Does not work well on trained doc2vec model

I trained a doc2vec model on the [large movie review dataset](http://ai.stanford.edu/~amaas/data/sentiment/) and then tried to use UMAP to reduce the dimensions of the resulting document vectors. I had hoped that it would be possible to separate the documents by sentiment (positive and negative), but unfortunately the embedding is one big blob. A notebook can be seen [here](https://github.com/gclen/umap-doc2vec-testing/blob/master/umap_with_imdb.ipynb) and the rest of the files for training the doc2vec model are in that [repository](https://github.com/gclen/umap-doc2vec-testing) as well.

Start with the linked umap_with_imdb.ipynb and the linked repository's doc2vec training files; reproduce the reduction on the trained document vectors and compare the embedding with the reported sentiment expectation. Determine whether the blob reflects UMAP behavior or the input embeddings, then document the conclusion or next steps.

Written by the indexing model from the issue text.