Join Our 5-Week ML/AI Engineer Interview Bootcamp 🚀 led by ML Tech Leads at FAANGs
43. Mean Reciprocal Rank
GeneralMean Reciprocal Rank (MRR) is a common evaluation metric for NLP retrieval and ranking tasks where each query has exactly one correct answer. You’ll compute MRR from ranked prediction lists by finding the rank position of the first relevant item per query.
Requirements
Implement the function
Rules:
- For each query
q, find the 1-based index oftargets[q]insiderankings[q], then compute its reciprocal. - If the target is not present in the ranking list, its reciprocal rank is
0.0. - Return the average reciprocal rank across all queries.
- Use NumPy for vectorized operations where helpful.
- Don’t use any prebuilt ranking-metric functions (e.g., from
sklearnortorchmetrics).
Example
Output:
| Argument | Type |
|---|---|
| targets | np.ndarray |
| rankings | list[np.ndarray] |
| Return Name | Type |
|---|---|
| value | float |
Constraints
Input
rankingsis a list of NumPy arrays;targetsis a NumPy arrayNo sklearn/torchmetrics ranking functions
Missing target contributes reciprocal rank 0.0
Hint 1
MRR is the average over queries of 1 / (1-based rank of the target); if the target isn’t in the list, contribute 0.
Hint 2
For each (rank_list, target), find the target’s index. In Python you can use list.index with a try/except, or with NumPy use np.where(rank_list == target).
Hint 3
Implementation pattern: loop with zip(rankings, targets), compute recip = 0.0 if no match else 1.0/(idx+1), append, then return float(np.mean(recip_ranks)).
Roles
Companies
GeneralLevels
Tags
Python Editor