MediumNLPnlp-254

HyDE Concept Score

Problem

Score a toy HyDE expansion. Build a hypothetical document vector as the weighted sum of concept_vectors for every term in expansions, skipping terms with no vector, then return its cosine similarity with query_vector.

Return 0.0 when the hypothetical document or the query has zero norm.

Hint

Weighted sum first, cosine second; missing concept vectors are simply ignored.

Test cases

case 1

in: [[1,0],{"a":1,"b":1},{"a":[1,0],"b":[0,1]}]

expected: 0.7071067811865475

case 2

in: [[1,0],{},{"a":[1,0]}]

expected: 0

case 3

in: [[1,0],{"a":1,"z":5},{"a":[1,0],"b":[0,1]}]

expected: 1

case 4

in: [[0,0],{"a":1},{"a":[1,0]}]

expected: 0

case 5

in: [[0,1],{"a":2},{"a":[0,3]}]

expected: 1

Starter codePython

import math
def hyde_concept_score(query_vector, expansions, concept_vectors):
    # Your code here
    pass

5 test cases · run your code against them in the editor