我有一个主要类 - 模拟器 - 它使用另外两个类 - Producer 和 Evaluator 。生产者生成结果,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送给评估者来控制执行流程。
生产者和评估者的实际实现在运行时是已知的,在编译时我只知道他们的接口。下面我粘贴接口,示例实现和Simulator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;
}
}
此代码应编译并运行。无论选择哪个生成器实现,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:
所有这一切,模拟器工作。现在,我想摆脱编译警告。问题是,我不知道如何参数化IEkgator和IProducer接口,以及如何更改IProducer和IEvaluator的实现。
我有一些限制:
有什么想法吗?
使用下面答案中的一些想法,我进入了这个阶段,它编译并运行时没有警告,也不需要使用SuppressWarnings指令。这与Eero建议的非常相似,但主要方法略有不同。
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;
}
}
关键区别似乎在主要方法中,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);
}
darn的东西甚至都不会编译,说:绑定不匹配:IEvaluator类型的泛型方法evaluate(Map,Map)不适用于参数(Map&gt;,Map)。推断类型捕获#3-of? extends Comparable不是有界参数&gt;
的有效替代品这对我来说非常奇怪。如果我调用evaluateator.evaluate(producer.getResults(),groundTruth),代码就可以工作。但是,如果我首先调用producer.getResults()方法,并将其存储到变量中,然后使用该变量调用evaluate方法(evaluateator.evaluate(ranking,groundTruth)),我会得到编译错误(无论该变量是什么)类型)。
答案 0 :(得分:2)
您需要指定对象愿意与之比较的内容。类似的东西:
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>());
}
}
答案 1 :(得分:1)
我在下面发布了我的答案。一些说明:
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 %.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;
}
}
答案 2 :(得分:0)
警告是否只是要求你做这样的事情?
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 Comparable 参数化时,无需在 compareTo 中使用对象。