Re:Statistics, Fast and Slow
by dbxz on Wednesday May 16, 2018 @07:38PM
(#17087339)
To be a statistician is to ignore experience. I once interviewed someone who was a statistics major. When I asked him a performance optimization question, the first thing he told me is that he'd write a program to compute the standard deviation of running time. He could've solved it in two minutes just by running the thing. No hire.
[ Parent
| on HN ]
So you decided not to hire a statistician based on their intuition to first get a handle on the problem "What is the performance of the program?".
...I'd say he came out ahead in that exchange.
I think it's a pretty good idea actually, because with standard deviations (and means, of course) you can run statistical tests to make sure your improvements are statistically significant and not due to chance. I hope the candidate had a chance to explain themselves before he got no-hired.
The candidate is probably now happily working under a manager who doesn’t complain “why are you using your skills? Just [do the simplest thing that occurs to me], you idiot.”
A lot of managers out there like that unfortunately.
It also tells you how variable your performance is - for example, do you experience large fluctuations in runtime, and if so, why?
>asked him a performance optimization question
This is like asking a performance engineer to optimize the accuracy of a machine learning model, and the engineer fired up gprof. No hire.
Does anybody else find that people boasting about their "hires or no hires" a) come off as showing off, b) most of times are in the wrong end of the decision?
We can comment on why we think a practice is bad -- as in, with arguments. Not about how we ditched some candidate because they followed it.
"Producing Wrong Data Without Doing Anything Obviously Wrong!" http://www.ece.northwestern.edu/~robby/courses/322-2013-spri...