• FlapKap@feddit.dk
    link
    fedilink
    arrow-up
    6
    ·
    1 year ago

    According to the article the linear algebra algorithm has a running time of O(log n)

      • IAm_A_Complete_Idiot@sh.itjust.works
        link
        fedilink
        arrow-up
        7
        arrow-down
        1
        ·
        edit-2
        1 year ago

        According to the benchmark in the article it’s already way faster at n = 1000. I think you’re overestimating the cost of multiplication relative to just cutting down n logarithmically.

        log_2(1000) = roughly a growth factor of 10. 2000 would be 11, and 4000 would be 12. Logs are crazy.

        • cbarrick@lemmy.world
          link
          fedilink
          English
          arrow-up
          2
          ·
          1 year ago

          The article is comparing to the dynamic programming algorithm, which requires reading and writing to an array or hash table (the article uses a hash table, which is slower).

          The naive algorithm is way faster than the DP algorithm.

          • t_veor
            link
            fedilink
            English
            arrow-up
            2
            ·
            1 year ago

            It’s not that hard to check yourself. Running the following code on my machine, I get that the linear algebra algorithm is already faster than the naive algorithm at around n = 100 or so. I’ve written a more optimised version of the naive algorithm, which is beaten somewhere between n = 200 and n = 500.

            Try running this Python code on your machine and see what you get:

            import timeit
            
            def fib_naive(n):
                a = 0
                b = 1
                while 0 < n:
                    b = a + b
                    a = b - a
                    n = n - 1
                return a
            
            def fib_naive_opt(n):
                a, b = 0, 1
                for _ in range(n):
                    a, b = b + a, b
                return a
            
            def matmul(a, b):
                return (
                    (a[0][0] * b[0][0] + a[0][1] * b[1][0], a[0][0] * b[0][1] + a[0][1] * b[1][1]),
                    (a[1][0] * b[0][0] + a[1][1] * b[1][0], a[1][0] * b[0][1] + a[1][1] * b[1][1]),
                )
            
            def fib_linear_alg(n):
                z = ((1, 1), (1, 0))
                y = ((1, 0), (0, 1))
                while n > 0:
                    if n % 2 == 1:
                        y = matmul(y, z)
                    z = matmul(z, z)
                    n //= 2
            
                return y[0][0]
            
            def time(func, n):
                times = timeit.Timer(lambda: func(n)).repeat(repeat=5, number=10000)
                return min(times)
            
            for n in (50, 100, 200, 500, 1000):
                print("========")
                print(f"n = {n}")
                print(f"fib_naive:\t{time(fib_naive, n):.3g}")
                print(f"fib_naive_opt:\t{time(fib_naive_opt, n):.3g}")
                print(f"fib_linear_alg:\t{time(fib_linear_alg, n):.3g}")
            

            Here’s what it prints on my machine:

            ========
            n = 50
            fib_naive:      0.0296
            fib_naive_opt:  0.0145
            fib_linear_alg: 0.0701
            ========
            n = 100
            fib_naive:      0.0652
            fib_naive_opt:  0.0263
            fib_linear_alg: 0.0609
            ========
            n = 200
            fib_naive:      0.135
            fib_naive_opt:  0.0507
            fib_linear_alg: 0.0734
            ========
            n = 500
            fib_naive:      0.384
            fib_naive_opt:  0.156
            fib_linear_alg: 0.112
            ========
            n = 1000
            fib_naive:      0.9
            fib_naive_opt:  0.347
            fib_linear_alg: 0.152