Instructions on how to make word2vec deterministic?

[mralex...@gmail.com]

I am training word2vec with the same corpus but receiving different embeddings in multiple runs. In my code, I have the following settings:

i) np.random.seed(0)

ii) random.seed(0)

iii) Word2Vec(doc, seed=0, workers=1)

The code is in Python 3 and the code was running in different screens called from os.system() (i.e., os.system('screen -dm python ...'))

Am I missing anything?

[Radim Řehůřek]

Check out the FAQ: https://github.com/RaRe-Technologies/gensim/wiki/Recipes-&-FAQ#q11-ive-trained-my-word2vecdoc2vecetc-model-repeatedly-using-the-exact-same-text-corpus-but-the-vectors-are-different-each-time-is-there-a-bug-or-have-i-made-a-mistake-2vec-training-non-determinism

…and the PYTHONHASHSEED mentioned there.

HTH,

Radim