What is the bootstrapping for?
- A way to generated simulated datasets, randomly selection data and allowing for duplicates. (Sampling with replacement)
- A resampling method to estimate parametersinstead of repeating experiment many times for a reasonable distribution.
- data are non-normal
- unknown statistical properties (e.g. PCA )
- lack a standard calculation
bootstrapping steps
Application Scenario: only a single sample is available
preparaion: an exist/raw dataset
Step1: make a bootstrapped dataset (randomly select the same number of data from the existing dataset)
Step2: calculate statistics(e.g. mean)
Step3: keep track of the calculated statistics
Step4: repeat step1-3 a bunch of times
bootstrapping procedura can maintains data structure but reshuffles values; can sample each value multiple times, or not at all
