SoftCosineSimilarity broken in gensim==3.7 [AssertionError (WordEmbeddingSimilarityIndex problem)]

[tedo.v...@gmail.com]

I trained w2v model on my own corpus and try to use it to calculate SoftCosineSimilarity.

Folowing https://github.com/RaRe-Technologies/gensim/blob/develop/docs/notebooks/soft_cosine_tutorial.ipynb, when execute 

similarity_index = WordEmbeddingSimilarityIndex(w2v_model)

I've got AssertionError.

  File "/usr/local/lib/python2.7/dist-packages/gensim/models/keyedvectors.py", line 1389, in __init__
    assert isinstance(keyedvectors, WordEmbeddingsKeyedVectors)
AssertionError

This w2v model is 100% OK, I have been use it for Word2Vec based Cosine Similarity, Word2Vec based Word Mover's Distance and Word2Vec based Euclidean Distance.

Something went wrong with gensim==3.7 as I said in one of previous posts.

If I use old code, which worked on gensim==3.6

similarity_matrix = w2v_model.wv.similarity_matrix(dictionary)

now in gensim==3.7 gives mi error:

  File "/usr/local/lib/python2.7/dist-packages/gensim/utils.py", line 1447, in new_func1
    return func(*args, **kwargs)
  File "/usr/local/lib/python2.7/dist-packages/gensim/models/keyedvectors.py", line 660, in similarity_matrix
    index, dictionary, tfidf=tfidf, nonzero_limit=nonzero_limit, dtype=dtype)
  File "/usr/local/lib/python2.7/dist-packages/gensim/similarities/termsim.py", line 234, in __init__
    for term, similarity in index.most_similar(t1, num_rows)
  File "/usr/local/lib/python2.7/dist-packages/gensim/models/keyedvectors.py", line 1401, in most_similar
    for t2, similarity in most_similar:
TypeError: 'numpy.float32' object is not iterable

If I downgrade to gensim==3.6 the old code is working just fine.

[tedo.v...@gmail.com]

Probably doesn't matter but it is on Python2.7.

Please check this bug.