我需要一个算法来生成所有可能的正数分区,我想出了一个(作为答案发布),但它是指数时间。
算法应该返回所有可能的方式,数字可以表示为小于或等于其自身的正数之和。例如,对于数字 5 ,结果将是:
所以我的问题是:对此有更有效的算法吗?
编辑:问题的标题是“数字的总和分解”,因为我真的不知道这叫什么。 ShreevatsaR pointed out他们被称为“分区”,所以我相应地编辑了问题标题。
答案 0 :(得分:26)
它被称为Partitions。 [另见维基百科:Partition (number theory)。]
分区数p(n)呈指数级增长,因此生成所有分区所需的任何操作都必须采用指数时间。
也就是说,你可以比你的代码更好。 this之后的Python Algorithms and Data Structures中的David Eppstein或其更新版本。
答案 1 :(得分:20)
这是我在Python中的解决方案(指数时间):
q = { 1: [[1]] }
def decompose(n):
try:
return q[n]
except:
pass
result = [[n]]
for i in range(1, n):
a = n-i
R = decompose(i)
for r in R:
if r[0] <= a:
result.append([a] + r)
q[n] = result
return result
>>> decompose(5)
[[5], [4, 1], [3, 2], [3, 1, 1], [2, 2, 1], [2, 1, 1, 1], [1, 1, 1, 1, 1]]
答案 2 :(得分:5)
当你问更高效的算法时,我不知道要比较哪个。但是这里有一种直接编写的算法(Erlang):
-module(partitions).
-export([partitions/1]).
partitions(N) -> partitions(N, N).
partitions(N, Max) when N > 0 ->
[[X | P]
|| X <- lists:seq(min(N, Max), 1, -1),
P <- partitions(N - X, X)];
partitions(0, _) -> [[]];
partitions(_, _) -> [].
它在时间上是指数的(与Can Berk Güder's solution in Python相同)并且在堆栈空间中是线性的。但是使用相同的技巧,memoization,你可以通过节省一些内存和更少的指数来实现重大改进。 (N = 50时快十倍)
mp(N) ->
lists:foreach(fun (X) -> put(X, undefined) end,
lists:seq(1, N)), % clean up process dictionary for sure
mp(N, N).
mp(N, Max) when N > 0 ->
case get(N) of
undefined -> R = mp(N, 1, Max, []), put(N, R), R;
[[Max | _] | _] = L -> L;
[[X | _] | _] = L ->
R = mp(N, X + 1, Max, L), put(N, R), R
end;
mp(0, _) -> [[]];
mp(_, _) -> [].
mp(_, X, Max, R) when X > Max -> R;
mp(N, X, Max, R) ->
mp(N, X + 1, Max, prepend(X, mp(N - X, X), R)).
prepend(_, [], R) -> R;
prepend(X, [H | T], R) -> prepend(X, T, [[X | H] | R]).
无论如何,您应该根据您的语言和目的进行基准测试。
答案 3 :(得分:1)
这里有一个更冗长的方式(这是我在知道术语&#34;分区&#34;之前所做的,这使我能够进行谷歌搜索):
def magic_chunker (remainder, chunkSet, prevChunkSet, chunkSets):
if remainder > 0:
if prevChunkSet and (len(prevChunkSet) > len(chunkSet)): # counting down from previous
# make a chunk that is one less than relevant one in the prevChunkSet
position = len(chunkSet)
chunk = prevChunkSet[position] - 1
prevChunkSet = [] # clear prevChunkSet, no longer need to reference it
else: # begins a new countdown;
if chunkSet and (remainder > chunkSet[-1]): # no need to do iterations any greater than last chunk in this set
chunk = chunkSet[-1]
else: # i.e. remainder is less than or equal to last chunk in this set
chunk = remainder #else use the whole remainder for this chunk
chunkSet.append(chunk)
remainder -= chunk
magic_chunker(remainder, chunkSet, prevChunkSet, chunkSets)
else: #i.e. remainder==0
chunkSets.append(list(chunkSet)) #save completed partition
prevChunkSet = list(chunkSet)
if chunkSet[-1] > 1: # if the finalchunk was > 1, do further recursion
remainder = chunkSet.pop() #remove last member, and use it as remainder
magic_chunker(remainder, chunkSet, prevChunkSet, chunkSets)
else: # last chunk is 1
if chunkSet[0]==1: #the partition started with 1, we know we're finished
return chunkSets
else: #i.e. still more chunking to go
# clear back to last chunk greater than 1
while chunkSet[-1]==1:
remainder += chunkSet.pop()
remainder += chunkSet.pop()
magic_chunker(remainder, chunkSet, prevChunkSet, chunkSets)
partitions = []
magic_chunker(10, [], [], partitions)
print partitions
>> [[10], [9, 1], [8, 2], [8, 1, 1], [7, 3], [7, 2, 1], [7, 1, 1, 1], [6, 4], [6, 3, 1], [6, 2, 2], [6, 2, 1, 1], [6, 1, 1, 1, 1], [5, 5], [5, 4, 1], [5, 3, 2], [5, 3, 1, 1], [5, 2, 2, 1], [5, 2, 1, 1, 1], [5, 1, 1, 1, 1, 1], [4, 4, 2], [4, 4, 1, 1], [4, 3, 3], [4, 3, 2, 1], [4, 3, 1, 1, 1], [4, 2, 2, 2], [4, 2, 2, 1, 1], [4, 2, 1, 1, 1, 1], [4, 1, 1, 1, 1, 1, 1], [3, 3, 3, 1], [3, 3, 2, 2], [3, 3, 2, 1, 1], [3, 3, 1, 1, 1, 1], [3, 2, 2, 2, 1], [3, 2, 2, 1, 1, 1], [3, 2, 1, 1, 1, 1, 1], [3, 1, 1, 1, 1, 1, 1, 1], [2, 2, 2, 2, 2], [2, 2, 2, 2, 1, 1], [2, 2, 2, 1, 1, 1, 1], [2, 2, 1, 1, 1, 1, 1, 1], [2, 1, 1, 1, 1, 1, 1, 1, 1], [1, 1, 1, 1, 1, 1, 1, 1, 1, 1]]
答案 4 :(得分:0)
这是使用我在Haskell中编写的paramorphisms的解决方案。
import Numeric.Natural (Natural)
import Control.Monad (join)
import Data.List (nub)
import Data.Functor.Foldable (ListF (..), para)
partitions :: Natural -> [[Natural]]
partitions = para algebra
where algebra Nothing = []
algebra (Just (0,_)) = [[1]]
algebra (Just (_, past)) = (nub . (getAll =<<)) (fmap (1:) past)
getAll :: [Natural] -> [[Natural]]
getAll = fmap (dropWhile (==0) . sort) . subsets
where subsets xs = flip sumIndicesAt xs <$> indices xs
indices :: [Natural] -> [[Natural]]
indices = join . para algebra
where algebra Nil = []
algebra (Cons x (xs, [])) = [[x:xs]]
algebra (Cons x (xs, past)) = (:) <$> [x:xs,[]] <*> past
它绝对不是最有效的,但我觉得它非常优雅,而且肯定是有益的。
答案 5 :(得分:0)
static void printArray(int p[], int n){
for (int i = 0; i < n; i++)
System.out.print(p[i]+" ");
System.out.println();
}
// Function to generate all unique partitions of an integer
static void printAllUniqueParts(int n) {
int[] p = new int[n]; // An array to store a partition
int k = 0; // Index of last element in a partition
p[k] = n; // Initialize first partition as number itself
// This loop first prints current partition, then generates next
// partition. The loop stops when the current partition has all 1s
while (true) {
// print current partition
printArray(p, k + 1);
// Generate next partition
// Find the rightmost non-one value in p[]. Also, update the
// rem_val so that we know how much value can be accommodated
int rem_val = 0;
while (k >= 0 && p[k] == 1) {
rem_val += p[k];
k--;
}
// if k < 0, all the values are 1 so there are no more partitions
if (k < 0){
break;
}
// Decrease the p[k] found above and adjust the rem_val
p[k]--;
rem_val++;
while (rem_val > p[k]) {
p[k + 1] = p[k];
rem_val = rem_val - p[k];
k++;
}
p[k + 1] = rem_val;
k++;
}
}
public static void main(String[] args) {
System.out.println("All Unique Partitions of 5");
printAllUniqueParts(5);
System.out.println("All Unique Partitions of 7");
printAllUniqueParts(7);
System.out.println("All Unique Partitions of 9");
printAllUniqueParts(8);
}
答案 6 :(得分:0)
另一个Java解决方案。首先创建仅给定编号的第一个分区。然后,它进入while循环,该循环在最后一个创建的分区中找到大于1的最后一个数字。从该数字中将1移动到数组中的下一个数字。如果下一个数字最终与找到的数字相同,它将移至下一行。当最后创建的分区的第一个数字为1时,循环停止。之所以起作用,是因为在所有时间,所有分区中的数字始终以降序排列。
数字5的示例。首先创建第一个分区,即数字5。然后在最后一个分区中找到大于1的最后一个数字。由于我们的最后一个分区是数组[5,0,0,0,0]它在索引0处找到数字5。然后从5中取一个,并将其移至下一个位置。这就是我们获得分区[4,1,0,0,0]的方式。它再次进入循环。现在,它从4中取出1并向上移动,因此我们得到[3,2,0,0,0]。然后,同样的事情得到[3,1,1,0,0]。在下一次迭代中,我们得到[2,2,1,0,0]。现在,它从第二个2开始取一个,并尝试将其移至我们拥有1的索引2。它将跳至下一个索引,因为我们也将获得2,并且将有分区[2,1,2,0,0]只是最后一个的重复。相反,我们得到[2,1,1,1,0]。在最后一步中,我们到达[1,1,1,1,1]并存在循环,因为新分区的第一个数字是1。
private static List<int[]> getNumberPartitions(int n) {
ArrayList<int[]> result = new ArrayList<>();
int[] initial = new int[n];
initial[0] = n;
result.add(initial);
while (result.get(result.size() - 1)[0] > 1) {
int[] lastPartition = result.get(result.size() - 1);
int posOfLastNotOne = 0;
for(int k = lastPartition.length - 1; k >= 0; k--) {
if (lastPartition[k] > 1) {
posOfLastNotOne = k;
break;
}
}
int[] newPartition = new int[n];
for (int j = posOfLastNotOne + 1; j < lastPartition.length; j++) {
if (lastPartition[posOfLastNotOne] - 1 > lastPartition[j]) {
System.arraycopy(lastPartition, 0, newPartition, 0, lastPartition.length);
newPartition[posOfLastNotOne]--;
newPartition[j]++;
result.add(newPartition);
break;
}
}
}
return result;
}
答案 7 :(得分:0)
这是我的Rust实现(受Python Algorithms and Data Structures启发):
#[derive(Clone)]
struct PartitionIter {
pub n: u32,
partition: Vec<u32>,
last_not_one_index: usize,
started: bool,
finished: bool
}
impl PartitionIter {
pub fn new(n: u32) -> PartitionIter {
PartitionIter {
n,
partition: Vec::with_capacity(n as usize),
last_not_one_index: 0,
started: false,
finished: false,
}
}
}
impl Iterator for PartitionIter {
type Item = Vec<u32>;
fn next(&mut self) -> Option<Self::Item> {
if self.finished {
return None
}
if !self.started {
self.partition.push(self.n);
self.started = true;
return Some(self.partition.clone());
} else if self.n == 1 {
return None;
}
if self.partition[self.last_not_one_index] == 2 {
self.partition[self.last_not_one_index] = 1;
self.partition.push(1);
if self.last_not_one_index == 0 {
self.finished = true;
} else {
self.last_not_one_index -= 1;
}
return Some(self.partition.clone())
}
let replacement = self.partition[self.last_not_one_index] - 1;
let total_replaced = replacement + (self.partition.len() - self.last_not_one_index) as u32;
let reps = total_replaced / replacement;
let rest = total_replaced % replacement;
self.partition.drain(self.last_not_one_index..);
self.partition.extend_from_slice(&vec![replacement; reps as usize]);
if rest > 0 {
self.partition.push(rest);
}
self.last_not_one_index = self.partition.len() - (self.partition.last().cloned().unwrap() == 1) as usize - 1;
Some(self.partition.clone())
}
}
答案 8 :(得分:-1)
Java实现。可以从备忘录中受益。
public class Partition {
/**
* partition returns a list of int[] that represent all distinct partitions of n.
*/
public static List<int[]> partition(int n) {
List<Integer> partial = new ArrayList<Integer>();
List<int[]> partitions = new ArrayList<int[]>();
partition(n, partial, partitions);
return partitions;
}
/**
* If n=0, it copies the partial solution into the list of complete solutions.
* Else, for all values i less than or equal to n, put i in the partial solution and partition the remainder n-i.
*/
private static void partition(int n, List<Integer> partial, List<int[]> partitions) {
//System.out.println("partition " + n + ", partial solution: " + partial);
if (n == 0) {
// Complete solution is held in 'partial' --> add it to list of solutions
partitions.add(toArray(partial));
} else {
// Iterate through all numbers i less than n.
// Avoid duplicate solutions by ensuring that the partial array is always non-increasing
for (int i=n; i>0; i--) {
if (partial.isEmpty() || partial.get(partial.size()-1) >= i) {
partial.add(i);
partition(n-i, partial, partitions);
partial.remove(partial.size()-1);
}
}
}
}
/**
* Helper method: creates a new integer array and copies the contents of the list into the array.
*/
private static int[] toArray(List<Integer> list) {
int i = 0;
int[] arr = new int[list.size()];
for (int val : list) {
arr[i++] = val;
}
return arr;
}
}