python-大整数相乘-分治法

暴力相加

下面展示一些 内联代码片。

def simple_multi(stra, strb):
    aa = list(stra)
    bb = list(strb)
    lena = len(stra)
    lenb = len(strb)
    result = [0 for i in range(lena+lenb)]
    for i in range(lena):
        for j in range(lenb):
            result[lena-i-1+lenb-j-1] += int(aa[i])*int(bb[j])
    for i in range(len(result)-1):
        if result[i] >= 10:
            result[i+1] += result[i]//10
            result[i] = result[i] % 10
    return list2str(result[::-1])

分治法

下面展示一些 内联代码片。

# 首先将x，y分别拆开成为两部分，可得x1，x0，y1，y0。他们的关系如下：
# x = x1 * 10
# m + x0；
# y = y1 * 10
# m + y0。其中m为正整数，m < n，且x0，y0
# 小于
# 10
# m。
# 那么
# xy = (x1 * 10m + x0)(y1 * 10
# m + y0)
# =z2 * 102
# m + z1 * 10
# m + z0，其中：
# 此步骤共需4次乘法，
def fenzhi_multi(x1, x2):
    l1 = len(x1)
    l2 = len(x2)

    if x1 == '0' or x2 == '0' or x1 == '' or x2 == '':
        return '0'
    elif l1 == 1 or l2 == 1:
        val = str(int(x1) * int(x2))
        return val
    else:
        cut_len = int(min(l1, l2) / 2)
        cut_point1 = l1 - cut_len
        cut_point2 = l2 - cut_len
        x1_0 = x1[:cut_point1]
        x1_1 = x1[cut_point1:]

        x2_0 = x2[:cut_point2]
        x2_1 = x2[cut_point2:]

        ac = fenzhi_multi(x1_0, x2_0)
        bd = fenzhi_multi(x1_1, x2_1)
        ad = fenzhi_multi(x1_0, x2_1)
        bc = fenzhi_multi(x1_1, x2_0)

        ac_10n = ac + '0' * 2 * cut_len
        adbc = add(ad, bc) + '0' * cut_len
        res = add(add(ac_10n, adbc), bd)

        return res

改进的分治法

# 但是由Karatsuba改进以后仅需要3次乘法。因为：
# z1 = x1 * y0 + x0 * y1
# z1 = (x1 + x0) * (y1 + y0) - x1 * y1 - x0 * y0，
def changed_fenzhi_multi(x1, x2):

    if x1 == '0' or x2 == '0' or x1 == '' or x2 == '':
        return '0'

    sign = 1
    if x1[0] == '-':
        sign *= -1
        x1 = x1[1:]
    if x2[0] == '-':
        sign *= -1
        x2 = x2[1:]

    l1 = len(x1)
    l2 = len(x2)

    if l1 == 1 or l2 == 1:
        val = str(int(x1) * int(x2))
        if sign == -1:
            val = val[::-1] + '-'
            val = val[::-1]
        return val
    else:
        cut_len = int(min(l1, l2)/2)
        cut_point1 = l1 - cut_len
        cut_point2 = l2 - cut_len
        x1_0 = x1[:cut_point1]
        x1_1 = x1[cut_point1:]

        x2_0 = x2[:cut_point2]
        x2_1 = x2[cut_point2:]

        ac = changed_fenzhi_multi(x1_0, x2_0)
        bd = changed_fenzhi_multi(x1_1, x2_1)

        a_add_b = add(x1_0, x1_1)
        d_add_c = add(x2_1, x2_0)

        abcd = changed_fenzhi_multi(a_add_b, d_add_c)

        ac_10n = ac + '0'*2*cut_len
        mid = sub1(sub1(abcd, ac), bd)
        mid = mid + '0' * cut_len

        val = add(add(ac_10n, mid), bd)
        if sign == -1:
            val = val[::-1] + '-'
            val = val[::-1]
        return val

时间	普通大整数乘法	分治大整数乘法	改进的分治大整数乘法
10位	0.000044	0.0003371238708496094	0.000533342361450
100位	0.002929210662841	0.03171944618225098	0.04110884666442871
1000位	0.3304471969604492	3.543729782104492	1.418264627456665
10000位	33.64151573181152	300.8758816719055	58.203367471694946
100000位	5678.60	30478.068567276	56603.10471582413
			56603.10471582413

全部源码

import time
import random


def add(n1, n2):
    if n1[0] == '-' and n2[0] != '-':
        return sub1(n2, n1[1:])
    elif n1[0] != '-' and n2[0] == '-':
        return sub1(n1, n2[1:])
    elif n1[0] == '-' and n2[0] == '-':
        temp = add(n1[1:], n2[1:])
        temp = temp[::-1] + '-'
        temp = temp[::-1]
        return temp

    n1 = n1[::-1]
    n2 = n2[::-1]
    # 补齐到和的最大位数
    if len(n1)<len(n2):
        n2 += '0'
        n1 += '0'*(len(n2)-len(n1))
    else:
        n1 += '0'
        n2 += '0'*(len(n1)-len(n2))
    carry,sum = 0,''
    for i in range(len(n1)):
        cur_sum = int(n1[i]) + int(n2[i]) + carry
        carry = cur_sum // 10
        sum += str(cur_sum % 10)
    sum = sum[::-1]
    # 去掉前面的零
    while len(sum)>1 and sum[0]=='0':
        sum = sum[1:]
    return sum


def sub1(n1, n2):

    if n1 == '' and n2 == '':
        return '0'
    elif n1 == '':
        return sub1('0', n2)
    elif n2 == '':
        return n1

    if n1[0] != '-' and n2[0] == '-':
        return add(n1, n2[1:])
    elif n1[0] == '-' and n2[0] == '-':
        return sub1(n2[1:], n1[1:])
    elif n1[0] == '-' and n2[0] != '-':
        temp = add(n1[1:], n2)
        temp = temp[::-1] + '-'
        temp = temp[::-1]
        return temp

    n1 = n1[::-1]
    n2 = n2[::-1]

    if len(n1) < len(n2):
        n1 += '0' * (len(n2) -len(n1))
    else:
        n2 += '0' * (len(n1) -len(n2))
    carry, sub = 0, ''
    for i in range(len(n1)):
        cur_sub = int(n1[i]) - carry - int(n2[i])
        carry = 0
        if cur_sub < 0:
            carry = 1
            cur_sub += 10
        sub += str(cur_sub)

    if carry == 1:
        carry, sub = 0, ''
        for i in range(len(n1)):
            cur_sub = int(n2[i]) - carry - int(n1[i])
            carry = 0
            if cur_sub < 0:
                carry = 1
                cur_sub += 10
            sub += str(cur_sub)

        sub += '-'

        sub = sub[::-1]
        while len(sub) > 1 and sub[1] == '0':
            sub = sub[0] + sub[2:]

    else:
        sub = sub[::-1]
        while len(sub)>1 and sub[0] == '0':
            sub = sub[1:]

    return sub


def list2str(li):
    while li[0] == 0:
        del li[0]
    res = ''
    for i in li:
        res += str(i)
    return res


def simple_multi(stra, strb):
    aa = list(stra)
    bb = list(strb)
    lena = len(stra)
    lenb = len(strb)
    result = [0 for i in range(lena+lenb)]
    for i in range(lena):
        for j in range(lenb):
            result[lena-i-1+lenb-j-1] += int(aa[i])*int(bb[j])
    for i in range(len(result)-1):
        if result[i] >= 10:
            result[i+1] += result[i]//10
            result[i] = result[i] % 10
    return list2str(result[::-1])


def number_generate(n):
    num1 = ''
    num2 = ''
    for i in range(n):
        num1 += str(random.randint(0, 9))
        num2 += str(random.randint(0, 9))
    return num1, num2


# 首先将x，y分别拆开成为两部分，可得x1，x0，y1，y0。他们的关系如下：
# x = x1 * 10
# m + x0；
# y = y1 * 10
# m + y0。其中m为正整数，m < n，且x0，y0
# 小于
# 10
# m。
# 那么
# xy = (x1 * 10m + x0)(y1 * 10
# m + y0)
# =z2 * 102
# m + z1 * 10
# m + z0，其中：
# 此步骤共需4次乘法，但是由Karatsuba改进以后仅需要3次乘法。因为：
# z1 = x1 * y0 + x0 * y1
# z1 = (x1 + x0) * (y1 + y0) - x1 * y1 - x0 * y0，
def fenzhi_multi(x1, x2):
    l1 = len(x1)
    l2 = len(x2)

    if x1 == '0' or x2 == '0' or x1 == '' or x2 == '':
        return '0'
    elif l1 == 1 or l2 == 1:
        val = str(int(x1) * int(x2))
        return val
    else:
        cut_len = int(min(l1, l2) / 2)
        cut_point1 = l1 - cut_len
        cut_point2 = l2 - cut_len
        x1_0 = x1[:cut_point1]
        x1_1 = x1[cut_point1:]

        x2_0 = x2[:cut_point2]
        x2_1 = x2[cut_point2:]

        ac = fenzhi_multi(x1_0, x2_0)
        bd = fenzhi_multi(x1_1, x2_1)
        ad = fenzhi_multi(x1_0, x2_1)
        bc = fenzhi_multi(x1_1, x2_0)

        ac_10n = ac + '0' * 2 * cut_len
        adbc = add(ad, bc) + '0' * cut_len
        res = add(add(ac_10n, adbc), bd)

        return res


def changed_fenzhi_multi(x1, x2):

    if x1 == '0' or x2 == '0' or x1 == '' or x2 == '':
        return '0'

    sign = 1
    if x1[0] == '-':
        sign *= -1
        x1 = x1[1:]
    if x2[0] == '-':
        sign *= -1
        x2 = x2[1:]

    l1 = len(x1)
    l2 = len(x2)

    if l1 == 1 or l2 == 1:
        val = str(int(x1) * int(x2))
        if sign == -1:
            val = val[::-1] + '-'
            val = val[::-1]
        return val
    else:
        cut_len = int(min(l1, l2)/2)
        cut_point1 = l1 - cut_len
        cut_point2 = l2 - cut_len
        x1_0 = x1[:cut_point1]
        x1_1 = x1[cut_point1:]

        x2_0 = x2[:cut_point2]
        x2_1 = x2[cut_point2:]

        ac = changed_fenzhi_multi(x1_0, x2_0)
        bd = changed_fenzhi_multi(x1_1, x2_1)

        a_add_b = add(x1_0, x1_1)
        d_add_c = add(x2_1, x2_0)

        abcd = changed_fenzhi_multi(a_add_b, d_add_c)

        ac_10n = ac + '0'*2*cut_len
        mid = sub1(sub1(abcd, ac), bd)
        mid = mid + '0' * cut_len

        val = add(add(ac_10n, mid), bd)
        if sign == -1:
            val = val[::-1] + '-'
            val = val[::-1]
        return val


if __name__ == '__main__':
    for i in [10, 100, 1000, 10000, 100000, 1000000, 10000000, 100000000]:
        a, b = number_generate(i)
        print(len(a))

        start_time = time.time()

        res = fenzhi_multi(a, b)

        end_time = time.time()

        print("fenzhi:time:", end_time - start_time)
        print('f', res)
        print("s", simple_multi(a, b))
        print('c', changed_fenzhi_multi(a, b))

折线图

import matplotlib.pyplot as plt

names = ['10', '100', '1000', '10000', '100000']

x1 = range(len(names))

y1 = [0.000044, 0.002929, 0.330447, 33.641515, 5678.60]

y2 = [0.000337, 0.031719, 3.543729, 300.875881, 3047.06]

y3 = [0.000533, 0.041108, 1.418264, 58.203367, 1902.112911]


# a normal scatter plot with default features
plt.plot(x1, y1, c='r', alpha=0.8, marker='o', label='simple')

plt.plot(x1, y2, c='g', alpha=0.8, marker='*', label='divide-and-conquer')

plt.plot(x1, y3, c='b', alpha=0.8, marker='+', label='changed divide-and-conquer')

plt.xticks(x1, names, rotation=45)
plt.xlabel('Number of digits')
plt.ylabel('Time')
plt.title('Multiplication of large integers')

plt.legend()
plt.show()