Prodigy terms.teach does not give any meaningful suggestions with en_vectors_web_lg , en_core_web_lg

Hello, new user here. First of all well done on the excellent library, I'm loving it so far.

I've been using your tool with success for several CV tasks, but I'm struggling replicating the insult classifier from the Youtube video.

For both en_vectors_web_lg and en_core_web_lg, no matter the seeds I give I seem to get the most generic words as suggestions. I've tried the example from insults, my own seeds, and
prodigy terms.teach prog_lang_terms en_vectors_web_lg --seeds Python,C++,Ruby
from the documentation, but all results were words like
you re doin got not a somethin need ca

I saw a related topic, but there OP solved it by using a new env. I'm on a new env, with spacy installed recently. I'm not well versed in Spacy, so any pointers on how to debug this would be helpful. Thanks in advance

spacy==2.3.0
prodigy==1.10.0

Hi! As a quick sanity check, which versions of the vector model packages do you have installed? You should be able to just run pip list in your env and they should show up in the list.

I think this was fixed for terms.teach with spaCy v2.3 in prodigy 1.10.1.

1 Like

These are the versions in my environment

en-core-web-lg               2.3.0
en-core-web-sm               2.3.0
en-vectors-web-lg            2.3.0

@skoetje Thanks! I didn't see that you weren't running the latest version of Prodigy, so as Adriane suggested above, could you try and upgrade to the latest v1.10.3 and try again? :slightly_smiling_face:

I've updated Prodigy, which caused a new error with missing strings in the vector space. However, I saw a related issue posted before and that solution solved it. Thanks a lot for the quick help!