最快的家用Prime发电机


23

什么是家庭必备?

例如,以HP(4)为例。首先,找到主要因素。4的质因子(按从最小到最大的数字顺序,始终为)为2。2。将这些因子作为文字数。2,2变成22。分解的过程一直持续到您达到素数为止。

number    prime factors
4         2, 2
22        2, 11
211       211 is prime

达到质数后,序列结束。HP(4)= 211。这是一个更长的例子,有14个:

number    prime factors
14        2, 7
27        3, 3, 3
333       3, 3, 37
3337      47, 71
4771      13, 367
13367     13367 is prime

您面临的挑战是创建一个程序,该程序将在给定x的情况下计算HP(x)并尽快执行。您可以使用所需的任何资源,而不是已知的家庭素数列表。

请注意,这些数字很快变得非常大。在x = 8时,HP(x)一直跳到3331113965338635107。尚未找到HP(49)。

平均在以下输入下,将在Raspberry Pi 2上测试程序速度:

16
20
64
65
80

如果您有Raspberry Pi 2,请自行安排程序时间,然后平均时间。


3
尽快定义。
LegionMammal978

1
@ LegionMammal978具有最佳的运行时性能。这是最快的代码挑战。
Noah L


1
我们如何知道哪个代码更快?有些人可能会在五十岁的笔记本电脑(被检测咳嗽像我咳嗽),而另一些则可能使用一些高端台式机/服务器。同样,同一语言的口译员的表现也不同。
JungHwan Min

1
是否允许使用诸如Miller-Rabin之类的概率素数检验?
英里

Answers:


6

Mathematica,HP(80)在〜0.88s

NestWhile[
  FromDigits[
    Flatten[IntegerDigits /@ 
      ConstantArray @@@ FactorInteger[#]]] &, #, CompositeQ] &

匿名函数。将数字作为输入并返回数字作为输出。


1在年底不应该在那里...
JungHwan民

我没有数学在我的电脑上,这意味着我必须对我的树莓派2进行测试(和程序的其余部分)
诺阿大号

由于我们不打高尔夫球:有CompositeQ!PrimeQ(这也保证你的回答永远不循环输入1)。
马丁·恩德

Mathematica如何HP(80)在如此短的时间内执行而没有素数在某处进行硬编码?我的酷睿i7笔记本电脑正在小时进行素性检查,以及发现的主要因素,HP(80)当它到达47109211289720051
马里奥(Mario)

@NoahL Mathematica可以在线进行测试。meta.codegolf.stackexchange.com/a/1445/34718
mbomb007'1

5

PyPy 5.4.1 64位(Linux),HP(80)〜1.54s

32位版本的时间会稍慢一些。

我使用四种不同的因式分解方法,凭经验确定断点:

  • n <2 30 →通过试验除法分解
  • n <2 70Pollard的Rho(布伦特的变种)
  • elseLenstra ECF(使用蒙哥马利曲线,Suyama参数),测试约100条曲线
  • 如果ECF失败多项式二次筛(使用蒙哥马利多项式)

我尝试了一段时间,以找到ECF和MPQS之间的明确中断,但似乎没有一个。但是,如果n包含一个很小的因子,ECF通常会几乎立即找到它,所以在选择MPQS之前,我选择仅测试几条曲线。

目前,它仅比Mathmatica慢2倍左右,我当然认为它是成功的。


home-prime.py

import math
import my_math
import mpqs

max_trial = 1e10
max_pollard = 1e22

def factor(n):
  if n < max_trial:
    return factor_trial(n)
  for p in my_math.small_primes:
    if n%p == 0:
      return [p] + factor(n/p)
  if my_math.is_prime(n):
    return [n]
  if n < max_pollard:
    p = pollard_rho(n)
  else:
    p = lenstra_ecf(n) or mpqs.mpqs(n)
  return factor(p) + factor(n/p)


def factor_trial(n):
  a = []
  for p in my_math.small_primes:
    while n%p == 0:
      a += [p]
      n /= p
  i = 211
  while i*i < n:
    for o in my_math.offsets:
      i += o
      while n%i == 0:
        a += [i]
        n /= i
  if n > 1:
    a += [n]
  return a


def pollard_rho(n):
  # Brent's variant
  y, r, q = 0, 1, 1
  c, m = 9, 40
  g = 1
  while g == 1:
    x = y
    for i in range(r):
      y = (y*y + c) % n
    k = 0
    while k < r and g == 1:
      ys = y
      for j in range(min(m, r-k)):
        y = (y*y + c) % n
        q = q*abs(x-y) % n
      g = my_math.gcd(q, n)
      k += m
    r *= 2
  if g == n:
    ys = (ys*ys + c) % n
    g = gcd(n, abs(x-ys))
    while g == 1:
      ys = (ys*ys + c) % n
      g = gcd(n, abs(x-ys))
  return g

def ec_add((x1, z1), (x2, z2), (x0, z0), n):
  t1, t2 = (x1-z1)*(x2+z2), (x1+z1)*(x2-z2)
  x, z = t1+t2, t1-t2
  return (z0*x*x % n, x0*z*z % n)

def ec_double((x, z), (a, b), n):
  t1 = x+z; t1 *= t1
  t2 = x-z; t2 *= t2
  t3 = t1 - t2
  t4 = 4*b*t2
  return (t1*t4 % n, t3*(t4 + a*t3) % n)

def ec_multiply(k, p, C, n):
  # Montgomery ladder algorithm
  p0 = p
  q, p = p, ec_double(p, C, n)
  b = k >> 1
  while b > (b & -b):
    b ^= b & -b
  while b:
    if k&b:
      q, p = ec_add(p, q, p0, n), ec_double(p, C, n)
    else:
      q, p = ec_double(q, C, n), ec_add(p, q, p0, n),
    b >>= 1
  return q

def lenstra_ecf(n, m = 5):
  # Montgomery curves w/ Suyama parameterization.
  # Based on pseudocode found in:
  # "Implementing the Elliptic Curve Method of Factoring in Reconfigurable Hardware"
  # Gaj, Kris et. al
  # http://www.hyperelliptic.org/tanja/SHARCS/talks06/Gaj.pdf
  # Phase 2 is not implemented.
  B1, B2 = 8, 13
  for i in range(m):
    pg = my_math.primes()
    p = pg.next()
    k = 1
    while p < B1:
      k *= p**int(math.log(B1, p))
      p = pg.next()
    for s in range(B1, B2):
      u, v = s*s-5, 4*s
      x = u*u*u
      z = v*v*v
      t = pow(v-u, 3, n)
      P = (x, z)
      C = (t*(3*u+v) % n, 4*x*v % n)
      Q = ec_multiply(k, P, C, n)
      g = my_math.gcd(Q[1], n)
      if 1 < g < n: return g
    B1, B2 = B2, B1 + B2


if __name__ == '__main__':
  import time
  import sys
  for n in sys.argv[1:]:
    t0 = time.time()
    i = int(n)
    f = []
    while len(f) != 1:
      f = sorted(factor(i))
      #print i, f
      i = int(''.join(map(str, f)))
    t1 = time.time()-t0
    print n, i
    print '%.3fs'%(t1)
    print

采样时间

    $ pypy home-prime.py 8 16 20 64 65 80
8 3331113965338635107
0.005s

16 31636373
0.001s

20 3318308475676071413
0.004s

64 1272505013723
0.000s

65 1381321118321175157763339900357651
0.397s

80 313169138727147145210044974146858220729781791489
1.537s

5的平均值约为0.39s。


依存关系

mpqs.py直接取自我对最快半素数分解的回答,并做了一些非常小的修改。

mpqs.py

import math
import my_math
import time

# Multiple Polynomial Quadratic Sieve
def mpqs(n, verbose=False):
  if verbose:
    time1 = time.time()

  root_n = my_math.isqrt(n)
  root_2n = my_math.isqrt(n+n)

  # formula chosen by experimentation
  # seems to be close to optimal for n < 10^50
  bound = int(5 * math.log(n, 10)**2)

  prime = []
  mod_root = []
  log_p = []
  num_prime = 0

  # find a number of small primes for which n is a quadratic residue
  p = 2
  while p < bound or num_prime < 3:

    # legendre (n|p) is only defined for odd p
    if p > 2:
      leg = my_math.legendre(n, p)
    else:
      leg = n & 1

    if leg == 1:
      prime += [p]
      mod_root += [int(my_math.mod_sqrt(n, p))]
      log_p += [math.log(p, 10)]
      num_prime += 1
    elif leg == 0:
      if verbose:
        print 'trial division found factors:'
        print p, 'x', n/p
      return p

    p = my_math.next_prime(p)

  # size of the sieve
  x_max = bound*8

  # maximum value on the sieved range
  m_val = (x_max * root_2n) >> 1

  # fudging the threshold down a bit makes it easier to find powers of primes as factors
  # as well as partial-partial relationships, but it also makes the smoothness check slower.
  # there's a happy medium somewhere, depending on how efficient the smoothness check is
  thresh = math.log(m_val, 10) * 0.735

  # skip small primes. they contribute very little to the log sum
  # and add a lot of unnecessary entries to the table
  # instead, fudge the threshold down a bit, assuming ~1/4 of them pass
  min_prime = int(thresh*3)
  fudge = sum(log_p[i] for i,p in enumerate(prime) if p < min_prime)/4
  thresh -= fudge

  sieve_primes = [p for p in prime if p >= min_prime]
  sp_idx = prime.index(sieve_primes[0])

  if verbose:
    print 'smoothness bound:', bound
    print 'sieve size:', x_max
    print 'log threshold:', thresh
    print 'skipping primes less than:', min_prime

  smooth = []
  used_prime = set()
  partial = {}
  num_smooth = 0
  prev_num_smooth = 0
  num_used_prime = 0
  num_partial = 0
  num_poly = 0
  root_A = my_math.isqrt(root_2n / x_max)

  if verbose:
    print 'sieving for smooths...'
  while True:
    # find an integer value A such that:
    # A is =~ sqrt(2*n) / x_max
    # A is a perfect square
    # sqrt(A) is prime, and n is a quadratic residue mod sqrt(A)
    while True:
      root_A = my_math.next_prime(root_A)
      leg = my_math.legendre(n, root_A)
      if leg == 1:
        break
      elif leg == 0:
        if verbose:
          print 'dumb luck found factors:'
          print root_A, 'x', n/root_A
        return root_A

    A = root_A * root_A

    # solve for an adequate B
    # B*B is a quadratic residue mod n, such that B*B-A*C = n
    # this is unsolvable if n is not a quadratic residue mod sqrt(A)
    b = my_math.mod_sqrt(n, root_A)
    B = (b + (n - b*b) * my_math.mod_inv(b + b, root_A))%A

    # B*B-A*C = n <=> C = (B*B-n)/A
    C = (B*B - n) / A

    num_poly += 1

    # sieve for prime factors
    sums = [0.0]*(2*x_max)
    i = sp_idx
    for p in sieve_primes:
      logp = log_p[i]

      inv_A = my_math.mod_inv(A, p)
      # modular root of the quadratic
      a = int(((mod_root[i] - B) * inv_A)%p)
      b = int(((p - mod_root[i] - B) * inv_A)%p)

      amx = a+x_max
      bmx = b+x_max

      ax = amx-p
      bx = bmx-p

      k = p
      while k < x_max:
        sums[k+ax] += logp
        sums[k+bx] += logp
        sums[amx-k] += logp
        sums[bmx-k] += logp
        k += p

      if k+ax < x_max:  
        sums[k+ax] += logp
      if k+bx < x_max:
        sums[k+bx] += logp
      if amx-k > 0:
        sums[amx-k] += logp
      if bmx-k > 0:
        sums[bmx-k] += logp
      i += 1

    # check for smooths
    x = -x_max
    for v in sums:
      if v > thresh:
        vec = set()
        sqr = []
        # because B*B-n = A*C
        # (A*x+B)^2 - n = A*A*x*x+2*A*B*x + B*B - n
        #               = A*(A*x*x+2*B*x+C)
        # gives the congruency
        # (A*x+B)^2 = A*(A*x*x+2*B*x+C) (mod n)
        # because A is chosen to be square, it doesn't need to be sieved
        sieve_val = (A*x + B+B)*x + C

        if sieve_val < 0:
          vec = {-1}
          sieve_val = -sieve_val

        for p in prime:
          while sieve_val%p == 0:
            if p in vec:
              # keep track of perfect square factors
              # to avoid taking the sqrt of a gigantic number at the end
              sqr += [p]
            vec ^= {p}
            sieve_val = int(sieve_val / p)

        if sieve_val == 1:
          # smooth
          smooth += [(vec, (sqr, (A*x+B), root_A))]
          used_prime |= vec
        elif sieve_val in partial:
          # combine two partials to make a (xor) smooth
          # that is, every prime factor with an odd power is in our factor base
          pair_vec, pair_vals = partial[sieve_val]
          sqr += list(vec & pair_vec) + [sieve_val]
          vec ^= pair_vec
          smooth += [(vec, (sqr + pair_vals[0], (A*x+B)*pair_vals[1], root_A*pair_vals[2]))]
          used_prime |= vec
          num_partial += 1
        else:
          # save partial for later pairing
          partial[sieve_val] = (vec, (sqr, A*x+B, root_A))
      x += 1

    prev_num_smooth = num_smooth
    num_smooth = len(smooth)
    num_used_prime = len(used_prime)

    if verbose:
      print 100 * num_smooth / num_prime, 'percent complete\r',

    if num_smooth > num_used_prime and num_smooth > prev_num_smooth:
      if verbose:
        print '%d polynomials sieved (%d values)'%(num_poly, num_poly*x_max*2)
        print 'found %d smooths (%d from partials) in %f seconds'%(num_smooth, num_partial, time.time()-time1)
        print 'solving for non-trivial congruencies...'

      used_prime_list = sorted(list(used_prime))

      # set up bit fields for gaussian elimination
      masks = []
      mask = 1
      bit_fields = [0]*num_used_prime
      for vec, vals in smooth:
        masks += [mask]
        i = 0
        for p in used_prime_list:
          if p in vec: bit_fields[i] |= mask
          i += 1
        mask <<= 1

      # row echelon form
      col_offset = 0
      null_cols = []
      for col in xrange(num_smooth):
        pivot = col-col_offset == num_used_prime or bit_fields[col-col_offset] & masks[col] == 0
        for row in xrange(col+1-col_offset, num_used_prime):
          if bit_fields[row] & masks[col]:
            if pivot:
              bit_fields[col-col_offset], bit_fields[row] = bit_fields[row], bit_fields[col-col_offset]
              pivot = False
            else:
              bit_fields[row] ^= bit_fields[col-col_offset]
        if pivot:
          null_cols += [col]
          col_offset += 1

      # reduced row echelon form
      for row in xrange(num_used_prime):
        # lowest set bit
        mask = bit_fields[row] & -bit_fields[row]
        for up_row in xrange(row):
          if bit_fields[up_row] & mask:
            bit_fields[up_row] ^= bit_fields[row]

      # check for non-trivial congruencies
      for col in null_cols:
        all_vec, (lh, rh, rA) = smooth[col]
        lhs = lh   # sieved values (left hand side)
        rhs = [rh] # sieved values - n (right hand side)
        rAs = [rA] # root_As (cofactor of lhs)
        i = 0
        for field in bit_fields:
          if field & masks[col]:
            vec, (lh, rh, rA) = smooth[i]
            lhs += list(all_vec & vec) + lh
            all_vec ^= vec
            rhs += [rh]
            rAs += [rA]
          i += 1

        factor = my_math.gcd(my_math.list_prod(rAs)*my_math.list_prod(lhs) - my_math.list_prod(rhs), n)
        if 1 < factor < n:
          break
      else:
        if verbose:
          print 'none found.'
        continue
      break

  if verbose:
    print 'factors found:'
    print factor, 'x', n/factor
    print 'time elapsed: %f seconds'%(time.time()-time1)
  return factor

if __name__ == "__main__":
  import argparse
  parser = argparse.ArgumentParser(description='Uses a MPQS to factor a composite number')
  parser.add_argument('composite', metavar='number_to_factor', type=long, help='the composite number to factor')
  parser.add_argument('--verbose', dest='verbose', action='store_true', help="enable verbose output")
  args = parser.parse_args()

  if args.verbose:
    mpqs(args.composite, args.verbose)
  else:
    time1 = time.time()
    print mpqs(args.composite)
    print 'time elapsed: %f seconds'%(time.time()-time1)

my_math.py取自与相同的帖子mpqs.py,但是我也添加了我在寻找良好素数之间最大差距的答案中使用的无界素数发生器。

my_math.py

# primes less than 212
small_primes = [
    2,  3,  5,  7, 11, 13, 17, 19, 23, 29, 31, 37,
   41, 43, 47, 53, 59, 61, 67, 71, 73, 79, 83, 89,
   97,101,103,107,109,113,127,131,137,139,149,151,
  157,163,167,173,179,181,191,193,197,199,211]

# pre-calced sieve of eratosthenes for n = 2, 3, 5, 7
indices = [
    1, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47,
   53, 59, 61, 67, 71, 73, 79, 83, 89, 97,101,103,
  107,109,113,121,127,131,137,139,143,149,151,157,
  163,167,169,173,179,181,187,191,193,197,199,209]

# distances between sieve values
offsets = [
  10, 2, 4, 2, 4, 6, 2, 6, 4, 2, 4, 6,
   6, 2, 6, 4, 2, 6, 4, 6, 8, 4, 2, 4,
   2, 4, 8, 6, 4, 6, 2, 4, 6, 2, 6, 6,
   4, 2, 4, 6, 2, 6, 4, 2, 4, 2,10, 2]

# tabulated, mod 105
dindices =[
  0,10, 2, 0, 4, 0, 0, 0, 8, 0, 0, 2, 0, 4, 0,
  0, 6, 2, 0, 4, 0, 0, 4, 6, 0, 0, 6, 0, 0, 2,
  0, 6, 2, 0, 4, 0, 0, 4, 6, 0, 0, 2, 0, 4, 2,
  0, 6, 6, 0, 0, 0, 0, 6, 6, 0, 0, 0, 0, 4, 2,
  0, 6, 2, 0, 4, 0, 0, 4, 6, 0, 0, 2, 0, 6, 2,
  0, 6, 0, 0, 4, 0, 0, 4, 6, 0, 0, 2, 0, 4, 8,
  0, 0, 2, 0,10, 0, 0, 4, 0, 0, 0, 2, 0, 4, 2]

max_int = 2147483647


# returns the index of x in a sorted list a
# or the index of the next larger item if x is not present
# i.e. the proper insertion point for x in a
def binary_search(a, x):
  s = 0
  e = len(a)
  m = e >> 1
  while m != e:
    if a[m] < x:
      s = m
      m = (s + e + 1) >> 1
    else:
      e = m
      m = (s + e) >> 1
  return m


# divide and conquer list product
def list_prod(a):
  size = len(a)
  if size == 1:
    return a[0]
  return list_prod(a[:size>>1]) * list_prod(a[size>>1:])


# greatest common divisor of a and b
def gcd(a, b):
  while b:
    a, b = b, a%b
  return a


# extended gcd
def ext_gcd(a, m):
  a = int(a%m)
  x, u = 0, 1
  while a:
    x, u = u, x - (m/a)*u
    m, a = a, m%a
  return (m, x, u)


# legendre symbol (a|m)
# note: returns m-1 if a is a non-residue, instead of -1
def legendre(a, m):
  return pow(a, (m-1) >> 1, m)


# modular inverse of a mod m
def mod_inv(a, m):
  return ext_gcd(a, m)[1]


# modular sqrt(n) mod p
# p must be prime
def mod_sqrt(n, p):
  a = n%p
  if p%4 == 3:
    return pow(a, (p+1) >> 2, p)
  elif p%8 == 5:
    v = pow(a << 1, (p-5) >> 3, p)
    i = ((a*v*v << 1) % p) - 1
    return (a*v*i)%p
  elif p%8 == 1:
    # Shank's method
    q = p-1
    e = 0
    while q&1 == 0:
      e += 1
      q >>= 1

    n = 2
    while legendre(n, p) != p-1:
      n += 1

    w = pow(a, q, p)
    x = pow(a, (q+1) >> 1, p)
    y = pow(n, q, p)
    r = e
    while True:
      if w == 1:
        return x

      v = w
      k = 0
      while v != 1 and k+1 < r:
        v = (v*v)%p
        k += 1

      if k == 0:
        return x

      d = pow(y, 1 << (r-k-1), p)
      x = (x*d)%p
      y = (d*d)%p
      w = (w*y)%p
      r = k
  else: # p == 2
    return a


#integer sqrt of n
def isqrt(n):
  c = n*4/3
  d = c.bit_length()

  a = d>>1
  if d&1:
    x = 1 << a
    y = (x + (n >> a)) >> 1
  else:
    x = (3 << a) >> 2
    y = (x + (c >> a)) >> 1

  if x != y:
    x = y
    y = (x + n/x) >> 1
    while y < x:
      x = y
      y = (x + n/x) >> 1
  return x


# integer cbrt of n
def icbrt(n):
  d = n.bit_length()

  if d%3 == 2:
    x = 3 << d/3-1
  else:
    x = 1 << d/3

  y = (2*x + n/(x*x))/3
  if x != y:
    x = y
    y = (2*x + n/(x*x))/3
    while y < x:
      x = y
      y = (2*x + n/(x*x))/3
  return x


# strong probable prime
def is_sprp(n, b=2):
  if n < 2: return False
  d = n-1
  s = 0
  while d&1 == 0:
    s += 1
    d >>= 1

  x = pow(b, d, n)
  if x == 1 or x == n-1:
    return True

  for r in xrange(1, s):
    x = (x * x)%n
    if x == 1:
      return False
    elif x == n-1:
      return True

  return False


# lucas probable prime
# assumes D = 1 (mod 4), (D|n) = -1
def is_lucas_prp(n, D):
  P = 1
  Q = (1-D) >> 2

  # n+1 = 2**r*s where s is odd
  s = n+1
  r = 0
  while s&1 == 0:
    r += 1
    s >>= 1

  # calculate the bit reversal of (odd) s
  # e.g. 19 (10011) <=> 25 (11001)
  t = 0
  while s:
    if s&1:
      t += 1
      s -= 1
    else:
      t <<= 1
      s >>= 1

  # use the same bit reversal process to calculate the sth Lucas number
  # keep track of q = Q**n as we go
  U = 0
  V = 2
  q = 1
  # mod_inv(2, n)
  inv_2 = (n+1) >> 1
  while t:
    if t&1:
      # U, V of n+1
      U, V = ((U + V) * inv_2)%n, ((D*U + V) * inv_2)%n
      q = (q * Q)%n
      t -= 1
    else:
      # U, V of n*2
      U, V = (U * V)%n, (V * V - 2 * q)%n
      q = (q * q)%n
      t >>= 1

  # double s until we have the 2**r*sth Lucas number
  while r:
    U, V = (U * V)%n, (V * V - 2 * q)%n
    q = (q * q)%n
    r -= 1

  # primality check
  # if n is prime, n divides the n+1st Lucas number, given the assumptions
  return U == 0


## Baillie-PSW ##
# this is technically a probabalistic test, but there are no known pseudoprimes
def is_bpsw(n):
  if not is_sprp(n, 2): return False

  # idea shamelessly stolen from Mathmatica's PrimeQ
  # if n is a 2-sprp and a 3-sprp, n is necessarily square-free
  if not is_sprp(n, 3): return False

  a = 5
  s = 2
  # if n is a perfect square, this will never terminate
  while legendre(a, n) != n-1:
    s = -s
    a = s-a
  return is_lucas_prp(n, a)


# an 'almost certain' primality check
def is_prime(n):
  if n < 212:
    m = binary_search(small_primes, n)
    return n == small_primes[m]

  for p in small_primes:
    if n%p == 0:
      return False

  # if n is a 32-bit integer, perform full trial division
  if n <= max_int:
    p = 211
    while p*p < n:
      for o in offsets:
        p += o
        if n%p == 0:
          return False
    return True

  return is_bpsw(n)


# next prime strictly larger than n
def next_prime(n):
  if n < 2:
    return 2

  # first odd larger than n
  n = (n + 1) | 1
  if n < 212:
    m = binary_search(small_primes, n)
    return small_primes[m]

  # find our position in the sieve rotation via binary search
  x = int(n%210)
  m = binary_search(indices, x)
  i = int(n + (indices[m] - x))

  # adjust offsets
  offs = offsets[m:] + offsets[:m]
  while True:
    for o in offs:
      if is_prime(i):
        return i
      i += o


# an infinite prime number generator
def primes(start = 0):
  for n in small_primes[start:]: yield n
  pg = primes(6)
  p = pg.next()
  q = p*p
  sieve = {221: 13, 253: 11}
  n = 211
  while True:
    for o in offsets:
      n += o
      stp = sieve.pop(n, 0)
      if stp:
        nxt = n/stp
        nxt += dindices[nxt%105]
        while nxt*stp in sieve:
          nxt += dindices[nxt%105]
        sieve[nxt*stp] = stp
      elif n < q:
        yield n
      else:
        sieve[q + dindices[p%105]*p] = p
        p = pg.next()
        q = p*p


# true if n is a prime power > 0
def is_prime_power(n):
  if n > 1:
    for p in small_primes:
      if n%p == 0:
        n /= p
        while n%p == 0: n /= p
        return n == 1

    r = isqrt(n)
    if r*r == n:
      return is_prime_power(r)

    s = icbrt(n)
    if s*s*s == n:
      return is_prime_power(s)

    p = 211
    while p*p < r:
      for o in offsets:
        p += o
        if n%p == 0:
          n /= p
          while n%p == 0: n /= p
          return n == 1

    if n <= max_int:
      while p*p < n:
        for o in offsets:
          p += o
          if n%p == 0:
            return False
      return True

    return is_bpsw(n)
  return False

2

Python 2 + Primefac 1.1

我没有Raspberry Pi可以对其进行测试。

from primefac import primefac

def HP(n):
    factors = list(primefac(n))

    #print n, factors

    if len(factors) == 1 and n in factors:
        return n

    n = ""
    for f in sorted(factors):
        n += str(f)
    return HP(int(n))

在线尝试

primefac函数返回的所有主要因子的列表n。在其定义中,它称为isprime(n),它结合了试验除法,欧拉方法和Miller-Rabin素数检验的组合。我建议下载软件包并查看源代码。

我尝试使用n = n * 10 ** int(floor(log10(f))+1) + f而不是字符串连接,但是速度慢得多。


pip install primefac尽管对65和80似乎由于fork在后台运行而无法在Windows上运行,但它为我工作。
primo

查看来源primefac非常有趣,因为其中有很多注释,TODO或带有find out why this is throwing errors
mbomb007

我也是。作者实际上使用了我的mpqs!...稍作修改。551行# This occasionally throws IndexErrors.是的,因为他取消了检查,检查是否存在比使用的素数更多的平滑度。
primo

你应该帮他 :)
mbomb007'1

解决完这个挑战后,我可能会联系他,我打算稍微优化一下mpq(一定要打败mathmatica,对吗?)。
primo

2

C#

using System;
using System.Linq;

public class Program
{
    public static void Main(string[] args) {

        Console.Write("Enter Number: ");

        Int64 n = Convert.ToInt64(Console.ReadLine());

        Console.WriteLine("Start Time: " + DateTime.Now.ToString("HH:mm:ss.ffffff"));
        Console.WriteLine("Number, Factors");

        HomePrime(n);

        Console.WriteLine("End Time: " + DateTime.Now.ToString("HH:mm:ss.ffffff"));
        Console.ReadLine();
    }

    public static void HomePrime(Int64 num) {
        string s = FindFactors(num);
        if (CheckPrime(num,s) == true) {
            Console.WriteLine("{0} is prime", num);
        } else {
            Console.WriteLine("{0}, {1}", num, s);
            HomePrime(Convert.ToInt64(RemSp(s)));
        }
    }

    public static string FindFactors(Int64 num) {
        Int64 n, r, t = num;
        string f = "";
        for (n = num; n >= 2; n--) {
            r = CalcP(n, t);
            if (r != 0) {
                f = f + " " + r.ToString();
                t = n / r;
                n = n / r + 1;
            }
        }
        return f;
    }

    public static Int64 CalcP(Int64 num, Int64 tot) {
        for (Int64 i = 2; i <= tot; i++) {
            if (num % i == 0) {
                return i;
            } 
        }
        return 0;
    }

    public static string RemSp(string str) {
        return new string(str.ToCharArray().Where(c => !Char.IsWhiteSpace(c)).ToArray());
    }

    public static bool CheckPrime(Int64 num, string s) {
        if (s == "") {
            return false;
        } else if (num == Convert.ToInt64(RemSp(s))) {
            return true;
        }
        return false;
    }

}

这是先前代码的优化版本,其中删除了一些不必要的冗余部分。

输出(在我的i7笔记本电脑上):

Enter Number: 16
Start Time: 18:09:51.636445
Number, Factors
16,  2 2 2 2
2222,  2 11 101
211101,  3 11 6397
3116397,  3 163 6373
31636373 is prime
End Time: 18:09:51.637954

在线测试


我认为,不允许使用具有预定素数/值的数组进行创建,因为这是一个标准漏洞。
P. Ktinos

@ P.Ktinos我也这么认为...无论如何它太大了。
马里奥(Mario)

1

Perl + ntheory,HP(80)在PC上仅需0.35s

暂无Raspberry Pi。

use ntheory ":all";
use feature "say";
sub hp {
  my $n = shift;
  while (!is_prime($n)) {
    $n = join "",factor($n);
  }
  $n;
}
say hp($_) for (16,20,64,65,80);

素数测试是ES BPSW,再加上一个随机数较大的MR。在此大小下,我们可以使用is_provable_prime(n-1和/或ECPP),但速度没有明显差异,但是对于> 300位数的值,它会发生变化,没有任何实际好处。分解包括试算,功率,Rho-Brent,P-1,SQUFOF,ECM,QS,具体取决于大小。

对于这些输入,它的运行速度与OEIS网站上Charles'Pari / GP代码的运行速度相同。ntheory对小数的分解速度更快,我的P-1和ECM很好,但是QS不好,所以我希望Pari在某个时候更快。


1
我发现P-1发现的任何因素也早已被ECM发现,因此我删除了它(威廉姆斯P + 1也是如此)。也许我会尝试添加SQUFOF。辉煌的图书馆,顺便说一句。
primo

1
另外,use feature "say";
primo
By using our site, you acknowledge that you have read and understand our Cookie Policy and Privacy Policy.
Licensed under cc by-sa 3.0 with attribution required.