Use standard deviation to have significant analysis about fast-pandas

[mm-mansour]

Well I agree with you that the mean is definitely not the best marker here, at the very least median is better as mentioned in the last section of readme; however, there is a different view to all this. One could argue that the minimum time across all runs is the most representative measure and any other run with a larger time was the result of some noise. Hence, you could take the minimum across all runs or take the average of the fastest 3 runs, which is also a common approach.
Based on this, I don't think that a boxplot is representative of how long the operation is taking. I believe that it would probably just show the distribution of the added noise caused by other processes in the background or some OS operations.