项目中使用到List求交集,很容易想到collecion.retainAll()方法,但是在数据量比较大时,这个方法效率并不高。本文研究了几种常用的方法,以供大家参考。
【首先】梳理下思路,List去重一般有几种方法。
复杂度O(NM) ,一般使用contains()检查是否包含
复杂度O(N),一般将内层List转化为HashSet实现
复杂度O(N),一般将内层List转化为字节映射实现
【其次】这里其实忽略了一个点,就是 『单层遍历』中,检查 元素不包含 时,需要将这个元素移除(即remove方法)。remove时,也会导致性能问题。
这里面我们使用Java8中java.util.AbstractCollection#retainAll
方法来验证下我们的思路。
// Java8 中 方法:java.util.AbstractCollection#retainAll public boolean retainAll(Collection<?> c) { Objects.requireNonNull(c); boolean modified = false; Iterator<E> it = iterator(); // 1. 外层遍历 while (it.hasNext()) { // 2. 内层查找『是否包含』 if (!c.contains(it.next())) { // 3. 不包含时,移除外层元素 it.remove(); modified = true; } } return modified; }
这里用图总结下,求交集的流程:
java中常用2种遍历查找的List:ArrayList、LinkedList,在内外层中测试。
// 外层:ArrayList,内层:ArrayList private void outArrayListInnerArrayList(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); ArrayList<Long> setA = new ArrayList<>(listA); ArrayList<Long> setB = new ArrayList<>(listB); setA.retainAll(setB); long end = System.currentTimeMillis(); System.out.println("[ArrayList-ArrayList]RetainAll耗时:" + (end - begin)); } // 外层:LinkedList,内层:ArrayList private void outLinkedListInnerArrayList(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); LinkedList<Long> setA = new LinkedList<>(listA); ArrayList<Long> setB = new ArrayList<>(listB); setA.retainAll(setB); long end = System.currentTimeMillis(); System.out.println("[LinkedList-ArrayList]RetainAll耗时:" + (end - begin)); } // 外层:ArrayList,内层:LinkedList private void outArrayListInnerLinkedList(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); LinkedList<Long> setA = new LinkedList<>(listA); ArrayList<Long> setB = new ArrayList<>(listB); setA.retainAll(setB); long end = System.currentTimeMillis(); System.out.println("[LinkedList-ArrayList]RetainAll耗时:" + (end - begin)); } // 外层:LinkedList,内层:LinkedList private void outLinkedListInnerLinkedList(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); LinkedList<Long> setA = new LinkedList<>(listA); ArrayList<Long> setB = new ArrayList<>(listB); setA.retainAll(setB); long end = System.currentTimeMillis(); System.out.println("[LinkedList-LinkedList]RetainAll耗时:" + (end - begin)); }
java中常用HashSet,内层替换为HashSet查找。
// 外层:ArrayList,内层:HashSet private void outArrayListInnerHashSet(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); ArrayList<Long> setA = new ArrayList<>(listA); HashSet<Long> setB = new HashSet<>(listB); setA.retainAll(setB); long end = System.currentTimeMillis(); System.out.println("[ArrayList-HashSet]RetainAll耗时:" + (end - begin)); } // 外层:LinkedList,内层:HashSet private void outLinkedListInnerHashSet(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); LinkedList<Long> setA = new LinkedList<>(listA); HashSet<Long> setB = new HashSet<>(listB); setA.retainAll(setB); long end = System.currentTimeMillis(); System.out.println("[LinkedList-HashSet]RetainAll耗时:" + (end - begin)); }
BitSet也称作BitMap,它是一种通用的快速数据结构,不幸的是它太费内存,所以通常我们使用压缩位图。RoaringBitmap是一种压缩位置,它提供更好的压缩效果,在某些情况下比其它压缩位图快好几百倍。
https://github.com/RoaringBitmap/RoaringBitmap
RoaringBitmap已经使用在很多知名的开源项目中:
Roaringbitmap中在Long类型中,提供了2种实现Roaring64NavigableMap
和Roaring64Bitmap
。Roaring64NavigableMap
基于红黑树实现,Roaring64Bitmap
基于ART(The Adaptive Radix Tree: ARTful Indexing for Main-Memory Databases )数据结构实现。
那么,外层使用ArrayList、LinkedList,内层使用Roaring64NavigableMap、Roaring64Bitmap。
// 外层:ArrayList,内层:Roaring64NavigableMap private void outArrayListInnerRoaring64NavigableMap(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); ArrayList<Long> setA = new ArrayList<>(listA); Roaring64NavigableMap ansB = new Roaring64NavigableMap(); listB.forEach(ansB::addLong); setA.removeIf(e -> !ansB.contains(e)); long end = System.currentTimeMillis(); System.out.println("[ArrayList-Roaring64NavigableMap]RetainAll耗时:" + (end - begin)); } // 外层:LinkedList,内层:Roaring64NavigableMap private void outLinkedListInnerRoaring64NavigableMap(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); LinkedList<Long> setA = new LinkedList<>(listA); Roaring64NavigableMap ansB = new Roaring64NavigableMap(); listB.forEach(ansB::addLong); setA.removeIf(e -> !ansB.contains(e)); long end = System.currentTimeMillis(); System.out.println("[LinkedList-Roaring64NavigableMap]RetainAll耗时:" + (end - begin)); } // 外层:ArrayList,内层:Roaring64Bitmap private void outArrayListInnerRoaring64Bitmap(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); ArrayList<Long> setA = new ArrayList<>(listA); Roaring64NavigableMap ansB = new Roaring64NavigableMap(); listB.forEach(ansB::addLong); setA.removeIf(e -> !ansB.contains(e)); long end = System.currentTimeMillis(); System.out.println("[ArrayList-Roaring64Bitmap]RetainAll耗时:" + (end - begin)); } // 外层:LinkedList,内层:Roaring64Bitmap private void outLinkedListInnerRoaring64Bitmap(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); LinkedList<Long> setA = new LinkedList<>(listA); Roaring64Bitmap ansB = new Roaring64Bitmap(); listB.forEach(ansB::addLong); setA.removeIf(e -> !ansB.contains(e)); long end = System.currentTimeMillis(); System.out.println("[LinkedList-Roaring64Bitmap]RetainAll耗时:" + (end - begin)); }
使用Mac Pro 2021 M1 + JDK8测试,100万级的数据太慢,实行中没有太大参考意义,有兴趣可以自行测试。
说明:由于受数据,以及电脑本身负载影响,测试结果可能不一致,仅做量级参考。
查找方法(外层-内层) | 1万(毫秒) | 10万(毫秒) | 20万(毫秒) | 50万(毫秒) | |
『外层遍历+内层遍历』查找 | ArrayList-ArrayList | 67 | 5502 | 26280 | 264133 |
LinkedList-ArrayList | 63 | 5418 | 27961 | 272362 | |
LinkedList-ArrayList | 57 | 5436 | 21330 | 260976 | |
LinkedList-LinkedList | 59 | 5153 | 23251 | 252472 | |
『外层遍历+内层Hash』查找 | ArrayList-HashSet | 8 | 46 | 75 | 102 |
LinkedList-HashSet | 8 | 17 | 49 | 82 | |
『外层遍历+内层bitMap』查找 | ArrayList-Roaring64NavigableMap | 91 | 265 | 719 | 876 |
LinkedList-Roaring64NavigableMap | 26 | 125 | 562 | 876 | |
ArrayList-Roaring64Bitmap | 20 | 78 | 572 | 801 | |
LinkedList-Roaring64Bitmap | 119 | 171 | 221 | 384 |
pom.xml
<dependency> <groupId>org.roaringbitmap</groupId> <artifactId>RoaringBitmap</artifactId> <version>0.9.23</version> </dependency>
完整测试代码
import org.apache.commons.lang.math.RandomUtils; import org.junit.Test; import org.roaringbitmap.longlong.Roaring64Bitmap; import org.roaringbitmap.longlong.Roaring64NavigableMap; import java.util.ArrayList; import java.util.HashSet; import java.util.LinkedList; import java.util.List; import java.util.Random; public class SetOperation { /** * 集合的运算方法用时测试 */ @Test public void setOperation() { int size = 50_0000; List<Long> listA = new ArrayList<>(size); List<Long> listB = new ArrayList<>(size); initData(size, listA, listB); //『外层遍历+内层遍历』查找 System.out.println("1. 『外层遍历+内层遍历』查找"); outArrayListInnerArrayList(new ArrayList<>(listA), new ArrayList<>(listB)); outLinkedListInnerArrayList(new ArrayList<>(listA), new ArrayList<>(listB)); outArrayListInnerLinkedList(new ArrayList<>(listA), new ArrayList<>(listB)); outLinkedListInnerLinkedList(new ArrayList<>(listA), new ArrayList<>(listB)); //『外层遍历+内层Hash』查找: System.out.println("2.『外层遍历+内层Hash』查找:"); outArrayListInnerHashSet(new ArrayList<>(listA), new ArrayList<>(listB)); outLinkedListInnerHashSet(new ArrayList<>(listA), new ArrayList<>(listB)); //『外层遍历+内层bitMap』查找 System.out.println("3.『外层遍历+内层bitMap』查找"); outArrayListInnerRoaring64NavigableMap(new ArrayList<>(listA), new ArrayList<>(listB)); outLinkedListInnerRoaring64NavigableMap(new ArrayList<>(listA), new ArrayList<>(listB)); outArrayListInnerRoaring64Bitmap(new ArrayList<>(listA), new ArrayList<>(listB)); outLinkedListInnerRoaring64Bitmap(new ArrayList<>(listA), new ArrayList<>(listB)); } // 外层:ArrayList,内层:ArrayList private void outArrayListInnerArrayList(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); ArrayList<Long> setA = new ArrayList<>(listA); ArrayList<Long> setB = new ArrayList<>(listB); setA.retainAll(setB); long end = System.currentTimeMillis(); System.out.println("[ArrayList-ArrayList]RetainAll耗时:" + (end - begin)); } // 外层:LinkedList,内层:ArrayList private void outLinkedListInnerArrayList(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); LinkedList<Long> setA = new LinkedList<>(listA); ArrayList<Long> setB = new ArrayList<>(listB); setA.retainAll(setB); long end = System.currentTimeMillis(); System.out.println("[LinkedList-ArrayList]RetainAll耗时:" + (end - begin)); } // 外层:ArrayList,内层:LinkedList private void outArrayListInnerLinkedList(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); LinkedList<Long> setA = new LinkedList<>(listA); ArrayList<Long> setB = new ArrayList<>(listB); setA.retainAll(setB); long end = System.currentTimeMillis(); System.out.println("[LinkedList-ArrayList]RetainAll耗时:" + (end - begin)); } // 外层:LinkedList,内层:LinkedList private void outLinkedListInnerLinkedList(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); LinkedList<Long> setA = new LinkedList<>(listA); ArrayList<Long> setB = new ArrayList<>(listB); setA.retainAll(setB); long end = System.currentTimeMillis(); System.out.println("[LinkedList-LinkedList]RetainAll耗时:" + (end - begin)); } // 外层:ArrayList,内层:HashSet private void outArrayListInnerHashSet(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); ArrayList<Long> setA = new ArrayList<>(listA); HashSet<Long> setB = new HashSet<>(listB); setA.retainAll(setB); long end = System.currentTimeMillis(); System.out.println("[ArrayList-HashSet]RetainAll耗时:" + (end - begin)); } // 外层:LinkedList,内层:HashSet private void outLinkedListInnerHashSet(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); LinkedList<Long> setA = new LinkedList<>(listA); HashSet<Long> setB = new HashSet<>(listB); setA.retainAll(setB); long end = System.currentTimeMillis(); System.out.println("[LinkedList-HashSet]RetainAll耗时:" + (end - begin)); } // 外层:ArrayList,内层:Roaring64NavigableMap private void outArrayListInnerRoaring64NavigableMap(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); ArrayList<Long> setA = new ArrayList<>(listA); Roaring64NavigableMap ansB = new Roaring64NavigableMap(); listB.forEach(ansB::addLong); setA.removeIf(e -> !ansB.contains(e)); long end = System.currentTimeMillis(); System.out.println("[ArrayList-Roaring64NavigableMap]RetainAll耗时:" + (end - begin)); } // 外层:LinkedList,内层:Roaring64NavigableMap private void outLinkedListInnerRoaring64NavigableMap(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); LinkedList<Long> setA = new LinkedList<>(listA); Roaring64NavigableMap ansB = new Roaring64NavigableMap(); listB.forEach(ansB::addLong); setA.removeIf(e -> !ansB.contains(e)); long end = System.currentTimeMillis(); System.out.println("[LinkedList-Roaring64NavigableMap]RetainAll耗时:" + (end - begin)); } // 外层:ArrayList,内层:Roaring64Bitmap private void outArrayListInnerRoaring64Bitmap(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); ArrayList<Long> setA = new ArrayList<>(listA); Roaring64NavigableMap ansB = new Roaring64NavigableMap(); listB.forEach(ansB::addLong); setA.removeIf(e -> !ansB.contains(e)); long end = System.currentTimeMillis(); System.out.println("[ArrayList-Roaring64Bitmap]RetainAll耗时:" + (end - begin)); } // 外层:LinkedList,内层:Roaring64Bitmap private void outLinkedListInnerRoaring64Bitmap(List<Long> listA, List<Long> listB) { long begin = System.currentTimeMillis(); LinkedList<Long> setA = new LinkedList<>(listA); Roaring64Bitmap ansB = new Roaring64Bitmap(); listB.forEach(ansB::addLong); setA.removeIf(e -> !ansB.contains(e)); long end = System.currentTimeMillis(); System.out.println("[LinkedList-Roaring64Bitmap]RetainAll耗时:" + (end - begin)); } private void initData(int size, List<Long> listA, List<Long> listB) { Random random = new Random(); Random random2 = new Random(); random.longs(size).forEach(e -> { listA.add(e); if (random2.nextFloat() > 0.5) { listB.add(e); } else { listB.add(RandomUtils.nextLong()); } }); } }
|