Solving a verbal reasoning test with word embeddings (Analogies)

I tried two different approaches, none of which is a standard functionality to calculate analogies according to the word embeddings model. But they both use functionality from the libraries to calculate the arithmetic of words and the cosine similarity. As this method is often used, I implemented it as a function to keep me in check and debug the code throughout the experiment. However, this approach is really returning the words that best fit the analogy, not the score. This method is really the one that I finally used to calculate scores for analogies.