Prediction Post-Processing
Post-processing transforms raw model outputs into final usable predictions while respecting domain constraints, calibration and operational decisions. The topic is split into focused lessons so definitions, implementation decisions, diagnostics and common failure modes can be learned separately.
Learn the mechanism one decision at a time
Work through the lessons in order if the topic is new. If you already know the basics, open the specific leaf lesson that matches the operation, diagnostic or failure mode you need.
1Definition→
2Mechanism→
3Example→
4Diagnostic→
5Decision
ClassificationThreshold optimisation, probability calibration, top-k rules, class-specific thresholds and abstention/reject options. The important practical question is not only how the technique is defined, but what assumptions it introduces, which data are allowed to influence it, and how its effect should be validated on unseen evidence.
02RegressionInverse target transforms, bias correction, constraint clipping and predictive intervals. The important practical question is not only how the technique is defined, but what assumptions it introduces, which data are allowed to influence it, and how its effect should be validated on unseen evidence.
03ForecastingTrend/seasonality restoration, smoothing, reconciliation across hierarchies and domain constraints. The important practical question is not only how the technique is defined, but what assumptions it introduces, which data are allowed to influence it, and how its effect should be validated on unseen evidence.
04Computer visionConfidence filtering, Intersection-over-Union thresholds and Non-Maximum Suppression turn dense detector outputs into final boxes. The important practical question is not only how the technique is defined, but what assumptions it introduces, which data are allowed to influence it, and how its effect should be validated on unseen evidence.
05Generative/NLPTemperature, top-k/top-p sampling, beam search and repetition controls shape decoded outputs after the model produces logits. The important practical question is not only how the technique is defined, but what assumptions it introduces, which data are allowed to influence it, and how its effect should be validated on unseen evidence.