Jitter adds a small random perturbation to generated values, spreading observations around histogram bin centers instead of producing only a few repeated discrete outputs. section 11.2.2
Use jitter when generating synthetic numeric features from empirical distributions:
- Sample histogram bin centers according to their frequencies, then perturb each sampled value by a random amount. section 11.2.2
- Vary the perturbation amount to adjust how realistic the synthetic values appear. section 11.2.2
- Interpret jitter as a feature-level operation: it preserves marginal distributions approximately but does not restore associations between features. section 11.2.2
Baseball features
For the baseball dataset, jittered histogram samples can closely match the real distribution of At_bats while producing continuous values; independently generated features can still lose their relationships with RBIs. section 11.2.2
