如何参数化可比接口？

Question

我有一个主类——模拟器——它使用另外两个类——生产者和评估者。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并将结果传递给评估器来控制执行流程。

Producer和Evaluator的实际实现在运行时已知，在编译时我只知道它们的接口。下面我将粘贴接口、示例实现和模拟器类的内容。

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;
    }
}

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

编译器在以下几个地方警告我不要使用泛型：

• 在Simulator类中，在接口IEEvaluator和IProducer中，以及在实现IProducer接口的类中，每当我引用可比接口时，我都会收到以下警告：Comparable是原始类型。对可比泛型类型的引用应参数化
• 在实现IEEvaluator的类中，我得到以下警告（在对Map的值调用compareTo（）时）：类型安全：方法compareTo（Object）属于原始类型Comparable。对泛型类型Compariable的引用应参数化

尽管如此，模拟器仍然有效。现在，我想摆脱编译警告。问题是，我不知道，如何参数化接口IEValator和IProducer，以及如何改变IProducer和IEValator的实现。

我有一些限制：

• 我无法知道制作者将返回的映射中的值的类型。但我知道，它们都将是相同的类型，并且它们将实现可比较的接口
• 类似地，IEEvaluator实例不需要知道任何有关正在评估的结果的信息，除非它们是相同类型的并且具有可比性（IEEvaluator implementations需要能够调用compareTo（）方法）
• 我必须让Simulator类远离这种“可比较”的困境——它不需要知道关于这些类型的任何信息（除了是相同的类型，这也是可比较的）。它的工作就是简单地将结果从制作者传递给评估者

有什么想法吗？

使用下面答案中的一些想法，我到了这个阶段，它编译和运行没有警告，也不需要使用禁止警告指令。这与埃罗建议的非常相似，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;
    }
}

关键的区别似乎在于main方法，它目前看起来是这样的。

    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);
    }

这很有效。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，比如：绑定不匹配：IEEvaluator类型的泛型方法evaluate（Map，Map）不适用于参数（Map）

这对我来说太奇怪了。如果调用evaluator，代码就会工作。评估（producer.getResults（），groundTruth）。不过，如果我先打电话给制片人。getResults（）方法，并将其存储到一个变量中，然后使用该变量调用evaluate方法（evaluator.evaluate（ranks，groundTruth）），我得到编译错误（不管该变量的类型如何）。

Answer 1

我已经把我的答案贴在下面了。一些注意事项：

• 制片人显然知道自己的类型
• 在调用该方法之前，评估者不知道它在处理什么
• ProducerImplements包含几种不同的类型，因此，当您实际选择其中一种类型时，您最终只能选择一种类型

代码如下：

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 their actual type is, but the
 * values in the map have to be comparable amongst themselves.
 */
interface IProducer<T extends Comparable<T>> {
    public Map<Integer, T> getResults();
}

/**
 * This example 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 example 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, metric Kendall Tau-B. Don't bother with
 * semantics, all that matters is that I want to be able to call
 * r1.compareTo(r2) for every (r1, r2) that appear in Map<Integer, T> results.
 */
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 ranksDiff = r1.compareTo(r2);
                    final int actualDiff = c1.compareTo(c2);

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

                        if (actualDiff == 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) {
        // example of a ground truth
        Map<Integer, Double> groundTruth = new HashMap<Integer, Double>();
        groundTruth.put(1, 1d);
        groundTruth.put(2, 2d);
        groundTruth.put(3, 3d);

        // dynamically load producers
        List<IProducer<?>> producerImplementations = lookUpProducers();

        // dynamically load evaluators
        List<IEvaluator> evaluatorImplementations = lookUpEvaluators();

        // pick a producer
        IProducer<?> producer = producerImplementations.get(0);

        // pick an evaluator
        IEvaluator evaluator = evaluatorImplementations.get(0);

        // evaluate the result against the ground truth
        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 interfaces IProducer and IEvaluator
    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;
    }
}

Answer 2

您需要指定对象愿意与自己进行比较的对象类型。比如：

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>());
  }
}

Answer 3

警告只是要求你做这样的事情吗？

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

每当我实现可比性（或扩展比较器）时，我总是这样做：

public class Dog implements Comparable<Dog> {

    private String breed;

    public String getBreed() {
        return breed;
    }

    public void setBreed(String s) {
        breed = s;
    }

    public int compareTo(Dog d) {
        return breed.compareTo(d.getBreed());
    }

}

请注意，当Comparable参数化时，不需要在compareTo中使用对象。