我有一个主类Simulator,它使用另外两个类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,然后将结果传达给评估者来控制执行流程。
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。下面我粘贴了接口、示例实现和Simulator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 %.2fn", 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)。
编译器警告我不要在几个地方使用泛型:
- 在Simulator类、接口IEvaluator和IProducer以及实现IProduceer接口的类中,每当我引用Comparable接口时,我都会收到以下警告:Comparable是一个原始类型。对泛型类型Comparable的引用应参数化
- 在实现IEvaluator的类中,我收到以下警告(当对Map的值调用compareTo()时):类型安全:compareTo方法(Object)属于原始类型Comparable。对泛型类型Comparable的引用应参数化
话虽如此,模拟器仍然有效。现在,我想去掉编译警告。问题是,我不知道如何参数化接口IEvaluator和IProducer,以及如何更改IProduceer和IEvaluater的实现。
我有一些限制:
- 我不知道生产者将返回的Map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Comparable接口
- 类似地,IEvaluator实例不需要知道任何关于正在评估的结果的信息,除非它们是相同类型的并且是可比较的(IEvaluatorimplementation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——它不需要知道任何关于这些类型的信息(除了是相同的类型,这也是可比较的)。它的工作是简单地将结果从生产者传达给评估者
有什么想法吗?
编辑和修订版本
使用下面答案中的一些想法,我进入了这个阶段,它在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 %.2fn", 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主要的方法,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 %.2fn", score);
}
这是有效的。然而,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 %.2fn", score);
}
该死的东西甚至不会编译,说:绑定不匹配:IEvaluator类型的泛型方法evaluate(Map,Map)不适用于参数(Map>,Map)。推断类型捕获#3?extends Comparable不是有界参数的有效替代项>
这对我来说很奇怪。如果我调用evaluator.eevaluate(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中,然后用该变量调用evaluate方法(evaluator.eevaluate(ranks,groundTruth)),我就会得到编译错误(无论该变量的类型如何)。
您需要指定对象愿意将自己与哪些类型的东西进行比较
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>());
}
}
我在下面发布了我的答案。一些注意事项:
- 制片人显然知道他们自己的类型
- 在调用该方法之前,Evaluator不知道它在处理什么
- producerImplementations包含几种不同的类型,因此当您实际选择其中一种类型时,您最终会得到一个转换
以下代码:
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 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 %.2fn", 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;
}
}
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Comparable(或扩展Comparator)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,当参数化Comparable时,不需要在compareTo中使用对象。