Bootstrapping

Suppose you want to calculate a quantity \(f(X)\) on your data frame \(X\). Bootstrapping samples multiple versions \(Y_i\) of \(X\) by drawing elements with replacement from the data frame with the same length the data frame itself. On all these \(Y_i\), the function \(f\) is evaluated, creating a distribution of possible values for \(f(X)\). The standard deviation of this distribution is the (symmetric) uncertainty returned by uncertain_panda. If you request the asymmetric uncertainty, the 1 sigma quantile in both directions around the median is returned. You can find some more information on bootstrapping in the net, e.g. on wikipedia.