faster n choose k for combination of array ruby - ruby

Question

While trying to solve the "paths on a grid" problem, I have written the code
def paths(n, k)
p = (1..n+k).to_a
p.combination(n).to_a.size
end
The code works fine, for instance if n == 8 and k == 2 the code returns 45 which is the correct number of paths.
However the code is very slow when using larger numbers and I'm struggling to figure out how to quicken the process.

Answer 1

Rather than building the array of combinations just to count it, just write the function that defines the number of combinations. I'm sure there are also gems that include this and many other combinatorics functions.
Note that I am using the gem Distribution for the Math.factorial method, but that is another easy one to write. Given that, though, I'd suggest taking #stefan's answer, as it's less overhead.
def n_choose_k(n, k)
Math.factorial(n) / (Math.factorial(k) * Math.factorial(n - k))
end
n_choose_k(10, 8)
# => 45
Note that the n and k here refer to slightly different things than in your method, but I am keeping them as it is highly standard nomenclature in combinatorics for this function.

Answer 2

def combinations(n, k)
return 1 if k == 0 or k == n
(k + 1 .. n).reduce(:*) / (1 .. n - k).reduce(:*)
end
combinations(8, 2) #=> 28
Explanation about the math part
The original equation is
combinations(n, k) = n! / k!(n - k)!
Since n! / k! = (1 * 2 * ... * n) / (1 * 2 * ... * k), for any k <= n there is a (1 * 2 * ... * k) factor both in the numerator and in the denominator, so we can cancel this factor. This makes the equation become
combinations(n, k) = (k + 1) * (k + 2) * ... * (n) / (n - k)!
which is exactly what I did in my Ruby code.

Answer 3

The answers that suggest computing full factorials will generate lots of unnecessary overhead when working with big numbers. You should use the method below for calculating the binomial coefficient: n!/(k!(n-k)!)
def n_choose_k(n, k)
return 0 if k > n
result = 1
1.upto(k) do |d|
result *= n
result /= d
n -= 1
end
result
end
This will perform the minimum operations needed. Note that incrementing d while decrementing n guarantees that there will be no rounding errors. For example, {n, n+1} is guaranteed to have at least one element divisible by two, {n, n+1, n+2} is guaranteed to have at least one element divisible by three and so on.
Your code can be rewritten as:
def paths(x, y)
# Choice of x or y for the second parameter is arbitrary
n_choose_k(x + y, x)
end
puts paths(8, 2) # 45
puts paths(2, 8) # 45
I assume that n and k in the original version were meant to be dimensions so i labeled them x and y instead. There's no need to generate an array here.
Edit: Here is a benchmark script...
require 'distribution'
def puts_time
$stderr.puts 'Completed in %f seconds' % (Time.now - $start_time)
$start_time = Time.now
end
def n_choose_k(n, k)
return 0 if k > n
result = 1
1.upto(k) do |d|
result *= n
result /= d
n -= 1
end
result
end
def n_choose_k_distribution(n, k)
Math.factorial(n) / (Math.factorial(k) * Math.factorial(n - k))
end
def n_choose_k_inject(n, k)
(1..n).inject(:*) / ((1..k).inject(:*) * (1..n-k).inject(:*))
end
def benchmark(&callback)
100.upto(300) do |n|
25.upto(75) do |k|
callback.call(n, k)
end
end
end
$start_time = Time.now
puts 'Distribution gem...'
benchmark { |n, k| n_choose_k_distribution(n, k) }
puts_time
puts 'Inject method...'
benchmark { |n, k| n_choose_k_inject(n, k) }
puts_time
puts 'Answer...'
benchmark { |n, k| n_choose_k(n, k) }
puts_time
Output on my system is:
Distribution gem...
Completed in 1.141804 seconds
Inject method...
Completed in 1.106018 seconds
Answer...
Completed in 0.150989 seconds

Answer 4

Since you're interested in the count rather than the actual combination sets, you should do this with a choose function. The mathematical definition involves evaluating three different factorials, but there's a lot of cancellation going on so you can speed it up by using ranges to avoid the calculations that will be cancelled anyway.
class Integer
def choose(k)
fail 'k > n' if k > self
fail 'args must be positive' if k < 0 or self < 1
return 1 if k == n || k == 0
mm = [self - k, k].minmax
(mm[1]+1..self).reduce(:*) / (2..mm[0]).reduce(:*)
end
end
p 8.choose 6 # => 28
To solve your paths problem, you could then define
def paths(n, k)
(n + k).choose(k)
end
p paths(8, 2) # => 45

Answer 5

The reduce/inject versions are nice. But since speed seemed to be a bit of an issue, I'd suggest the n_choose_k versions from #google-fail.
It is quite insightful and suggests a ~10-fold speed increase.
I would suggest that the iteration use the Lesser of k and ( n - k ).
N-choose-K and N-choose-(N-K) produce the same result (the factors in the denominator are simply reversed). So something like a 52-choose-51 could be done in one iteration.

Answer 6

I usually do the following:
class Integer
def !
(2..self).reduce(1, :*)
end
def choose(k)
self.! / (k.! * (self-k).!)
end
end
Benchmarking:
k = 5
Benchmark.bm do |x|
[10, 100, 1000, 10000, 100000].each do |n|
x.report("#{n}") { n.choose(k) }
end
end
On my machine I get:
user system total real
10 0.000008 0.000001 0.000009 ( 0.000006)
100 0.000027 0.000003 0.000030 ( 0.000031)
1000 0.000798 0.000094 0.000892 ( 0.000893)
10000 0.045911 0.013201 0.059112 ( 0.059260)
100000 4.885310 0.229735 5.115045 ( 5.119902)
Not the fastest thing on the planet, but it's okay for my uses. If it ever becomes a problem, then I can think about optimizing