liveBook · Manning

Poisson bootstrap

As explained in Machine Learning System Design

A bootstrap approximation that independently assigns Poisson-distributed multiplicities to observations for each resampling iteration.

Poisson bootstrap approximates repeated resampling by independently assigning Poisson-distributed multiplicities to observations in each iteration. section 7.4

Combined with noise addition, it helps estimate how precision and recall vary, exposing narrow decision-boundary margins and weak robustness. section 7.4

Confidence bounds for recall

Bootstrapped datasets produce distributions from which lower confidence bounds can be calculated, such as recall at a given precision or specificity. section 7.4

Distribution recall with every point being a bootstrapped original dataset; the red line is a 99% lower confidence bound.
sidebar figure

A plotted recall distribution marks a 99% lower confidence bound, making the reliability of the estimate visible. section 7.4

How Machine Learning System Design covers it

How deep it goes
Explains Poisson resampling through performance distributions and confidence bounds, without deriving the procedure mathematically.
Examples
Uses bootstrapped precision and recall distributions, including noisy and noiseless samples and a 99% recall bound.
What you can do after
Evaluate model performance using lower confidence bounds for recall at specified precision or specificity levels.
How it connects
Connects resampling with precision, recall, specificity, noise addition, confidence intervals, and decision-boundary robustness.
Cover of Machine Learning System Design
liveBookmanning.com
Read Machine Learning System Design on liveBook Get the book at manning.com

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Machine Learning System Design. The section links point to the parts of the book it draws on.