Efficiency of the way comparator works

问题

I am trying to use a comparator to help sort a list of objects. I have a question about how exactly the comparator works and what it would be doing exactly in the following example:

private static Comparator<Student> comparator()
{
        return (Student a, Student b) ->
        {  
                return Integer.compare(complexOperation(a), complexOperation(b));
        }
}

As you can see above, there is a need to compare and sort students according to an integer rank returned by the complexOperation() method. As the name suggests, it is a heavy operation. Would the above approach be the most efficient? Or would it be better to essentially run through each student in the list I am trying to sort, perform the complexOperation() per student and store the result in a field in the Student object. Then the comparator would just do an:

Integer.compare(a.getRank(), b.getRank())

Would both these approaches be comparable or, due to the way the comparator works (perhaps compares the same object more than once with others hence running complexOperation() multiple times per Student during the compare), would it be faster to do the pre computation of the complexOperation() result in a student field?

The above would be called like so:

Collections.sort(students, comparator());

Hope that was clear!

Edit: Lets say, for the sake of it, it is not possible to add a field to the Student object (This is a toy problem for a more complex situation where I am not at liberty to modify the Student object). Would it still be better to perhaps create a custom Object with Student sitting inside with another field added rather than doing the complexOperation() right in the comparator? Or is there another way to approach the problem? I can think of creating a Hashmap that takes student id as key and the result of the complexOperation() as value and just creates/access that record within the comparator?

回答1:

On average, your sort algorithm will call complexOperation() method about log₂N times for an array of N students. If the operation is really slow, you may be better off running it once for each student. This could bring an order of magnitude improvement for an array of 1,000 students.

However, you do not have to do it explicitly: you could make complexOperation(...) store the result for each student, and then return the cached value on subsequent requests:

private Map<Student,Integer> cache = new HashMap<Student,Integer>();

private int complexOperation(Student s) {
    // See if we computed the rank of the student before
    Integer res = cache.get(s);
    if (res != null) {
        // We did! Just return the stored result:
        return res.intValue();
    }
    ... // do the real computation here
    // Save the result for future invocations
    cache.put(s, result);
    return result;
}

Note that in order for this approach to work, Student class needs to implement hashCode and equals.

回答2:

Basically, you want to compare students by comparing some values that each maps to. This is usually done by

    static Comparator<Student> comparator()
    {
        return Comparator.comparing( Foo::complexOperation );
    }

However, since the function complexOperation is too expensive, we want to cache its results. We can have a general purpose utility method Function cache(Function)

    static Comparator<Student> comparator()
    {
        return Comparator.comparing( cache(Foo::complexOperation) );
    }

In general, it is better that the caller can supply a Map as the cache

public static <K,V> Function<K,V> cache(Function<K,V> f, Map<K,V> cache)
{
    return k->cache.computeIfAbsent(k, f);
}

We can use IdentityHashMap as the default cache

public static <K,V> Function<K,V> cache(Function<K,V> f)
{
    return cache(f, new IdentityHashMap<>());
}

来源：https://stackoverflow.com/questions/32878398/efficiency-of-the-way-comparator-works

标签

java

performance

sorting

comparator