Ask HN: Am I being too hard on the developers I interview? | Hacker News Reader

Ask HN: Am I being too hard on the developers I interview?

18 pointsjlangenauer72 comments

I'm the CTO at a startup in London, so I'm spending a lot of my time interviewing developers as we try to grow our team. As part of that, I ask each of them to do what I consider to be a very simple programming exercise (See below.)

Yet, the results of this are disappointing. About 10% of candidates will be able to produce a solution in 10 minutes or less, another 30% of candidates will eventually produce something close to a solution in 30 minutes (though often with errors), and 60% of candidates will simply be unable to produce anything close to a solution. These are not junior or graduate developers, but senior, experienced programmers asking for £50k or £60k.

I try to make the candidates as comfortable as possible while doing this, but even so, the inability of 90% of the people I interview to develop a solution to a simple question is astonishing.

Am I being too hard on the candidates in asking them to do this?

-------------------------------------------------------------------

This is, verbatim, the problem sheet I show them:

Pascal’s Triangle

Each element in Pascal’s Triangle is found by adding the two numbers to the right and left of it on the row above it; where these don’t exist, they are assumed to be zero.

1 1

1 2 1

1 3 3 1

1 4 6 4 1

1 5 10 10 5 1

(This is correctly centered in the actual document.)

Write a function that takes a single integer n as a parameter and returns the nth row of Pascal’s Triangle.

For example:

pascal(1) should return [1]

pascal(4) should return [1, 3, 3, 1]

View on HN

> Each element in Pascal’s Triangle is found by adding the two numbers to the right and left of it on the row above it; where these don’t exist, they are assumed to be zero.

That gibberish. Unless you know what "Pascal’s Triangle" is prior to seeing this you have no shot of even understanding the exercise.

You haven't even said how you're meant to iterate through the numbers. If this is meant to be an explanation of a "Pascal’s Triangle" it fails at that.

You aren't testing programming, you're testing obscure maths knowledge and the ability to decrypt your question/explanation.

I bet a lot of the other posters had to google this just to answer it in their impressively quick times.

I remember seeing Pascal's triangle in high school algebra as a way to obtain the coefficients of a polynomial in the form (a+b)^n, i don't see how can one forget the concept, this is far from obscure maths knowledge.

There's nothing obscure about this. This is basic secondary school math.

I never studied it or even heard of it. So from my perspective it is obscure.

I've never needed this in my programming career up until now. Have you?

I agree. "adding the two numbers to the right and left of it on the row above it" makes no sense to me whatsoever. I could interpret that in about many different ways, and during an interview I'd probably crack. To put this in perspective, I got a 1450 on the SATs.

My thought process: I take a number, add the right and left on the same row, then put the result on the upper row. Or, add the left and right and current and put it on the row above. After a while I guess you mean any position is the sum of the numbers to the 1 position upper left and upper right.

> I could interpret that in about many different ways

If you look at the graphic that always comes with this question, there's only one way you could possibly interpret it. If you somehow misinterpret the text, you would know immediately when you test your mental model against the graphic.

To be fair, the screwed up formatting in the OP makes it a little harder to understand; in an interview it would be laid out correctly.

> My thought process:

Would you still think all of those with the diagram?

          1

         1 1

        1 2 1

       1 3 3 1

      1 4 6 4 1

    1 5 10 10 5 1

This phrase:

"...adding the two numbers to the right and left of it on the row above it..."

Doesn't make a whole lot of sense. What is the "it"?

row 1: is obviously easy. row 2: I think you are adding 0 to 1 and 1 to 0 to get 1 and 1 row 3: You are adding 0 + 1 for the first item, 1 + 1 for the second, and 1 + 0 for the third. row 4: You are adding 0 + 1 for the first number, 1 + 2 for the second...but the directions say "to the left AND right of it" which makes me think that the answer should be 4, not 3. 1 + 2 + 1. Really, you are only adding the number to the left of "it" (which still isn't clear). And actually, for the first half of the line you are adding the number to the left and for the second half of the line you are adding the number to the right.

What you want:

row 1: 0 + 1 row 2: 0 + 1, 1 + 0 row 3: 0 + 1, 1 + 1, 1 + 0 row 4: 0 + 1, 1 + 2, 2 + 1, 1 + 0 row 5: 0 + 1, 1 + 3, 3 + 3, 3 + 1, 1 + 0 row 6: 0 + 1, 1 + 4, 4 + 6, 6 + 4, 1 + 0

So I think your directions are probably confusing people.

Of course, you still probably aren't going to increase your percentages much, but you could be clearer in what the problem actually is.

Smart people suffer from analysis paralysis - that's why they suck at simple multiple choice questions. "But it could mean [...]"

No, just take the obvious.

> Each element in Pascal’s Triangle is found by adding the two numbers to the right and left of it on the row above it; where these don’t exist, they are assumed to be zero.

Subject of the sentence: 'Each element'

So, "adding two numbers to the right and left of it on the row above it" means "adding the two numbers to the above right and above left of an element."

> So I think your directions are probably confusing people

Not my directions!

Still, this thread should be useful to OP. There are people who get the problem, and there are people who don't get the problem, and smartness isn't the deciding factor.

(I'm stupid, and not a programmer, but I understood what the problem was and what the desired result would be. Other people are smarter than I am, and are programmers, but didn't seem to grok the question.)

Oops, didn't mean your directions but the OP's, of course. Just thought this thread was a good place for the answer and you had a nicely formatted pyramid.

And, for the record, I'm a programmer, think I'm reasonably intelligent, and still not entirely sure I get the question based on the description. Language is tough sometimes.

I don't understand why are people having troubles interpreting the text.

P(i,j) = the number at i-th row and j-th column, could be defined as:

P(i, j) = P(i-1, j-1) + P(i-1, j+1) - for the indented pyramid diagram

P(i, j) = P(i-1, j-1) + P(i-1, j) - for the non indented triangle (the way it would appear in a basic 2D data structure, eg: int P[][]).

Also, you will have to check if P(i-1,j-1) exists.

> That gibberish

No, it's a fairly straight forward and simple algorithm.

The only ambiguity is due to the fact that it references relative spatial positions...but this question always provides a graphic that makes everything completely clear:

If someone couldn't understand the question with that description and graphic, they are either having a "brain fart"/interview jitters or have some deficiency in their reading comprehension, spatial reasoning, etc that they need to work on. The former is perfectly understandable and will happen no matter what you do in the interview; the latter tells you that the interviewee needs work in some area which is useful information.

> You aren't testing programming, you're testing obscure maths knowledge

Not at all, to answer this question requires absolutely no previous knowledge. All you need to answer is in the question.

You can pretty much just translate the question directly into code:

    function getRow(n) {
        if (n == 1) return [1];
        if (n == 2) return [1, 1];
        var previous_row = getRow(n - 1);
        var current_row = []
        for (var i = 0; i < n; i++) {
            current_row.push(
            	(previous_row[i - 1] || 0) +
                (previous_row[i] || 0)
            );
        }
        return current_row;
    }

Took me maybe a couple minutes to translate the question into a fairly straightforward recursive function and type it out. From there it would be easy to improve in a lot of ways: caching, making it iterative, etc...all of which are relatively obvious and easy.

I doubt there are any interviewers who wouldn't be basically satisfied with the most brain dead implementation. That demonstrates basic competence. The question has ample room to demonstrate a variety of skills beyond that. I'd expect the better interviewees (even with little math skill/background knowledge) to at least be able to make some observations about the properties of the triangle. I'd also expect the better ones to be able to talk about different approaches and their benefits/drawbacks.

For candidates with better math skills or background knowledge, they might know offhand/be able to write a better algorithm for computing the nth row. If the interviewer is expecting that as a baseline, sure I agree that's a bit unreasonable...but I don't think that ever happens.

I think this is a perfectly fine interview question with the caveats that it's very well known (and hence there will be some with a solution memorized) and one shouldn't rely on it too much.

It helps to have a good interviewer who will be able to explain the purpose of the question, and guide the interviewee if they get stuck. This is holds true for any interview question though.

I think this is something you've studied and have entirely lost perspective. The way it is explained on the test actually does not make sense.

For example, does the triangle already exist or are we generating it/calculating it? Are we to iterate over the table? And if so then how? What data format is it in?

Let's also discuss this line:

> Each element in Pascal’s Triangle is found by adding the two numbers to the right and left of it on the row above it; where these don’t exist, they are assumed to be zero.

The way it is worded is just horrible. Try this one:

"At the tip of Pascal's Triangle is the number 1, which makes up the zeroth row. The first row (1 & 1) contains two 1's, both formed by adding the two numbers above them to the left and the right, in this case 1 and 0 (all numbers outside the Triangle are 0's). Do the same to create the 2nd row: 0+1=1; 1+1=2; 1+0=1. And the third: 0+1=1; 1+2=3; 2+1=3; 1+0=1. In this way, the rows of the triangle go on infinitly. A number in the triangle can also be found by nCr (n Choose r) where n is the number of the row and r is the element in that row. For example, in row 3, 1 is the zeroth element, 3 is element number 1, the next three is the 2nd element, and the last 1 is the 3rd element. "

With this diagram: http://www.mathsisfun.com/images/pascals-triangle-1.gif

> I think this is something you've studied and have entirely lost perspective.

I've studied programming sure; anyone who's studied programming should have no problem with this. In terms of this particular problem, not really. I've come across it before, and understood it immediately then as well.

I think you are fundamentally misunderstanding the process/reason behind such a question.

> For example, does the triangle already exist or are we generating it/calculating it? Are we to iterate over the table? And if so then how? What data format is it in?

These aren't reasonable/good questions. That is, only a person who hasn't paid attention, misunderstood the question, etc would ask these. They represent a flaw in the reader, not the problem as it was phrased.

I think the issue here is you just don't understand the process and reasoning behind this question. You seem to think it's some sort of "gotcha" when that's not the case at all. There's absolutely no reason you can't ask questions like you pose. Everyone sometimes has issues with the way things are worded...and being able to ask for clarification is important.

I don't see how your explanation improves anything; it doesn't add any information it's just more explicit and convoluted. Many (I'd say most) programmers can solve the problem easily the first time they encounter it with no background knowledge whatsoever. I did it, as have many others. There will be many people who need a little help, that's fine...this is not a test to weed them out. It's to weed out the people who don't ask for a little help, or just can't do it at all.

Most of the times and interviewer would intentionally omit some specifics about the problem the candidate must resolve, thus leaving room for questions. The questions the candidate asks in order to complete a certain task are as important as the task itself.

Regarding the above problem, in my opinion it has everything you need for a good understanding. The only thing it's missing would be some limits for n.

If you have trouble understanding this task, it means you have trouble understanding basic algorithmic problems as most of them are defined in a similar manner.

To also, answer jlangenauer's question: this would be a perfect interview task for a junior. If you find many candidates that fail this kind of problems, you should seriously consider a way of better screening your candidates. (phone interview, online tasks, etc).

     def cal(n):
         if n == 1:
            return [1]
         else:
            x = cal(n-1)
            return [1] + [x[i]+x[i-1] for i in range(1,len(x))] + [1]

    I didnt know "Pascal’s Triangle" , nor did I google, maybe I just got lucky.
    (No CS degree, self-taught)

----------------------------------------------------------------------------------------------------------------------------

I'm not yet a great programmer, but no doubt I'm becoming one

Wanna hire me, for free :)

Location: SF,CA

----------------------------------------------------------------------------------------------------------------------------

my site: codefella.herokuapp.com

Many developers who don't come from a CS or Math background will run from such questions for the following reasons:

1) They've never solved problems like this with code before.

2) They fail to grasp how such problem solving skills can judge their "real world" experience.

3) Someone across the street/city/village will pay them more than you are paying and have more softball interview questions because they are more desperate to bring in any programming talent they can find.

While reasons 1 and 2 are good reasons to pass on candidates, reason 3 is unfortunately a bigger challenge to solve.

It may be unpopular to say this, but I would take a hard look at the problems you need your developers to solve and come up with a more real world scenario to use for interviews and see if you get better results.

That would probably make a better homework problem. People get nervous in interviews and can't even do fizzbuzz.

In general though it's better to ask people problems related to things they'll actually be working on. E.g., have them analyze a bug you recently found in your own code.

You might be right, though one of the reasons I like this particular problem is that it lends itself to a recursive solution, and some of our production code is heavily recursive.

I have to agree. This probably isn't all that difficult to work out when you're in the zone, home alone but in an interview is completely different. You're in a completely different state and it's hardly reflective of a normal working environment.

I think if you are going to give this to people in an interview you should a) make sure you've told them to expect puzzles and b) give them a few questions, some as a warm up with plenty of time to settle in.

I never make a candidate take tests. After 20 years in this industry I have come to the conclusion that they are useless. It's applied Curse of knowledge, one assumes because they can do it, all others should be able to. I have seen some of the best developers in the world that cannot do fizzbuzz in an interview situation no matter how comfortable you make them.

Personally, I have a really good track record of finding and hiring talent. So much so that, friends in other companies will ask me to find them someone. Anyways, one of the most powerful things I do to find talent is ask them either send me some code ahead of time or better yet bring a laptop with a working project they have done, on it. I then ask one simple question, show me the piece of code in this project that you wrote and are most proud of. Someone that is BS'ing can't BS through that. It shows real world application, and people are generally most proud of things that they had to overcome adversity on, they are happy to talk about how they solved the problem. People try to make interviewing hard, but with that single question you will know everything you need to know about their technical skills. Personality skills is another matter, but I just take them to lunch and hang out with them to figure out if they are a fit personality wise.

What a great philosophy! Are you hiring now??

Take a look at my comment to klaut, the same applies for you, feel free to contact me, but I am not actively looking for people.

That is an excellent way of conducting an interview. Are you hiring? ;)

At the moment I have no plans to expand. My current company is just a one man consultancy (service based). At some point though I will start or join another product based start-up, at which point I will be. If you would like, contact me via my info in my profile and when that time comes I will let you know.

The question completely confused me, probably because I've never seen it before and the triangle is apparently not printing correctly here.

I don't really care for interview questions that are unlikely to show up in the real world. This feels like sort of a "gotcha" question. Perhaps you will get the best of the best in your 10%. But I suspect you are missing a lot of candidates who are very good at real-world programming.

Took a minute to code a solution using the formula for line n of Pascal's triangle, and ten minutes (my usual mix of 30% manic typing, 70% bug-fixin' and finding the typo where I wrote "naemspace" :p ) to get an iterative line-by-line working solution in C++. You say you're hiring? I'm in the UK and I'm more than a little bored where I am.

We do mostly Ruby on Rails as well as C#/Unity 3D. Send me an email at jason@beyondthestory.com .

I tick precisely none of those boxes. Ah well, back to debugging today's C++ infinite template error :p

Don't stick with a job that bores you. It's soul destroying. My site might have a few bits and pieces that may interest you: http://hackerjobs.co.uk/jobs

I don't think you're being too hard, but your instrument may be too blunt. If you basing a large part of your hiring decision off this one test then you may be eliminating people unnecessarily and occasionally including someone who doesn't belong. For example I'm pretty rusty as a programmer, having graduated to management many years ago (and a linux admin before that) and I could pass this test, but you wouldn't want me as a senior programmer.

The point I'm getting at, is that like any science experiment you need to control for the variables. Just like an economics experiment might control for wealth, or education, you have to control for people who get nervous during interviews, people who have a blind spot for this particular problem, etc.

An easy addition to your methodology would be to have three different problems and let them choose. I would see how that affects their performance.

In my company I give potential hires a test for them to complete at home, before even interviewing them. The average time to complete the test is 6 hours, but we've never had anyone outright refuse to attempt it, and it gives us a giant steaming pile of data. So if my experience is any guide (and it might not be) I would think you could make your test more intensive, and potentially end up with more and better candidates.

Do you actually ask them to do this on a piece of paper?

because I went through an interview process not too long ago, and they asked me to take a test that was actually very simple stuff, fibonacci sequence, dealing with strings, etc.

And for the life of me I just couldn't do it on paper, if I was still in college I might have done it, but now that I'm a professional developer, I DON'T code on paper anymore, it feels weird and awkward. Maybe do a test, and let a few take the test on the computer, doesn't have to be in an IDE, just let them hack away at notepad, or something.

I know most people believe developers should be those guys that can code their way out of any problem with just matchsticks or something, but really, would you ever encounter a situation where you need a developer to actually write code on paper?

And before anyone says I'm just a bad programmer, I am usually the go to guy at my workplace, the guy that usually solves hard issues, and the more competent I become at actually useful stuff, and actual day to day programming challenges, the more I feel far from things like your test.

Alternatively, you could also use a real world problem, I think there is nothing that can get a guy to prove they are a good developer than making subtle bugs in some code and asking the guy to fix them.

No, it's definitely done at a computer, I know I'd find it more difficult to write code on paper too.

Also consider that you might be below market rate and all the good guys are passing on you.

The story reminds me of a blog post from a while ago [1] which concluded with an update that after raising his rates by $5 the first candidate to apply passed with flying colors.

[1] http://yakovfain.com/2012/10/11/the-degradation-of-java-deve....

I have interviewed people at several companies, but at one company I went through hundreds of resumes, probably over 100 phone screens, and dozens of interviews. I noticed some patterns.

One was that every technical question after the third one was usually a waste. Anyone who hit it out of the park on the first three, hit almost all of the subsequent ones out of the park. No one who struggled through the first three would suddenly start doing well on the subsequent questions.

Another thing I noticed is people fell into a Gaussian distribution. Some were on the left end of it, but most of those were weeded out by the recruiters, or by resumes or finally by phone screens. Just like a bell curve, most of the people were in the middle. Of the dozens I interviewed in person, only three were on the right end of the bell curve. Which is before you even get to soft skills.

Your results speak for themselves. You have found an exercise that helps mark the top 10% from the 11-40% from the 41-100%.

You might want to read some blog posts like Jeff Atwood's "Why can't programmers program?"

Because I wanted distracting for 5 min, in python:

  def pascal(row, prev_row=None):
      if row == 1:
          return [1]
      else:
          if not prev_row:
              prev_row = pascal(row-1)
          new_row = []
          for i in xrange(row):
              if (i == 0) or (i == row-1):
                  new_row.append(1)
              else:
                  new_row.append(prev_row[i-1]+prev_row[i])
          return new_row

  def none_recursive_pascal(row):
      current_row = None
      for i in xrange(1,row+1):
          current_row = pascal(i, current_row)
      return current_row


  for i in xrange(1,10):
      print pascal(i)

  for i in xrange(1,10):
      print none_recursive_pascal(i)

EDIT: Added a none recursive version.

I'm not sure what your startup does, but I would bet there are better things to ask a candidate than this. You get these problems in school and become quite good at them in the moment but after doing real world work, this mindset will fade. I consider myself a pretty good coder. I've made full production apps that work just fine. But who knows how I would handle this problem if handed to me at an interview I'm already nervous about.

Give this problem in the same setting to some of your currently employed programmers. If they have similar results as your candidate then your candidate filter sucks and you should re-evaluate it.

I wouldn't expect them to write the code for it, maybe some pseudo code, but a candidate should be able to talk through the problem and come to a solution for this in 10-15 minutes. It also depends on what kind of work you expect them to do. If you are building iPhone apps or doing plain old web dev, you probably don't need candidates who can solve this in <10 minutes.

I'm not surprised by your 90% no-solve stat. Candidates I interview who put Java/C# on their resumes can barely "write a function that returns true if the first letter of a string parameter is capitalized". At least half flat out can't do it.

It took me a couple of minutes to remember the pascal triangle due to the way you are presenting the triangle. I'd rather have it structured as such:

It makes it easier to understand when you say 'the two numbers to the right and left on the row above'. However, the way you present it may make it easier to find a solution since it hints to storing the numbers in an array that can be used in each iteration and highlights the relations between the positions of the numbers. But i guess much of this shouldn't be a problem to experienced programmers.

That's an artifact of the HN display system. When it's presented to the candidates, it's definitely correctly centered.

I would say that for an average group of candidates the results were about right. About 10% of "programmers" can write a simple program that solves a simple problem when asked - most can't.

Hi, new guy here. My solution took me about 15 minutes to code in Python, but I looked up my algorithm (I am not nearly as math-oriented as I should be; programming came later in life to me). I'm genuinely curious about anyone's opinion of where I would be ranked among candidates.

def gen_pascal_row(n): row = [1] for i in range(1,n+1): row.append(row[-1] * n/i) n -= 1 return row

ETA: This might be better. https://gist.github.com/4612320

I'm a firm believer in talent dilution being a huge problem, so if this is the standard that you set for full time hires then keep it. Although it may be a good idea to think about wether or not this problem is representative of someones abilities to fill the position. You don't want to fall into the trap of asking an interview question (albeit a fairly easy on in my opinion) that may not tell you how good someone would be at filling the postion your hiring for.

Unless you are hiring a developer for a critical banking software or something similar, you should never base your interviews on Math. Rather, interviews should be based on the problem solving attitude of the candidate. I'd prefer a candidate who can come up with an innovative and cheaper solution to the problem at hand than a person who is better at Math. Though that person would also need to be good at the programming skill that he/she is being hired for.

Are you asking them to do this in a particular language, or just to come up with an algorithm? This might work better as the sort of problem you ask an interviewee to talk you through, rather than an "I'll check on you in 10 minutes" kind of exercise. But in general, no, this kind of thing isn't out of bounds, and I'd say if 10% of your candidates are passing it, you're in the right sort of neighborhood.

I tell them to do it in any language they're comfortable with, though obviously, if they're interviewing for a Ruby rule, I'd be surprised if they didn't do it in Ruby.

I actually do ask them to write the code - I'd expect most programmers to know the basic syntax of the language the claim to be proficient in, but I've seen Ruby developers who are unable to correctly use Array#each, which would be a red flag.

Also, they don't need to come up with an algorithm: to solve the problem, you simply need to implement exactly what's written in the problem statement - calculating each element based on the previous row, and with a couple of conditional statements (or other ways) to handle the edge cases at the, er, edge of each row. There is a bit of a conceptual leap in working out a recursive solution is possible (as is an iterative one), so I'll sometimes gently drop hints if a candidate seems to be struggling on this point.

It took me about 20 minutes to solve using a recursive solution error-free, and I have not been doing any math intense work lately and am well out of practice.

Anyone with decent exposure to math has at least SEEN pascals triangle, so I think it is a fair question. Try as you might to make them comfortable, they are probably still nervous regardless.

This doesn't seem too hard to me but maybe that's because I'm mathy. At one of previous places my boss had found that he was able to boil down the interview to a take home problem that the candidate should just be prepared to discuss, not even solve. Literally 1 out of 10 did this. The rest gave up before hand.

It's hard to find good people.

Something like this should do the trick, good old coffee.

pascal_r = (n) -> lines = [[1]] for i in [1...n] last = lines[i-1] lines[i] = [] for j in [0..i] lines[i][j] = (last[j-1] or 0) + (last[j] or 0) lines[n-1]

console.log pascal_r(1) console.log pascal_r(4)

Edit: Bloody HN formatting got the better of me again.

given that "Ruby Programmer" >= "PHP Programmer" (not necessarily true, but be honest PHP is considered pretty low on the food chain around here :)), and given that I have no CS or Math background, I whipped up an answer to this in about 5 mins:

    https://gist.github.com/4611869

Even taking into account the effect of 'nerves' in an interview setting, on the face of things the 10% mentioned seems a bit depressing.

I do wonder how relevant or helpful this particular exercise/test is with regard to accurately gauging a candidate's potential efficacy within [a|your] development team. My instinct suggests to me that being able to write an algorithm under scrutiny and/or pressure is actually a fairly low priority skill (in most development shops) when compared to the other many other skills that are required to be a successful senior developer.

data point: I started learning Python/Django 4 months ago (after a 20 year break from programmiung). Took me 40 minutes to solve. I'm not yet fluent with syntax so I had to look up (and still managed to mangle list processing syntax and was slow to interpret error messages related to accidentally swapping my variables for size of triangle and the list. My logic was fine once I got got the syntax down and swaps corrected. So your test would have caused me to be weeded out, I guess.

Had you asked me to pseudo code it I could have done it in less than 10 minutes.

Note, however, that learning web development requires learning many different pieces in addition to Python, so I simply haven't had the time to thoroughly learn all the syntax of any given piece.

Worried that I'm too slow, I then tracked down the reference to the fizzbuzz test: http://www.codinghorror.com/blog/2007/02/why-cant-programmer... That one took me 11 minutes. There was less syntax and more logic tested with fizzbuzz. The only thing that tripped me up for a few minutes was I found out (through debugging) that integers in Python stay integers when divided. So I had to float(i) before dividing by 3 or 5 to get it to work.

I don't know if you're trying to weed out people weak on syntax but your test may very well do that. Syntax fluency generally gets better over time, I would think. Not sure basic logic and debugging skill gets better over time.

     def cal(n):
         if n == 1:
            return [1]
         else:
            x = cal(n-1)
            return [1] + [x[i]+x[i-1] for i in range(1,len(x))] + [1]

who wanna hire a junior developer? I will work for free in SF,CA

I actually had a bit of trouble with this problem, it took a while to iron out the bugs. I copied down the triangle for reference but accidentally had 1 2 2 1 as the third row.

Took me about 15 minutes to do. Doesn't seem like an unreasonable question to ask.

The solution (in Objective-C)

- (NSArray)pascal:(int)n { if (n == 1) return @[@(1)];

NSArray array = [self pascal:n-1]; NSMutableArray* mArray = [@[] mutableCopy]; for(int i = 0; i < array.count+1; i++) { int r = i - 1; if (r >= 0 && i < array.count) [mArray addObject:@([array[r] integerValue] + [array[i] integerValue])]; else if (i < array.count) [mArray addObject:@(0 + [array[i] integerValue])]; else [mArray addObject:@(0 + [array[i-1] integerValue])]; }

return [mArray copy]; }

Have you set the same challenge with the same time constraints for your current team? If you have and they all pass with flying colours then don't change a thing. If you haven't, then it might be worth testing the challenge with them.

10% is disappointing but it also means it's a good filter, particularly when you think recursive algorithms are critical in your projects. A question with 90% passing rate is useless as a filter.

What is your business domain? You might find it more beneficial to ask your candidates to solve something relevant to the work they would be doing for your company.

Are other senior developer positions in your area offering £50k to £60k? The equivalent $80k to $95k seems to be the going rate for junior/mid developers in the US.

We don't have a specific salary figure for the positions - those numbers are what the candidates are asking for.

£50k - £60k would be considered senior territory in London. Most senior developers are closer to £70k and there are a smaller percentage earning closer to £100k.

As a 'junior/mid' dev in the UK salary range is £25-30k on average.

Outside of SF and NYC, where are all of these $80k junior positions?

Boston, DC area, Atlanta, to name a few

$80k in Boston is definitely closer to mid/senior starting salaries.

Depends on how you define mid/senior. I started at a Boston company with about 10 years of experience in '08. My "mid-level" salary was ~ $90K. I ended up around $100K after about 3 years working at the same company (my title got prepended with "Senior" at some point, but it didn't start that way). These numbers do not include bonuses, which varied.

  pascal(n) = [binomial(n, i) for i=0:n]

lol, now define your binomial function.

  binomial(n, k) = factorial(n)/(factorial(k)*factorial(n-k))
  factorial(n) = n <= 1 ? 1 : n*factorial(n-1)

I'm working from random statistics for London that I found on the internet, so please forgive me if this doesn't apply.

You seem to be paying average - average+10k (for all, not just senior) developer salaries, yet you specify senior programmer. At this level, you're only going to be attracting mid-level talent. There could a problem with your recruitment targeting, which could mean you're not getting ideal people in the door to begin with.

At the mid-level, you should expect a high failure rate on simple tasks. You're dealing with people who wound up on the job market after getting a bunch of experience. This means that a lot of them were fired, came from untenable/poorly managed companies which went under/downsized, or were working under such unbearable conditions that they quit. Employed devs trade up - unless your pay packet, benefits or feel-good factor (eg if you're an NGO, charity, working on 'real' problems, etc) give your company an edge, you're not dragging people away from jobs where they're encouraged to develop their skills, work under good conditions and understand software engineering processes. Job postings regularly ask for ridiculous and irrelevent things, so a lot of the employed devs who read your job ad and apply will not meet the criteria you specify. I know it's a huge generalisation, but the best devs are usually only on the market a couple of times in their lives, and they're quickly snapped up by companies at a premium and kept happy with various incentives. Tl;dr: you should be prepared for most candidates that you interview to not be appropriate for the job you've posted.

Thirdly, there's a problem with your, ahem, problem. It's not a great problem, imo because there's a fair amount of nuance to it. I'd want to shine by giving a good solution - vaguely remembering that a value in Pascal's triangle can be derived from combinations or permutations or something similar I'd spend some time trying to figure out if there's some algorithm "n!/(n+k)! - no, that's not it...", then I'd try to do it in-place calculating only half of the triangle, and then probably wind up using two rows of half length to calculate the numbers there. I'd be trying to impress you by coming up with a solution with more useful properties (memory boundedness, complexity, speed) than "returns the row of pascal's triangle". You're probably not even hiring people for their comp sci experience.

Maybe you just wanted to vent, or to advertise here - that's ok, but keep in mind that you're going to have to quickly reject unsuitable candidates and quickly pick up suitable ones, that it's just a numbers game, and that this is why most companies that offer mid-range compensation are "always hiring". The problem isn't too hard as such, it's perhaps a little undirected. Also, some people just interview terribly.

Dunno, just some food for thought.

No, you're not being too hard on your interview candidates. That's not a hard problem, at least under basic assumptions (not concerned with performance, integer overflow, etc.)

I would say you should pass on the 60% who can't solve it, unless they take the initiative and send you a solution after the interview (I've done this, when I could solve the problem but not in time). That might prevent you from passing on an otherwise good candidate who happened to choke once. The 30% in the middle you can attribute to interview stress, if other indications are that they know how to write code.

The truth is, though, that a large percentage of "professional programmers" have never built anything from scratch and have been wearing training wheels the whole time. 90 percent of what professional programmers do is maintenance at ~10 LoC per day, never leaving the comfort of the IDE or building a system from the ground, and so there are a lot who've never been in a green field or had to solve a hard programming problem. You don't want to hire such people to build new stuff. You're not being "too hard" on them by filtering them out.

In fact, I'm surprised that your fail rate is only 60%.

Reply on news.ycombinator.com