In the bar graph, you used benchmark_std as error bars. These are calculated as the standard deviation of sample means, which is incorrect.
Instead, they should be the (symmetric) bounds such that ex. 67% or 95% of the means are inside that bound, depending on which error bars you want.
In the bar graph, you used benchmark_std as error bars. These are calculated as the standard deviation of sample means, which is incorrect.
Instead, they should be the (symmetric) bounds such that ex. 67% or 95% of the means are inside that bound, depending on which error bars you want.