//这是Java中的递归解决方案 //从main product(a,1,0)调用如下;

public static double product(double[] a, double fwdprod, int index){
    double revprod = 1;
    if (index < a.length){
        revprod = product2(a, fwdprod*a[index], index+1);
        double cur = a[index];
        a[index] = fwdprod * revprod;
        revprod *= cur;
    }
    return revprod;
}

还有一个O(N^(3/2))非最优解。不过，这很有趣。

首先预处理大小为N^0.5的每个部分乘法(这在O(N)时间复杂度中完成)。然后，计算每个数字的其他值的倍数可以在2*O(N^0.5)时间内完成(为什么?因为您只需要将其他((N^0.5) - 1)数字的最后一个元素相乘，并将结果与属于当前数字组的((N^0.5) - 1)数字相乘。对每一个数都这样做，可以得到O(N^(3/2))时间。

例子:

4, 6, 7, 2, 3, 1, 9, 5, 8

部分结果: 4*6*7 = 168 2*3*1 = 6 9*5*8 = 360

要计算3的值，需要将其他组的值乘以168*360，然后乘以2*1。

这是ptyhon版本

  # This solution use O(n) time and O(n) space
  def productExceptSelf(self, nums):
    """
    :type nums: List[int]
    :rtype: List[int]
    """
    N = len(nums)
    if N == 0: return

    # Initialzie list of 1, size N
    l_prods, r_prods = [1]*N, [1]*N

    for i in range(1, N):
      l_prods[i] = l_prods[i-1] * nums[i-1]

    for i in reversed(range(N-1)):
      r_prods[i] = r_prods[i+1] * nums[i+1]

    result = [x*y for x,y in zip(l_prods,r_prods)]
    return result

  # This solution use O(n) time and O(1) space
  def productExceptSelfSpaceOptimized(self, nums):
    """
    :type nums: List[int]
    :rtype: List[int]
    """
    N = len(nums)
    if N == 0: return

    # Initialzie list of 1, size N
    result = [1]*N

    for i in range(1, N):
      result[i] = result[i-1] * nums[i-1]

    r_prod = 1
    for i in reversed(range(N)):
      result[i] *= r_prod
      r_prod *= nums[i]

    return result

{-
Recursive solution using sqrt(n) subsets. Runs in O(n).

Recursively computes the solution on sqrt(n) subsets of size sqrt(n). 
Then recurses on the product sum of each subset.
Then for each element in each subset, it computes the product with
the product sum of all other products.
Then flattens all subsets.

Recurrence on the run time is T(n) = sqrt(n)*T(sqrt(n)) + T(sqrt(n)) + n

Suppose that T(n) ≤ cn in O(n).

T(n) = sqrt(n)*T(sqrt(n)) + T(sqrt(n)) + n
    ≤ sqrt(n)*c*sqrt(n) + c*sqrt(n) + n
    ≤ c*n + c*sqrt(n) + n
    ≤ (2c+1)*n
    ∈ O(n)

Note that ceiling(sqrt(n)) can be computed using a binary search 
and O(logn) iterations, if the sqrt instruction is not permitted.
-}

otherProducts [] = []
otherProducts [x] = [1]
otherProducts [x,y] = [y,x]
otherProducts a = foldl' (++) [] $ zipWith (\s p -> map (*p) s) solvedSubsets subsetOtherProducts
    where 
      n = length a

      -- Subset size. Require that 1 < s < n.
      s = ceiling $ sqrt $ fromIntegral n

      solvedSubsets = map otherProducts subsets
      subsetOtherProducts = otherProducts $ map product subsets

      subsets = reverse $ loop a []
          where loop [] acc = acc
                loop a acc = loop (drop s a) ((take s a):acc)

鬼鬼祟祟地绕过“不划分”规则:

sum = 0.0
for i in range(a):
  sum += log(a[i])

for i in range(a):
  output[i] = exp(sum - log(a[i]))

最近有人问我这个问题，虽然我不能得到O(N)，但我有一个不同的方法(不幸的是O(N²))，但我想无论如何都要分享。

首先转换为列表<Integer>。

遍历原始数组array.length()次。

使用while循环乘下一组所需的数字:

while (temp < list.size() - 1) {
    res *= list.get(temp);
    temp++;
}

然后将res添加到一个新数组(当然，您已经在前面声明了)，然后将数组[i]的值添加到List，依此类推。

我知道这不会有太大的用处，但这是我在面试的压力下想到的:)

    int[] array = new int[]{1, 2, 3, 4, 5};
    List<Integer> list = Arrays.stream(array).boxed().collect(Collectors.toList());
    int[] newarray = new int[array.length];
    int res = 1;
    for (int i = 0; i < array.length; i++) {
        int temp = i;
        while (temp < list.size() - 1) {
            res *= list.get(temp);
            temp++;
        }
        newarray[i] = res;
        list.add(array[i]);
        res = 1;
    }

输出:[24,120,60,40,30]