Reference: http://www.programcreek.com/2013/10/efficient-counter-in-java/
You may often need a counter to understand the frequency of something (e.g., words) from a database or text file. A counter can be easily implemented by using a HashMap in Java. This article compares different approaches to implement a counter. Finally, an efficient one will be concluded.
UPDATE: Check out Java 8 counter, writing a counter is just 2 simple lines now.
1. The Naive Counter
Naively, it can be implemented as the following:
| String s = "one two three two three three";String[] sArr = s.split(" "); //naive approach HashMap<String, Integer> counter = new HashMap<String, Integer>(); for (String a : sArr) { if (counter.containsKey(a)) { int oldValue = counter.get(a); counter.put(a, oldValue + 1); } else { counter.put(a, 1); }} |
In each loop, you check if the key exists or not. If it does, increment the old value by 1, if not, set it to 1. This approach is simple and straightforward, but it is not the most efficient approach. This method is considered less efficient for the following reasons:
- containsKey(), get() are called twice when a key already exists. That means searching the map twice.
- Since Integer is immutable, each loop will create a new one for increment the old value
2. The Better Counter
Naturally we want a mutable integer to avoid creating many Integer objects. A mutable integer class can be defined as follows:
| class MutableInteger { private int val; public MutableInteger(int val) { this.val = val; } public int get() { return val; } public void set(int val) { this.val = val; } //used to print value convinently public String toString(){ return Integer.toString(val); }} |
And the counter is improved and changed to the following:
| HashMap<String, MutableInteger> newCounter = new HashMap<String, MutableInteger>(); for (String a : sArr) { if (newCounter.containsKey(a)) { MutableInteger oldValue = newCounter.get(a); oldValue.set(oldValue.get() + 1); } else { newCounter.put(a, new MutableInteger(1)); }} |
This seems better because it does not require creating many Integer objects any longer. However, the search is still twice in each loop if a key exists.
3. The Efficient Counter
The HashMap.put(key, value) method returns the key‘s current value. This is useful, because we can use the reference of the old value to update the value without searching one more time!
| HashMap<String, MutableInteger> efficientCounter = new HashMap<String, MutableInteger>(); for (String a : sArr) { MutableInteger initValue = new MutableInteger(1); MutableInteger oldValue = efficientCounter.put(a, initValue); if(oldValue != null){ initValue.set(oldValue.get() + 1); }} |
4. Performance Difference
To test the performance of the three different approaches, the following code is used. The performance test is on 1 million times. The raw results are as follows:
Naive Approach : 222796000Better Approach: 117283000Efficient Approach: 96374000
The difference is significant - 223 vs. 117 vs. 96. There is huge difference between Naive and Better, which indicates that creating objects are expensive!
| String s = "one two three two three three";String[] sArr = s.split(" "); long startTime = 0;long endTime = 0;long duration = 0; // naive approachstartTime = System.nanoTime();HashMap<String, Integer> counter = new HashMap<String, Integer>(); for (int i = 0; i < 1000000; i++) for (String a : sArr) { if (counter.containsKey(a)) { int oldValue = counter.get(a); counter.put(a, oldValue + 1); } else { counter.put(a, 1); } } endTime = System.nanoTime();duration = endTime - startTime;System.out.println("Naive Approach : " + duration); // better approachstartTime = System.nanoTime();HashMap<String, MutableInteger> newCounter = new HashMap<String, MutableInteger>(); for (int i = 0; i < 1000000; i++) for (String a : sArr) { if (newCounter.containsKey(a)) { MutableInteger oldValue = newCounter.get(a); oldValue.set(oldValue.get() + 1); } else { newCounter.put(a, new MutableInteger(1)); } } endTime = System.nanoTime();duration = endTime - startTime;System.out.println("Better Approach: " + duration); // efficient approachstartTime = System.nanoTime(); HashMap<String, MutableInteger> efficientCounter = new HashMap<String, MutableInteger>(); for (int i = 0; i < 1000000; i++) for (String a : sArr) { MutableInteger initValue = new MutableInteger(1); MutableInteger oldValue = efficientCounter.put(a, initValue); if (oldValue != null) { initValue.set(oldValue.get() + 1); } } endTime = System.nanoTime();duration = endTime - startTime;System.out.println("Efficient Approach: " + duration); |
When you use a counter, you probably also need a function to sort the map by value. You can check out the frequently used method of HashMap.
5. Solutions from Keith
Added a couple tests:
1)
Refactored "better approach" to just call get instead of containsKey. Usually,
the elements you want are in the HashMap so that reduces from two searches to
one.
2)
Added a test with AtomicInteger, which michal mentioned.
3)
Compared to singleton int array, which uses less memory according to http://amzn.com/0748614079
I
ran the test program 3x and took the min to remove variance from other programs.
Note that you can‘t do this within the program or the results are affected too
much, probably due to gc.
Naive: 201716122Better Approach: 112259166Efficient
Approach: 93066471Better Approach (without containsKey): 69578496Better Approach
(without containsKey, with AtomicInteger): 94313287Better Approach (without
containsKey, with int[]): 65877234
Better
Approach (without containsKey):
| HashMap<String, MutableInteger> efficientCounter2 = new HashMap<String, MutableInteger>();for (int i = 0; i < NUM_ITERATIONS; i++) { for (String a : sArr) { MutableInteger value = efficientCounter2.get(a); if (value != null) { value.set(value.get() + 1); } else { efficientCounter2.put(a, new MutableInteger(1)); } }} |
Better
Approach (without containsKey, with AtomicInteger):
| HashMap<String, AtomicInteger> atomicCounter = new HashMap<String, AtomicInteger>();for (int i = 0; i < NUM_ITERATIONS; i++) { for (String a : sArr) { AtomicInteger value = atomicCounter.get(a); if (value != null) { value.incrementAndGet(); } else { atomicCounter.put(a, new AtomicInteger(1)); } }} |
Better
Approach (without containsKey, with int[]):
|
HashMap<String, int[]> intCounter = new HashMap<String, int[]>(); for { |
Guava‘s
MultiSet is probably faster still.
6.
Conclusion
The
winner is the last one which uses int arrays.