MediumNLPnlp-173

Embedding Drift Rate

Problem

Return the average embedding drift 1 - cosine(old, new) over words present in both vector dictionaries.

Return 0.0 when there is no overlap between the two vocabularies.

Hint

Cosine distance is 1 minus cosine similarity, and the rate averages it over shared words.

case 1

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

expected: 0.5

case 2

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

expected: 0

case 3

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

expected: 0

case 4

in: [{},{}]

expected: 0

case 5

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

expected: 2

Starter codePython

import math
def embedding_drift_rate(old_vectors, new_vectors):
    # Your code here
    pass

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