Thursday, July 25, 2013
More on the Programming Tests Saga

A couple of people had asked how the story with the company that triggered the "I Hate Programming Tests" post ended, so I figured I'd follow up with the rest of that story, and some thoughts.

After handing in the disjoint-set solution I'd come up with, the VP pondered things for a bit, then decided to bring me in for an in-person interview loop with a half-dozen of the others that work there. I said I'd be happy to, and came in, did a brief meet-and-greet with the group of folks I'd be interviewing with (plus, I think, a few others), and then we got to the first interview mono-a-mono, and after a brief "Are you familiar with MVC?", we get into...

... another algorithm challenge. A walk-up-to-the-whiteboard-and-code-this challenge.

OK, whatever. I already said I'm not great with algorithmic challenges like this, but maybe this guy didn't get the memo or he's just trying to see how I reason things through. So, sure, let's attack this, even though I haven't done this kind of problem in like twenty years. (One of the challenges was "How do you sort a file of integer numbers when you can't store the entire collection of numbers in memory?", which wasn't an unfair challenge, just not something that I generally have to mess with. Honestly, in the working world, I'll start by going through the file number by number--or do chunks of the file in parallel using actors, if the file is large enough--and shove them into a database that's indexed on that number. But, of course, as with all of these kinds of challenges, the interviewer continues to throw constraints at the problem until we either get to the solution he wants or Ted runs out of imagination; in this case, I think it was the latter.) End result: not a positive win.

Next interviewer walks in, he wasn't there for the meet-and-greet, which means he has even less context about me than the guy before me, and he immediately asks... another algorithmic challenge. "If you have a tree of nodes, and you want to get a list of the nodes in rank order" (meaning, a breadth-first search, where each node now gets a "sibling" pointer pointing to the sibling on its right in the tree, or null if it's the rightmost node at that depth level) "how would you do it?" Again, a fail, and now I'm getting annoyed. I admitted, from the outset, that this is not the kind of stuff I'm good at. We've already made that point. I accept the "F" on that part of my report card. What's even more annoying, the interviewer keeps sighing and drumming his fingers in an obvious state of "Why is this bozo wasting my time like this, I could be doing something vastly more important" and so on, which, gotta say, was kind of distracting. End result: total fail.

By this point, I'm really annoyed. The VP comes to meet me, asks how it's going, and I tell him, flatly, "Sucks." He nods, says, yeah, we're going to kill the interview loop early, but I want to talk to you over lunch (with another employee along for company) and then have you meet with one more person before we end the exercise.

Lunch goes quite well, actually, and the last interview of the day is with their Product Manager, who then presents me with a challenge: "Suppose I want to build an online system for ordering pizzas. Customers can order pizzas, in other words. Build for me either the UI or the data model for this system." OK, this is different. I choose the data model, and build a ridiculously simple one-to-many relationship of customers to orders, and a similar one-to-many for orders to pizzas. She then proceeds to complicate the model step by step, sometimes in response to my questions, sometimes out of the blue, until we have a fairly complex roughly-sketched data model on the whiteboard. Result: win.

The VP at this point is on the horns of a dilemma: two of the engineers in the interview loop are convinced I'm an idiot. They're clearly voting no on this. But he's read my articles, he's seen some of my presentations, he knows I'm not the idiot the others assume me to be, and he's now trying to figure out what his next steps are. He takes a week to think about it, then emails me yesterday to say that it's not going to work.

Here's my thoughts, and folks, if you interview people or are part of an interview process, I'm trying to generalize this beyond this one experience to take it into a larger context:

  • Know what you want to prove with your interview. I get the feeling that this interview loop was essentially a repeat of every interview loop they've ever done before, with no consideration to the candidate himself. An interview is a chance for the company to get to know the candidate better, in order to make a well-informed decision. In this particular case, trying to suss out my skills around algorithms was a wasted effort--I'd already conceded that point. Therefore, find new questions! Find new areas in which to challenge the candidate to see what their skills are. (If you can't think of something else to ask, then you're not really thinking about the interview all that hard, and you're just going through the motions.)
  • Look for the proof you seek in other areas. With the growth of things like Github and open source projects in general, it's becoming easier and easier to prove to yourself as a company that a candidate does or does not have the coding skills you're looking for. Did this guy submit some pull requests to a project? Did this guy post some blogs about interesting technical tidbits? (Or, Lord help us, write articles for major publications?) Did this guy author an open-source project, or work on a project that other people know about? Look at it this way: If Anders Heljsberg, Bjarne Stroustrup or James Gosling walk through the door, are you going to put them through the same interview questions you put the random recruiter-found candidate goes through? Or are you willing to consider their established body of work and call it covered? As an interviewer, it behooves you to look for that established body of work, so that you can spend the interview loop looking at other things.
  • Be clear in what you want. One of the things the VP said to me was that he was looking for somebody who had a similar skillset to what he had; that is, had a architectural view of things and an interest in managing the people involved. By then submitting my candidacy to a series of tests that didn't really test for those things, he essentially torpedoed whatever chances it might have had.
  • Be willing to assert your authority. If you're the VP of the company, and the people who work for you disagree with your decisions, sometimes the Right Thing To Do is to simply overrule them. Yes, I know, it's not all politically correct to do that, and if you do it too often you'll ruin whatever sense of empowerment that you want your employees to have within the company, but there are times when you just need to assert that authority and say, "You know what? I appreciate y'all's input, but this is one of those cases where I think I have a different enough perspective that I am going to just overrule and do it anyway." Sometimes you'll be right, yay, and sometimes you'll be wrong, boo, but there is a reason you're the VP or the Director or the Team Lead, and not the others. Leadership means making hard decisions sometimes.
  • Be willing to change up the process. So your candidate comes in, and they're a junior programmer who's just graduated college, with zero experience. Do you then start asking them questions about their experience? That would be a waste of their time and yours. So you'll have to come up with new questions and a new approach. Not all interviews have to be carbon copies of each other, because certainly your candidates aren't carbon copies of each other. (At least, you'd better hope not, or else you're going to end up with a pretty single-dimensional staff.) If they've proven their strength in some category, or admitted a lack in another, then drop your standard set of questions, and go to something different. There is no honor in asking the exact same questions of every candidate.
  • Be willing to hire somebody that offers complementary skills. If your company already has a couple of engineers who know algorithms really well, then hire somebody for a different skillset. Likewise, if your company already has a couple of people who are really good with customers, you don't need another one. Look for people that have skills that fall outside the realm of what you currently have, and trust that when that individual is presented with a problem that attacks their weakness, they'll turn to somebody else in the firm to help them with it. When presented with an algorithmic challenge, you're damn well sure that I'm going to turn to somebody next to me and say, "Hey, dude? Help me walk through this for a bit, would you?" And, in turn, if that engineer has to give a presentation to a customer, and they turn to me and say, "Hey, dude? Help me work on this presentation, would you?", I'm absolutely ready to chip in. That's how teams are built. That's why we have teams in the first place.
In the end, this is probably the best of all possible scenarios, not working for them, particularly since I have some other things brewing that will likely consume all of my attention in the coming months, but there's that part of me that hates the fact that I failed at this. That same part of me is now going back through a few of the "interview challenges" books that I picked up, ironically, for my eldest son when he goes out and does his programming interviews, just to work through a few of the problems because I HATE feeling inadequate to a challenge.

And that, in turn, raises my next challenge: I want to create a website, just a static thing, that has a series of questions that, I think, are far better coding challenges than the ones I was given. I don't know when or if I'm going to get to this, but I gotta believe that any of the problems out of the book "Programming Challenges" (by Skiena and Revilla, Springer-Verlag, 2003) or the website from which those challenges were drawn, would be a much better test of the candidate's ability, particularly if you look at the ancillary parts of the challenge: do they write tests, how do they write their tests, do they pair well with somebody, and so on. THOSE are the things you really care about, not how well they remember their college lessons, which are easily accessible over Google or StackOverflow.

Bottom line: Your time is precious, people. Interview well, or just don't bother.

.NET | Android | C# | C++ | Conferences | Development Processes | F# | Industry | iPhone | Java/J2EE | Languages | Mac OS | Personal | Ruby | Scala

Thursday, July 25, 2013 1:19:48 PM (Pacific Standard Time, UTC-08:00)
Comments [2]  | 
 Tuesday, July 9, 2013
Programming Tests

It's official: I hate them.

Don't get me wrong, I understand their use and the reasons why potential employers give them out. There's enough programmers in the world who aren't really skilled enough for the job (whatever that job may be) that it becomes necessary to offer some kind of litmus test that a potential job-seeker must pass. I get that.

And it's not like all the programming tests in the world are created equal: some are pretty useful ways to demonstrate basic programming facilities, a la the FizzBuzz problem. Or some of the projects I've seen done, a la the "Robot on Mars" problem that ThoughtWorks handed out to candidates (a robot lands on Mars, which happens to be a cartesian grid; assuming that we hand the robot these instructions, such as LFFFRFFFRRFFF, where "L" is a "turn 90 degrees left", "R" is a "turn 90 degrees right", and "F" is "go forward one space, please write control code for the robot such that it ends up at the appropriate-and-correct destination, and include unit tests), are good indicators of how a candidate could/would handle a small project entirely on his/her own.

But the ones where the challenge is to implement some algorithmic doodad or other? *shudder*.

For example, one I just took recently asks candidates to calculate the "disjoint sets" of a collection of sets; in other words, given sets of { 1, 2, 3 }, { 1, 2, 4 } and { 1, 2, 5 }, the result should be sets of {1,2},{3},{4}, and {5}. Do this and calculate the big-O notation for your solution in terms of time and of space/memory.

I hate to say this, but in twenty years of programming, I've never had to do this. Granted, I see the usefulness of it, and granted, it's something that, given large enough sets and large enough numbers of sets, will make a significant difference that it bears examination, but honestly, in times past when I've been confronted with this problem, I'm usually the first to ask somebody next to me how best to think about this, and start sounding out some ideas with them before writing any bit of code. Unit tests to test input and its expected responses are next. Then I start looking for the easy cases to verify before I start attacking the algorithm in its entirety, usually with liberal help from Google and StackOverflow.

But in a programming test, you're doing this alone (which already takes away a significant part of my approach, because being an "external processor", I think by talking out loud), and if it's timed (such as this one was), you're tempted to take a shortcut and forgo some of the setup (which I did) in order to maximize the time spent hacking, and when you end up down a wrong path (such as I did), you have nothing to fall back on.

Granted, I screwed up, in that I should've stuck to my process and simply said, "Here's how far I got in the hour". But when you've been writing code for twenty years, across three major platforms, for dozens of Fortune 500 companies and architected platforms that others will use to build software and services for thousands of users and customers, you feel like you should be able to hack something like this out fairly quickly.

And when you can't, you feel like a failure.

I hate programming tests.

Update: By the way, as always, I would love some suggestions on how to accomplish the disjoint-set problem. I kept thinking I was close, but was missing one key element. I particularly would LOVE a nudge in doing it in a object-functional language, like F# or Scala (I've only attempted it in C# so far). Just a nudge, though--I want to work through it myself, so I learn.

Postscript An analogy hit me shortly after posting this: it's almost as if, in order to test a master carpenter's skill at carpentry, you ask him to build a hammer. After all, if he's all that good, he should be able to do something as simple as affix a metal head to a wooden shaft and have the result be a superior device to anything he could buy off the shelf, right?

Further update: After writing this, I took a break, had some dinner, played a game of Magic: The Gathering with my wife and kids (I won, but I can't be certain they didn't let me win, since they knew I was grumpy about not getting this test done in time), and then came back to it. I built up a series of little steps, backed by unit tests to make sure I was stepping through my attempts at reasoning out the algorithm correctly, backed up once or twice with a new approach, and finally solved it in about three hours, emailing it to the company at 6am (0600 for those of you reading this across the Atlantic or from a keyboard marked "Property of US Armed Forces"), just for grins. I wasn't expecting to get a response, since I was grossly beyond the time allotted, but apparently it was good enough to merit a follow-up interview, so yay for me. :-) Upshot is, though, I have an implementation that works, though now I find myself wondering if there's a way to do it in a functional/no-side-effect/persistent-data-structure kind of way....

I still hate them, though, at least the algorithm-based ones, and in a fleeting moment of transparent honesty, I will admit it's probably because I'm not very good at them, but if you repeat that to anyone I'll deny it as outrageous slander and demand satisfaction, Nerf guns at ten paces.

.NET | Android | Azure | C# | C++ | Conferences | Development Processes | F# | Industry | iPhone | Java/J2EE | Languages | Objective-C | Parrot | Personal | Python | Ruby | Scala | Social | Visual Basic

Monday, July 8, 2013 11:02:11 PM (Pacific Standard Time, UTC-08:00)
Comments [3]  |