When running the below code. this Python 3.6, latest Gensim library in Jupyter

for model in models:print(str(model))pprint(model.docvecs.most_similar(positive=["Machine learning"], topn=20))

[1]: https://github.com/RaRe-Technologies/gensim/blob/develop/docs/notebooks/doc2vec-wikipedia.ipynb

string= "machine learning".split()doc_vector = model.infer_vector(string)
out= model.docvecs.most_similar([doc_vector])

I'm not sure 100% since I'm using a more recent release, but I think that the issue is connected to the fact that the most_similar function is expecting a string mapped in the feature space and not the raw string.