I have a piece of code that I'm adding a feature onto, and I need to test performance before and after the feature addition. So I used a test instance with a large amount of data and triggered the endpoint with and without my feature (flag on and flag off) and now I have a data set with the elapsed times as follows: (The same endpoint was called 100 times parallely for each flag value):

Flag off Flag on
32412 24568
28550 22996
26574 25395
35558 21346
...... .....

What is the best way to show a comparison between the elapsed times? The purpose is to show there was no regression with the flag on.

If there was a better way to conduct the pexperiment, what would it be?


1 Answer 1


The best solution depends on whether the values in each column are paired or not.

If paired, a scatterplot with a 1:1 line would be best. Depending on the data distribution, you may want to use some transparency for the points or even consider using a hexbin plot instead.

If unpaired, a split violin plot, or overlapping density plots with some transparency would be good options.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.