analysis

README

This tool generates CSV files for various statistical analyses of the t-digest's behavior. They are similar to the CSVs generated by Dunning's own tests here. To use:

go build .
./analysis [<rounds> [<samples> [<seed>]]]

This will populate a t-digest with samples samples in it, drawn from a uniform distribution seeded by seed, and then dump statistical information about the t-digest to various CSV files. It will repeat this process rounds times for the uniform distribution, and then another rounds times for a normal distribution, and then another rounds times for an exponential distribution. The default values will run 10 rounds with 100,000 samples each, using a seed from the current time. This behavior is equivalent to Dunning's original tests.

An R script has been included to draw various graphs from the data. It depends on data.table and ggplot2. Note that, depending on the size of the ata, it can take upwards of 10 minutes for this script to finish.