RecommenderEvaluator接口用于实现对特定算法的评估. 目前实现对于ranking的评估器有AUC, AveragePrecision, AverageReciprocalHitRank, Diversity, HitRate, IdealDCG, Normalized, Precision, Recall, REciprocalRank十类评估器. 对于rating实现评估器MAE, MPE, MSE, RMSE四类.
若在配置文件中不进行评估器的指定, 最终评估时使用对应类型的全部评估器.
指定评估器的配置示例:
rating:
rec.recommender.isranking=false
rec.eval.enable=true
rec.eval.class=mse # if ratingranking:
rec.recommender.isranking=true
rec.eval.enable=true
rec.eval.class=auc
rec.recommender.ranking.topn=10不同评估器相应的简写见下表
| Evaluator | configuration |
|---|---|
| AUCEvaluator | auc |
| AveragePrecisionEvaluator | ap |
| AverageReciprocalHitRankEvaluator | arhr |
| DiversityEvaluator | diversity |
| HitRateEvaluator | hitrate |
| IdealDCGEvaluator | idcg |
| NormalizedDCGEvaluator | ndcg |
| PrecisionEvaluator | precision |
| RecallEvaluator | recall |
| ReciprocalRankEvaluator | rr |
| MAEEvaluator | mae |
| MPEEvaluator | mpe |
| MSEEvaluator | mse |
| RMSEEvaluator | rmse |