Are SemanticSimilarity and AnswerSimilarity supposed to be different?

[sanjeed5]

Problem

The collections API has two implementations that appear to be identical:

• SemanticSimilarity (_semantic_similarity.py)
• AnswerSimilarity (_answer_similarity.py)

The implementations are 100% identical - same cosine similarity algorithm, same parameters. The only differences are:

1. Class name
2. Default metric name: "semantic_similarity" vs "answer_similarity"

Context

In the legacy API (_answer_similarity.py), AnswerSimilarity was simply a subclass alias of SemanticSimilarity:

class AnswerSimilarity(SemanticSimilarity):
name: str = "answer_similarity"When migrating to collections API, both were implemented as separate standalone classes with duplicated code.

Current State

• SemanticSimilarity has documentation and is in the metrics index
• AnswerSimilarity has no documentation and is not in the metrics index
• Both are exported in ragas.metrics.collections.__all__

Question

Should we consolidate these, or is there a reason to keep both as separate implementations?

[anistark]

Good catch. Can merge them together with AnswerSimilarity removed. with docs updated.