Incorrect standard error

[fzakaria]

pyperf/pyperf/_cli.py

Line 290 in bf02a8b

stdev = bench.stdev()

The docs link to https://en.wikipedia.org/wiki/Standard_error which claim it's: standard deviation / square-root(n).

pyperf seems to be only reporting the standard deviation?

Not sure how to use this information along with the t-score to get my confidence intervals.

[vstinner]

Did you try the https://pyperf.readthedocs.io/en/latest/cli.html#stats-cmd command?

The https://pyperf.readthedocs.io/en/latest/cli.html#compare-to-cmd command computes also the t-test:

pyperf determines whether two samples differ significantly using a Student’s two-sample, two-tailed t-test with alpha equals to 0.95.

[fzakaria]

I tried those commands; they do show the t-test value.
My issue I opened was because the documentation claims that the stdev shown is the "Standard Error" (it even links to the wiki page for Standard Error) but is in fact the sampled stdev.

Showing it as +- is a bit incorrect I felt. It probably needs to then be used with the t-score for a 95% confidence interval?
Anyways; if i'm wrong since i'm also poor at statistics that also makes sense.
I was using pyperformance to benchmark some stuff and wanted to highlight something I found confusing.

[vstinner]

Oh, stdev() is the standard deviation: https://docs.python.org/dev/library/statistics.html#statistics.stdev

[fzakaria]

Right and that's being reported as the Standard Error on the API was my read.

[vstinner]

I think that there is a mistake in the link: I wrote #204 to fix the doc.