Fast Fibonacci recursion

问题:

I'm trying to recall an algorithm on Fibonacci recursion. The following:

public int fibonacci(int n)  {
  if(n == 0)
    return 0;
  else if(n == 1)
    return 1;
  else
    return fibonacci(n - 1) + fibonacci(n - 2);
}

is not what I'm looking for because it's greedy. This will grow exponentially (just look at Java recursive Fibonacci sequence - the bigger the initial argument the more useless calls will be made).

There is probably something like a "cyclic argument shift", where calling previous Fibonacci value will retrieve value instead of calculating it again.

回答1:

maybe like this:

int fib(int term, int val = 1, int prev = 0)
{
 if(term == 0) return prev;
 return fib(term - 1, val+prev, val);
}

this function is tail recursive. this means it could be optimized and executed very efficiently. In fact, it gets optimized into a simple loop..

回答2:

This kind of problems are linear recurrence types and they are solved fastest via fast matrix exponentiation. Here's the blogpost that describes this kind of approach concisely.

回答3:

You can do a pretty fast version of recursive Fibonacci by using memoization (meaning: storing previous results to avoid recalculating them). for example, here's a proof of concept in Python, where a dictionary is used for saving previous results:

results = { 0:0, 1:1 }

def memofib(n):
    if n not in results:
        results[n] = memofib(n-1) + memofib(n-2)
    return results[n]

It returns quickly for input values that would normally block the "normal" recursive version. Just bear in mind that an int data type won't be enough for holding large results, and using arbitrary precision integers is recommended.

A different option altogether - rewriting this iterative version ...

def iterfib(n):
    a, b = 0, 1
    for i in xrange(n):
        a, b = b, a + b
    return a

... as a tail-recursive function, called loop in my code:

def tailfib(n):
    return loop(n, 0, 1)

def loop(i, a, b):
    if i == 0:
        return a
    return loop(i-1, b, a+b)

回答4:

I found interesting article about fibonacci problem

here the code snippet

# Returns F(n)
def fibonacci(n):
    if n < 0:
        raise ValueError("Negative arguments not implemented")
    return _fib(n)[0]


# Returns a tuple (F(n), F(n+1))
def _fib(n):
    if n == 0:
        return (0, 1)
    else:
        a, b = _fib(n // 2)
        c = a * (2 * b - a)
        d = b * b + a * a
        if n % 2 == 0:
            return (c, d)
        else:
            return (d, c + d)

# added iterative version base on C# example
def iterFib(n):
    a = 0
    b = 1
    i=31
    while i>=0:
        d = a * (b * 2 - a)
        e = a * a + b * b
        a = d
        b = e
        if ((n >> i) & 1) != 0:
            c = a + b;
            a = b
            b = c
        i=i-1
    return a

回答5:

Say you want to have the the n'th fib number then build an array containing the preceeding numbers

int a[n];
a[0] = 0;
a[1] =1;
a[i] = n[i-1]+n[n-2];

回答6:

An example in JavaScript that uses recursion and a lazily initialized cache for added efficiency:

var cache = {};

function fibonacciOf (n) {
  if(n === 0) return 0;
  if(n === 1) return 1;
  var previous = cache[n-1] || fibonacciOf(n-1);
  cache[n-1] = previous;
  return previous + fibonacciOf(n-2);
};

回答7:

duedl0r's algorithm translated to Swift:

func fib(n: Int, previous: (Int, Int) = (0,1)) -> Int {
    guard n > 0 else { return 0 }
    if n == 1 { return previous.1 }
    return fib(n - 1, previous: (previous.1, previous.0 + previous.1))
}

worked example:

fib(4)
= fib(4, (0,1) )
= fib(3, (1,1) )
= fib(2, (1,2) )
= fib(1, (2,3) )
= 3

回答8:

You need to memorize the calculated value in order to stop exponential growth.

1. Just use an array to store the value.
2. Check the array if you have already calculate it.
3. If it finds it,use it or otherwise calculate it and store it.

Here is an working example for faster recursion using memory.

Calculating fibonacci number

回答9:

A good algorithm for fast fibonacci calculations is (in python):

def fib2(n):
    # return (fib(n), fib(n-1))
    if n ==  0: return (0,  1)
    if n == -1: return (1, -1)
    k, r = divmod(n, 2) # n=2k+r
    u_k, u_km1 = fib2(k)
    u_k_s, u_km1_s = u_k**2, u_km1**2  # Can be improved by parallel calls
    u_2kp1 = 4 * u_k_s - u_km1_s + (-2 if k%2 else 2)
    u_2km1 = u_k_s + u_km1_s
    u_2k   = u_2kp1 - u_2km1
    return (u_2kp1, u_2k) if r else (u_2k, u_2km1)

def fib(n):
    k, r = divmod(n, 2) # n=2k+r
    u_k, u_km1 = fib2(k)
    return (2*u_k+u_km1)*(2*u_k-u_km1)+(-2 if k%2 else 2) if r else u_k*(u_k+2*u_km1)

If you need very fast computation, links to the libgmp and use mpz_fib_ui() or mpz_fib2_ui() functions.