Build a Vector Index From Scratch | Dataford Interview Questions

Implement a custom vector similarity search index structure from scratch using basic math libraries.

Build a sign-based locality-sensitive hashing index from supplied hyperplanes, then rerank candidates by cosine similarity. Implement vector_similarity_search(vectors, queries, hyperplanes, k, max_hamming_distance) and return one list of vector indices per query, ordered by descending similarity and then ascending index. Use only Python's standard library.

Constraints

1 <= len(vectors) <= 10^5
1 <= len(queries) <= 10^3
1 <= len(vectors[i]) == len(queries[j]) <= 256
1 <= len(hyperplanes) <= 20
Every hyperplane has the same dimension as the vectors
1 <= k <= len(vectors)
0 <= max_hamming_distance <= 2
Vectors and queries contain finite real numbers
A query may return fewer than k results if its probed buckets contain fewer than k candidates
Zero vectors and zero queries are valid and have cosine similarity 0

Function Signature

def vector_similarity_search(vectors, queries, hyperplanes, k, max_hamming_distance):

Your question is Build a Vector Index From Scratch. Start with the requirements in the Question tab.

Run and submit as often as you like. When you're ready, talk me through your approach or go straight to the code.

Problem

Implement a custom vector similarity search index structure from scratch using basic math libraries.

Constraints

1 <= len(vectors) <= 10^5
1 <= len(queries) <= 10^3
1 <= len(vectors[i]) == len(queries[j]) <= 256
1 <= len(hyperplanes) <= 20
Every hyperplane has the same dimension as the vectors
1 <= k <= len(vectors)
0 <= max_hamming_distance <= 2
Vectors and queries contain finite real numbers
A query may return fewer than k results if its probed buckets contain fewer than k candidates
Zero vectors and zero queries are valid and have cosine similarity 0

Function Signature

def vector_similarity_search(vectors, queries, hyperplanes, k, max_hamming_distance):

Your question is Build a Vector Index From Scratch. Start with the requirements in the Question tab.

Run and submit as often as you like. When you're ready, talk me through your approach or go straight to the code.