单独测量检索召回与排序质量

DoubleSummaryStatistics stats = new DoubleSummaryStatistics();
for (OptionalDouble metric : perQueryMetrics) {
    if (metric.isPresent()) stats.accept(metric.getAsDouble()); // 未定义值不伪装成零。
}
OptionalDouble mean = stats.getCount() == 0 ? OptionalDouble.empty()
        : OptionalDouble.of(stats.getAverage()); // 同时报告有效样本数。

import java.util.*;
public class Main {
    record Metrics(double hit,double recall,double mrr,double ndcg){}
    static double gain(int grade,int rank){return (Math.pow(2,grade)-1)/(Math.log(rank+1)/Math.log(2));}
    static Optional<Metrics> score(List<String> ranked,Map<String,Integer> grades,int k){
        if(k<1)throw new IllegalArgumentException("k");
        Set<String> seen=new HashSet<>();for(String id:ranked)if(id==null||!seen.add(id))throw new IllegalArgumentException("duplicate rank");
        List<Integer> ideal=new ArrayList<>();
        for(int g:grades.values()){if(g<0||g>3)throw new IllegalArgumentException("grade");if(g>0)ideal.add(g);}
        if(ideal.isEmpty())return Optional.empty(); // 未定义不能写成零分或满分。
        ideal.sort(Comparator.reverseOrder());double dcg=0,idcg=0,mrr=0;int hits=0;
        for(int i=0;i<Math.min(k,ranked.size());i++){
            int g=grades.getOrDefault(ranked.get(i),0);
            if(g>0){hits++;if(mrr==0)mrr=1.0/(i+1);} // 只保留首个相关位置。
            dcg+=gain(g,i+1);
        }
        for(int i=0;i<Math.min(k,ideal.size());i++)idcg+=gain(ideal.get(i),i+1); // 理想序列来自全部标注。
        return Optional.of(new Metrics(hits>0?1:0,(double)hits/ideal.size(),mrr,dcg/idcg));
    }
    static void near(double e,double a){if(!Double.isFinite(a)||Math.abs(e-a)>1e-12)throw new AssertionError(a);}
    public static void main(String[] args){
        Map<String,Integer> gold=Map.of("A",3,"B",1);
        Metrics x=score(List.of("X","A"),gold,2).orElseThrow();near(1,x.hit());near(.5,x.recall());near(.5,x.mrr());
        near(1,score(List.of("A","B"),gold,2).orElseThrow().ndcg());
        near((1+7/(Math.log(3)/Math.log(2)))/(7+1/(Math.log(3)/Math.log(2))),score(List.of("B","A"),gold,2).orElseThrow().ndcg());
        near(0,score(List.of(),gold,2).orElseThrow().recall());
        if(score(List.of("A"),Map.of(),2).isPresent())throw new AssertionError();
        try{score(List.of("A","A"),gold,2);throw new AssertionError();}catch(IllegalArgumentException expected){}
        System.out.println("PASS 6 cases");
    }
}