Re:How to Pass a Programming Interview
by aidenn0 on Tuesday March 08, 2016 @01:50PM
(#11247501)
I have more than once in my professional career run into programmers who tested on small inputs and assumed the timing would scale at least close to linearly.
[ Parent
| on HN ]
Fair point. But is it not much more common to encounter a programer who cannot break a complex system into loosely coupled components? Or who does not use consistent naming conventions? Or who is smart but lazy? I'd rate all of these as equally important things to try to evaluate in an interview.
I am a strong believer in looking for strength in an interview. Someone really rocking complexity analysis is a strong positive (shows that they are smart and pedantic in a good way). But so does clean, smart code and great loose coupling.
Oh, I'd never disqualify an interviewee purely for not knowing this stuff, but it is useful to know.
[edit]
Also, I care far less about whether they can solve some problem on paper about asymptotic complexity than if they have some sense of what it means. Someone with little formal training who discovered the classic python "Add to the end of string" algorithm is N^2 and figured out a working knowledge of N^2 vs. N is better than someone who memorized a bunch of math but can't apply it because it went in the "math box"[1]
1: http://zenoferox.blogspot.com/2009/10/deep-inside-math-box.h...
There are a lot of situations where a "worse" algorithm will be significantly faster that another algorithm that's faster in theory, due to memory locality. In practice, it is very hard to know beforehand what parts of your program will scale and what parts won't.
This is categorically untrue if you are talking about large numbers and asymptotic complexity. It is true that algorithms with a constant factor larger number of operations may have such properties, but O(n) will always beat O(n^2) eventually, and in nearly all real-world cases at a fairly small n (1000 L1 cache accesses is slower than 1 memory access so n=1000 will be enough to counter any locality issues).