Mean Average Precision at K (MAP@K)medium

Mean Average Precision at K (MAP@K)

medium Classical ML from Scratch 15 min

NumPy allowed · plain Python earns 10 bonus points

A query retrieves context

Embedding similarity ranks the documents before generation.

Open lesson →

01query

02embed

03retrieve

04context

What is RAG?→ embed →top context

retrieval.md44%

rag-systems.md33%

agents.md24%

Cosine-like similarity scores ranked before context assembly

Implement Average Precision@K for one user, the per-query building block of the MAP@K metric used to evaluate recommendation ranking quality.

Task

Implement average_precision_at_k(ranked_items, relevant_items, k). Walk ranked_items[:k] in order; every time you hit a relevant item, compute precision-so-far (hits found / position, 1-indexed) and record it. Return the average of those recorded precision values, or 0.0 if no relevant item was found in the top k.

Function Signature

average_precision_at_k(ranked_items: list[str], relevant_items: set, k: int) -> float

Examples

Example 1: all relevant, front-loaded
Input: ranked_items = ['a', 'b', 'c']
relevant_items = ['a', 'b', 'c']
k = 3
Output: 1

Example 2: no relevant found
Input: ranked_items = ['a', 'b']
relevant_items = ['z']
k = 2
Output: 0

Example 3: relevant items spread out
Input: ranked_items = ['a', 'x', 'b', 'y']
relevant_items = ['a', 'b']
k = 4
Output: 0.8333333333333333

Constraints

k >= 1

Python3Saved ✓

def average_precision_at_k(ranked_items, relevant_items, k):

# Your implementation here

pass

Case 1: all relevant, front-loaded

Input: ranked_items = ['a', 'b', 'c']
relevant_items = ['a', 'b', 'c']
k = 3

Expected: 1

Case 2: no relevant found

Input: ranked_items = ['a', 'b']
relevant_items = ['z']
k = 2

Expected: 0

Case 3: relevant items spread out

Input: ranked_items = ['a', 'x', 'b', 'y']
relevant_items = ['a', 'b']
k = 4

Expected: 0.8333333333333333