LINEAR & NONLINEARCORE
Categorical & Threshold Models
Buckets and cutoffs turn messy attributes into testable if-then rules.
How it works
- Chop messy attributes into buckets with explicit cutoffs.
- Fit separate responses per bucket instead of one grand line.
- Validate cutoffs out-of-sample; buckets love to overfit.
Use → Segmentation, eligibility, triage. Watch boundary gaming.
Watch out → Thresholds get gamed the moment rewards attach to them.
Try it → Bucket customers three ways and compare bucket means to the grand mean.
REDCAPE jobs → Reason · Predict · Communicate
Free sources → Coursera M3 (Categorical & Linear)
Attach to a lattice — combines with
Different families, shared jobs. Disagreement between them is where insight lives.
DISTRIBUTIONS
Normal Distribution
Bell-curve averaging: many small independent effects sum to a predictable center with thin tails.
SHARED JOBS: REASON · PREDICT · COMMUNICATE
DIFFUSIONBroadcast Model
One source reaches a fixed pool at a steady rate: fast early, then fading returns — the r-shaped curve.
SHARED JOBS: PREDICT · COMMUNICATE
LEARNINGDiversity Prediction Theorem
Crowd error equals average error minus diversity. Diverse models beat the lone genius.
SHARED JOBS: PREDICT · COMMUNICATE