Langchain::Evals::CosineSimilarity to calculate simple similarity between actual and expected outputs

[andreibondarev]

No description provided.

[andreibondarev]

@bborn Okay, I re-packaged it a bit to a small, simple PR. I'm thinking of adding instructions to the README, something like:

# Execute the Assistant.
assistant.add_message_and_run! content: "What is 2+2?"
actual_output = assistant.messages.last.content

# Compare it's output to the expected output
Langchain::Evals::CosineSimilarity.new(llm).score(actual_output: actual_output, expected_output: "4")

( ^^^ Need a better example of course)

[sergiobayona]

@andreibondarev is this only missing the README update?

[Copilot]

Pull Request Overview

This pull request introduces the Langchain::Evals::CosineSimilarity class to compute a similarity score between two output strings by comparing their embedding vectors. The implementation uses an underlying LLM to embed the texts and leverages a cosine similarity utility, with accompanying specifications to ensure correct behavior.

• Adds a new CosineSimilarity class that computes similarity using embeddings.
• Provides an RSpec test to verify that a score of 1.0 is returned when embeddings are identical.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
spec/langchain/evals/cosine_similarity_spec.rb	New test file to verify cosine similarity computation.
lib/langchain/evals/cosine_similarity.rb	Implements the cosine similarity calculation using LLM embeddings.

Comments suppressed due to low confidence (1)

spec/langchain/evals/cosine_similarity_spec.rb:11

• Consider adding additional tests with varied embedding values to verify that the cosine similarity calculation works correctly across different scenarios.

allow(subject.llm).to receive(:embed).and_return(double("Langchain::LLM::OpenAIResponse", embedding: [1, 0, 0]))