java - 如何参数化Comparable接口(interface)？

Question

我有一个主类 -- Simulator -- 使用另外两个类 -- Producer 和 Evaluator。生产者产生结果，而评估者评估这些结果。模拟器通过查询生产者控制执行流程，然后将结果传送给评估器。

Producer 和 Evaluator 的实际实现在运行时是已知的，在编译时我只知道它们的接口(interface)。下面我粘贴了接口(interface)、示例实现和模拟器类的内容。

旧代码

package com.test;

import java.util.ArrayList;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import java.util.Map.Entry;

/**
 * Producers produce results. I do not care what is their type, but the values
 * in the map have to be comparable amongst themselves.
 */
interface IProducer {
    public Map<Integer, Comparable> getResults();
}

/**
 * This implementation ranks items in the map by using Strings.
 */
class ProducerA implements IProducer {
    @Override
    public Map<Integer, Comparable> getResults() {
        Map<Integer, Comparable> result = new HashMap<Integer, Comparable>();
        result.put(1, "A");
        result.put(2, "B");
        result.put(3, "B");
        return result;
    }
}

/**
 * This implementation ranks items in the map by using integers.
 */
class ProducerB implements IProducer {
    @Override
    public Map<Integer, Comparable> getResults() {
        Map<Integer, Comparable> result = new HashMap<Integer, Comparable>();
        result.put(1, 10);
        result.put(2, 30);
        result.put(3, 30);

        return result;
    }
}

/**
 * Evaluator evaluates the results against the given groundTruth. All it needs
 * to know about results, is that they are comparable amongst themselves.
 */
interface IEvaluator {
    public double evaluate(Map<Integer, Comparable> results,
            Map<Integer, Double> groundTruth);
}

/**
 * This is example of an evaluator (a metric) -- Kendall's Tau B.
 */
class KendallTauB implements IEvaluator {
    @Override
    public double evaluate(Map<Integer, Comparable> results,
            Map<Integer, Double> groundTruth) {

        int concordant = 0, discordant = 0, tiedRanks = 0, tiedCapabilities = 0;

        for (Entry<Integer, Comparable> rank1 : results.entrySet()) {
            for (Entry<Integer, Comparable> rank2 : results.entrySet()) {
                if (rank1.getKey() < rank2.getKey()) {
                    final Comparable r1 = rank1.getValue();
                    final Comparable r2 = rank2.getValue();
                    final Double c1 = groundTruth.get(rank1.getKey());
                    final Double c2 = groundTruth.get(rank2.getKey());

                    final int rankDiff = r1.compareTo(r2);
                    final int capDiff = c1.compareTo(c2);

                    if (rankDiff * capDiff > 0) {
                        concordant++;
                    } else if (rankDiff * capDiff < 0) {
                        discordant++;
                    } else {
                        if (rankDiff == 0)
                            tiedRanks++;

                        if (capDiff == 0)
                            tiedCapabilities++;
                    }
                }
            }
        }

        final double n = results.size() * (results.size() - 1d) / 2d;

        return (concordant - discordant)
                / Math.sqrt((n - tiedRanks) * (n - tiedCapabilities));
    }
}

/**
 * The simulator class that queries the producer and them conveys results to the
 * evaluator.
 */
public class Simulator {
    public static void main(String[] args) {
        Map<Integer, Double> groundTruth = new HashMap<Integer, Double>();
        groundTruth.put(1, 1d);
        groundTruth.put(2, 2d);
        groundTruth.put(3, 3d);

        List<IProducer> producerImplementations = lookUpProducers();
        List<IEvaluator> evaluatorImplementations = lookUpEvaluators();

        IProducer producer = producerImplementations.get(1); // pick a producer
        IEvaluator evaluator = evaluatorImplementations.get(0); // pick an evaluator
        // Notice that this class should NOT know what actually comes from
        // producers (besides that is comparable)
        Map<Integer, Comparable> results = producer.getResults();
        double score = evaluator.evaluate(results, groundTruth);

        System.out.printf("Score is %.2f\n", score);
    }

    // Methods below are for demonstration purposes only. I'm actually using
    // ServiceLoader.load(Clazz) to dynamically discover and load classes that
    // implement these interfaces

    public static List<IProducer> lookUpProducers() {
        List<IProducer> producers = new ArrayList<IProducer>();
        producers.add(new ProducerA());
        producers.add(new ProducerB());

        return producers;
    }

    public static List<IEvaluator> lookUpEvaluators() {
        List<IEvaluator> evaluators = new ArrayList<IEvaluator>();
        evaluators.add(new KendallTauB());

        return evaluators;
    }
}

此代码应该编译并运行。无论您选择哪个生产者实现，您都应该得到相同的结果 (0.82)。

编译器在几个地方警告我不要使用泛型:

• 在 Simulator 类中，在接口(interface) IEvaluator 和 IProducer 中，以及在实现 IProducer 接口(interface)的类中，每当我引用 Comparable 接口(interface)时，我都会收到以下警告:Comparable 是原始类型。对通用类型 Comparable 的引用应该被参数化
• 在实现 IEvaluator 的类中，我收到以下警告(在对 Map 的值调用 compareTo() 时):类型安全:方法 compareTo(Object) 属于原始类型 Comparable。对通用类型 Comparable 的引用应该被参数化

综上所述，模拟器可以正常工作。现在，我想摆脱编译警告。问题是，我不知道如何参数化接口(interface) IEvaluator 和 IProducer，以及如何更改 IProducer 和 IEvaluator 的实现。

我有一些限制:

• 我不知道生产者将返回的 Map 中值的类型。但我知道，它们都属于同一类型，并且它们将实现 Comparable 接口(interface)。
• 同样，IEvaluator 实例不需要知道任何有关正在评估的结果的信息，除了它们属于同一类型并且它们是可比较的(IEvaluator 实现需要能够调用 compareTo() 方法。)。
• 我必须让 Simulator 类摆脱这种“可比较”的困境——它不需要了解这些类型的任何信息(除了属于同一类型，这也是可比较的)。它的工作只是简单地将结果从生产者传达给评估者。

有什么想法吗？

编辑修订版

使用下面答案中的一些想法，我进入了这个阶段，它编译和运行时没有警告，也不需要使用 SuppressWarnings 指令。这与 Eero 的建议非常相似，但主要方法有点不同。

package com.test;

import java.util.ArrayList;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import java.util.Map.Entry;

/**
 * Producers produce results. I do not care what is their type, but the values
 * in the map have to be comparable amongst themselves.
 */
interface IProducer<T extends Comparable<T>> {
    public Map<Integer, T> getResults();
}

/**
 * This implementation ranks items in the map by using Strings.
 */
class ProducerA implements IProducer<String> {
    @Override
    public Map<Integer, String> getResults() {
        Map<Integer, String> result = new HashMap<Integer, String>();
        result.put(1, "A");
        result.put(2, "B");
        result.put(3, "B");

        return result;
    }
}

/**
 * This implementation ranks items in the map by using integers.
 */
class ProducerB implements IProducer<Integer> {
    @Override
    public Map<Integer, Integer> getResults() {
        Map<Integer, Integer> result = new HashMap<Integer, Integer>();
        result.put(1, 10);
        result.put(2, 30);
        result.put(3, 30);

        return result;
    }
}

/**
 * Evaluator evaluates the results against the given groundTruth. All it needs
 * to know about results, is that they are comparable amongst themselves.
 */
interface IEvaluator {
    public <T extends Comparable<T>> double evaluate(Map<Integer, T> results,
            Map<Integer, Double> groundTruth);
}

/**
 * This is example of an evaluator (a metric) -- Kendall's Tau B.
 */
class KendallTauB implements IEvaluator {
    @Override
    public <T extends Comparable<T>> double evaluate(Map<Integer, T> results,
            Map<Integer, Double> groundTruth) {
        int concordant = 0, discordant = 0, tiedRanks = 0, tiedCapabilities = 0;

        for (Entry<Integer, T> rank1 : results.entrySet()) {
            for (Entry<Integer, T> rank2 : results.entrySet()) {
                if (rank1.getKey() < rank2.getKey()) {
                    final T r1 = rank1.getValue();
                    final T r2 = rank2.getValue();
                    final Double c1 = groundTruth.get(rank1.getKey());
                    final Double c2 = groundTruth.get(rank2.getKey());

                    final int rankDiff = r1.compareTo(r2);
                    final int capDiff = c1.compareTo(c2);

                    if (rankDiff * capDiff > 0) {
                        concordant++;
                    } else if (rankDiff * capDiff < 0) {
                        discordant++;
                    } else {
                        if (rankDiff == 0)
                            tiedRanks++;

                        if (capDiff == 0)
                            tiedCapabilities++;
                    }
                }
            }
        }

        final double n = results.size() * (results.size() - 1d) / 2d;

        return (concordant - discordant)
                / Math.sqrt((n - tiedRanks) * (n - tiedCapabilities));
    }
}

/**
 * The simulator class that queries the producer and them conveys results to the
 * evaluator.
 */
public class Main {
    public static void main(String[] args) {
        Map<Integer, Double> groundTruth = new HashMap<Integer, Double>();
        groundTruth.put(1, 1d);
        groundTruth.put(2, 2d);
        groundTruth.put(3, 3d);

        List<IProducer<?>> producerImplementations = lookUpProducers();
        List<IEvaluator> evaluatorImplementations = lookUpEvaluators();

        IProducer<?> producer = producerImplementations.get(0);
        IEvaluator evaluator = evaluatorImplementations.get(0);

        // Notice that this class should NOT know what actually comes from
        // producers (besides that is comparable)
        double score = evaluator.evaluate(producer.getResults(), groundTruth);

        System.out.printf("Score is %.2f\n", score);
    }

    // Methods below are for demonstration purposes only. I'm actually using
    // ServiceLoader.load(Clazz) to dynamically discover and load classes that
    // implement these interfaces
    public static List<IProducer<?>> lookUpProducers() {
        List<IProducer<?>> producers = new ArrayList<IProducer<?>>();
        producers.add(new ProducerA());
        producers.add(new ProducerB());

        return producers;
    }

    public static List<IEvaluator> lookUpEvaluators() {
        List<IEvaluator> evaluators = new ArrayList<IEvaluator>();
        evaluators.add(new KendallTauB());

        return evaluators;
    }
}

主要区别似乎在于 main 方法，目前看起来像这样。

    public static void main(String[] args) {
        Map<Integer, Double> groundTruth = new HashMap<Integer, Double>();
        groundTruth.put(1, 1d);
        groundTruth.put(2, 2d);
        groundTruth.put(3, 3d);

        List<IProducer<?>> producerImplementations = lookUpProducers();
        List<IEvaluator> evaluatorImplementations = lookUpEvaluators();

        IProducer<?> producer = producerImplementations.get(0);
        IEvaluator evaluator = evaluatorImplementations.get(0);

        // Notice that this class should NOT know what actually comes from
        // producers (besides that is comparable)
        double score = evaluator.evaluate(producer.getResults(), groundTruth);

        System.out.printf("Score is %.2f\n", score);
    }

这行得通。但是，如果我将代码更改为:

    public static void main(String[] args) {
        Map<Integer, Double> groundTruth = new HashMap<Integer, Double>();
        groundTruth.put(1, 1d);
        groundTruth.put(2, 2d);
        groundTruth.put(3, 3d);

        List<IProducer<?>> producerImplementations = lookUpProducers();
        List<IEvaluator> evaluatorImplementations = lookUpEvaluators();

        IProducer<?> producer = producerImplementations.get(0);
        IEvaluator evaluator = evaluatorImplementations.get(0);

        // Notice that this class should NOT know what actually comes from
        // producers (besides that is comparable)

        // Lines below changed
        Map<Integer, ? extends Comparable<?>> ranks = producer.getResults();            
        double score = evaluator.evaluate(ranks, groundTruth);

        System.out.printf("Score is %.2f\n", score);
}

该死的东西甚至不会编译，说:绑定(bind)不匹配:IEvaluator 类型的通用方法 evaluate(Map, Map) 不适用于参数 (Map>, Map)。推断类型捕获#3-of ? extends Comparable 不是有界参数的有效替代 >

这对我来说很奇怪。如果我调用 evaluator.evaluate(producer.getResults(), groundTruth)，代码就会工作。但是，如果我首先调用 producer.getResults() 方法，并将其存储到一个变量，然后使用该变量调用 evaluate 方法 (evaluator.evaluate(ranks, groundTruth))，我会得到编译错误(无论该变量的类型)。

Answer 1

您需要指定对象愿意将自己与哪些事物进行比较。像这样的东西:

import java.util.Map;
import java.util.HashMap;

interface IProducer<T extends Comparable<? super T>> {
    public Map<Integer, T> getResults();
}

interface IEvaluator {
    public <T extends Comparable<? super T>> double evaluate(Map<Integer, T> results,
                                                             Map<Integer, Double> groundTruth);
}

public class Main {
  public static void main(String[] args) {
    IProducer<String> producer = null;
    IEvaluator evaluator = null;
    Map<Integer, String> results = producer.getResults();
    double score = evaluator.evaluate(results, new HashMap<Integer, Double>());
  }
}