🚀 Feature
(Follow up to #3569 and #2631)
After adding HitRate@top_k and MRR@top_k metrics for recommendation systems, it makes sense to add NDCG@top_k as well.
NDCG complements MRR by considering all relevant items in the ranking, not just the first one. This makes it useful for:
- Recommendation systems evaluation
- RAG (Retrieval Augmented Generation) pipeline evaluation
- Information retrieval benchmarking
Similar to MRR, the implementation would:
- Follow the same structure as
HitRate and MRR
- Support multiple
top_k values in a single call
- Use ranx as reference implementation for tests
🚀 Feature
(Follow up to #3569 and #2631)
After adding
HitRate@top_kandMRR@top_kmetrics for recommendation systems, it makes sense to addNDCG@top_kas well.NDCG complements MRR by considering all relevant items in the ranking, not just the first one. This makes it useful for:
Similar to MRR, the implementation would:
HitRateandMRRtop_kvalues in a single call