最长递增子序列的潜在 O(n) 解
Posted
技术标签:
【中文标题】最长递增子序列的潜在 O(n) 解【英文标题】:potential O(n) solution to Longest Increasing Subsequence 【发布时间】:2014-02-25 15:36:10 【问题描述】:我试图回答这个问题,只使用递归(动态编程) http://en.wikipedia.org/wiki/Longest_increasing_subsequence
从这篇文章中,我意识到最有效的现有解决方案是 O(nlgn)。我的解决方案是 O(N),我找不到失败的案例。我包括了我使用的单元测试用例。
import static org.junit.Assert.assertEquals;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.List;
import org.junit.Test;
public class LongestIncreasingSubseq
public static void main(String[] args)
int[] arr = 0, 8, 4, 12, 2, 10, 6, 14, 1, 9, 5, 13, 3, 11, 7, 15, 1;
getLongestSubSeq(arr);
public static List<Integer> getLongestSubSeq(int[] arr)
List<Integer> indices = longestRecursive(arr, 0, arr.length-1);
List<Integer> result = new ArrayList<>();
for (Integer i : indices)
result.add(arr[i]);
System.out.println(result.toString());
return result;
private static List<Integer> longestRecursive(int[] arr, int start, int end)
if (start == end)
List<Integer> singleton = new ArrayList<>();
singleton.add(start);
return singleton;
List<Integer> bestRightSubsequence = longestRecursive(arr, start+1, end); //recursive call down the array to the next start index
if (bestRightSubsequence.size() == 1 && arr[start] > arr[bestRightSubsequence.get(0)])
bestRightSubsequence.set(0, start); //larger end allows more possibilities ahead
else if (arr[start] < arr[bestRightSubsequence.get(0)])
bestRightSubsequence.add(0, start); //add to head
else if (bestRightSubsequence.size() > 1 && arr[start] < arr[bestRightSubsequence.get(1)])
//larger than head, but still smaller than 2nd, so replace to allow more possibilities ahead
bestRightSubsequence.set(0, start);
return bestRightSubsequence;
@Test
public void test()
int[] arr1 = 0, 8, 4, 12, 2, 10, 6, 14, 1, 9, 5, 13, 3, 11, 7, 15, 1;
int[] arr2 = 7, 0, 9, 2, 8, 4, 1;
int[] arr3 = 9, 11, 2, 13, 7, 15;
int[] arr4 = 10, 22, 9, 33, 21, 50, 41, 60, 80;
int[] arr5 = 1, 2, 9, 4, 7, 3, 11, 8, 14, 6;
assertEquals(getLongestSubSeq(arr1), Arrays.asList(0, 4, 6, 9, 11, 15));
assertEquals(getLongestSubSeq(arr2), Arrays.asList(0, 2, 8));
assertEquals(getLongestSubSeq(arr3), Arrays.asList(9, 11, 13, 15));
assertEquals(getLongestSubSeq(arr4), Arrays.asList(10, 22, 33, 50, 60, 80));
assertEquals(getLongestSubSeq(arr5), Arrays.asList(1, 2, 4, 7, 11, 14));
由于关系 T(n) = T(n-1) + O(1) => T(n) = O(n),因此成本严格为 O(n)
任何人都可以找到失败的案例或存在任何错误吗?非常感谢。
更新: 感谢大家指出我在之前实施中的错误。下面的最终代码通过了它曾经失败的所有测试用例。
这个想法是列出(计算)所有可能的递增子序列(每个从索引 i 开始,从 0 到 N.length-1)并选择最长的子序列。我使用记忆(使用哈希表)来避免重新计算已计算的子序列 - 因此对于每个起始索引,我们只计算一次所有递增的子序列。
但是,在这种情况下,我不确定如何正式推导出 时间复杂度 - 如果有人能阐明这一点,我将不胜感激。非常感谢。
import static org.junit.Assert.assertEquals;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import org.junit.Test;
public class LongestIncreasingSubsequence
public static List<Integer> getLongestSubSeq(int[] arr)
List<Integer> longest = new ArrayList<>();
for (int i = 0; i < arr.length; i++)
List<Integer> candidate = longestSubseqStartsWith(arr, i);
if (longest.size() < candidate.size())
longest = candidate;
List<Integer> result = new ArrayList<>();
for (Integer i : longest)
result.add(arr[i]);
System.out.println(result.toString());
cache = new HashMap<>(); //new cache otherwise collision in next use - because object is static
return result;
private static Map<Integer, List<Integer>> cache = new HashMap<>();
private static List<Integer> longestSubseqStartsWith(int[] arr, int startIndex)
if (cache.containsKey(startIndex)) //check if already computed
//must always return a clone otherwise object sharing messes things up
return new ArrayList<>(cache.get(startIndex));
if (startIndex == arr.length-1)
List<Integer> singleton = new ArrayList<>();
singleton.add(startIndex);
return singleton;
List<Integer> longest = new ArrayList<>();
for (int i = startIndex + 1; i < arr.length; i++)
if (arr[startIndex] < arr[i])
List<Integer> longestOnRight = longestSubseqStartsWith(arr, i);
if (longestOnRight.size() > longest.size())
longest = longestOnRight;
longest.add(0, startIndex);
List<Integer> cloneOfLongest = new ArrayList<>(longest);
//must always cache a clone otherwise object sharing messes things up
cache.put(startIndex, cloneOfLongest); //remember this subsequence
return longest;
@Test
public void test()
int[] arr1 = 0, 8, 4, 12, 2, 10, 6, 14, 1, 9, 5, 13, 3, 11, 7, 15, 1;
int[] arr2 = 7, 0, 9, 2, 8, 4, 1;
int[] arr3 = 9, 11, 2, 13, 7, 15;
int[] arr4 = 10, 22, 9, 33, 21, 50, 41, 60, 80;
int[] arr5 = 1, 2, 9, 4, 7, 3, 11, 8, 14, 6;
int[] arr6 = 0,0,0,0,0,0,1,1,1,1,2,3,0,0,0,1,1,0,1,1,0,1,0,3;
int[] arr7 = 0,1,2,0,1,3;
int[] arr8 = 0,1,2,3,4,5,1,3,8;
assertEquals(getLongestSubSeq(arr1), Arrays.asList(0, 4, 6, 9, 13, 15));
assertEquals(getLongestSubSeq(arr2), Arrays.asList(0, 2, 8));
assertEquals(getLongestSubSeq(arr3), Arrays.asList(9, 11, 13, 15));
assertEquals(getLongestSubSeq(arr4), Arrays.asList(10, 22, 33, 50, 60, 80));
assertEquals(getLongestSubSeq(arr5), Arrays.asList(1, 2, 4, 7, 11, 14));
assertEquals(getLongestSubSeq(arr6), Arrays.asList(0,1,2,3));
assertEquals(getLongestSubSeq(arr7), Arrays.asList(0,1,2,3));
assertEquals(getLongestSubSeq(arr8), Arrays.asList(0, 1, 2, 3, 4, 5, 8));
public static void main(String[] args)
int[] arr1 = 7, 0, 9, 2, 8, 4, 1;
System.out.println(getLongestSubSeq(arr1));
【问题讨论】:
为什么要投反对票和接近投票?这个问题非常有意义,显示了研究成果,而且绝对有用。 您可能会更幸运地找到带有模糊/随机测试的反例。实现一个已知正确的算法,生成随机序列,并比较两种实现的结果。 @kba 既不赞成也不反对,但它看起来像一个典型的 find-the-bug-in-my-code ,它经常被相当严重地反对(而且不是那么有用)(我是无法看到这些和这之间的显着差异,但选票却大不相同)。此外,一般来说,除了实际代码之外,伪代码 / 高级描述而不是 / 可以更容易地看到发生了什么。 查看您链接的***页面,引用了一篇论文的Omega(nlogn)
下限。我自己没有验证它,但我认为Omega(nlogn)
确实是常见计算模型的下限。引用链接到这里,您可以在此处下载 pdf:sciencedirect.com/science/article/pii/0012365X7590103X
"""构建软件设计有两种方式:一种是简单到没有明显的缺陷,另一种是复杂到没有明显的缺陷。第一种方法要困难得多。"""
【参考方案1】:
这是我在 python3.x 中的潜在 O(N) 解决方案:
l = list(map(int,input().split()))
t = []
t2 = []
m = 0
for i in l:
if(len(t)!=0):
if(t[-1]<=i):
if(t[-1]!=1):
t.append(i)
else:
if(len(t)>m):
t2 = t
m = len(t)
t = [i]
else:
t.append(i)
print(t2,len(t2))
【讨论】:
【参考方案2】:这是一个 O(n^2) 算法。因为有两个循环。第二个循环隐藏在方法调用中。
这是第一个循环:for (int i = 0; i < arr.length; i++)
。在这个循环中,您调用了 longestSubseqStartsWith(arr, i);
。查看longestSubseqStartWith实现我们看到for (int i = startIndex + 1; i < arr.length; i++)
【讨论】:
【参考方案3】:很抱歉成为坏消息的承担者,但这实际上是 O(n2)。我不确定你是否有更正式的想法,但这是我的分析:
consider the case when the input is sorted in descending order
(longestRecursive is never executed recursively, and the cache has no effect)
getLongestSubSeq iterates over the entire input -> 1:n
each iteration calls longestRecursive
longestRecursive compares arr[startIndex] < arr[i] for startIndex+1:n -> i - 1
因此,比较 arr[startIndex] 2 )。您可以通过发送按升序排序的输入来强制使用最大缓存。在这种情况下,getLongestSubSeq 将调用longestRecursive n 次;其中第一个将触发 n - 1 次递归调用,每个调用都会导致缓存未命中并运行 i - 1 次比较 arr[startIndex]
【讨论】:
【参考方案4】:刚才我使用以下测试用例尝试了您的算法:
@Test
public void test()
int[] arr1 = 0,1,2,3,4,5,1,3,8;
assertEquals(getLongestSubSeq(arr1), Arrays.asList(0, 1, 2, 3, 4, 5, 8));
它失败了,因为它给出了输出 1, 3, 8 根据您的评论编辑。
【讨论】:
您的测试用例的预期结果应该是Arrays.asList(0, 1, 2, 3, 4, 5, 8)
。【参考方案5】:
你的程序在这个测试用例中失败了
int[] arr5 = 0,0,0,0,0,0,1,1,1,1,2,3,0,0,0,1,1,0,1,1,0,1,0,3;
您的结果[0, 1, 3]
不应该是[0,1,2,3]
【讨论】:
好收获。0,1,2,0,1,3
甚至失败以上是关于最长递增子序列的潜在 O(n) 解的主要内容,如果未能解决你的问题,请参考以下文章