare concept, data and prediction drift by what changes, whether labels are required, detection latency, monitoring signals, and the right response.
-
Population Stability Index (PSI): Formula and Thresholds
Learn the PSI formula, calculate it with NumPy or Evidently, interpret 0.1 and 0.25 thresholds, choose bins, and avoid common false alarms.
-
Data Drift Detection in ML: Methods, Tests, and Practice
Statistical tests, reference windows, multivariate checks, threshold tuning, and when a data drift alert should actually trigger a model retrain.