You may have to manually isntall numba for UMAP to run through

pip install -U numba

python -m spacy download en_core_web_sm

Also don't forget to setup an environment file which looks similar to this:

 EN_CORPUS="/home/david/_MasterThesis/data/corpus/english/training-monolingual/news.2007.en.shuffled.sample"
 SEMCOR_CORPUS="/home/david/_MasterThesis/data/semcor-corpus/semcor/semcor/"
 TOP_20000_EN="/home/david/_MasterThesis/data/most_common/20k.txt"

I think this is a proper way to fix out of embedding vectors (and I think this should be done to keep the expeirment fair w.r.t. BERT)

https://discuss.pytorch.org/t/updating-part-of-an-embedding-matrix-only-for-out-of-vocab-words/33297/3