Sense2Vec options appear to be invalid for terms.train-vectors

I’m attempting to use terms.train-vectors to build a sense2vec model. The latest (and the last 2 versions as well I believe) Prodigy documentation shows --s2v-merge-ents and --s2v-merge-nps as valid flags, but I keep getting the following errors

Any thoughts? Does train-vectors not actually support sense2vec?

Sorry I think there’s a mistake in the docs here.

You can get the details help for any command with --help. This should show you the correct spelling of the two flags you’re interested in, whose shortcuts are -ME and -MN.

Incidentally, note that the implementation of the merge_noun_phrases is currently a bit different in Prodigy from the sense2vec repo. The implementation in the recipes file (which you can read and modify — the source for that is shipped within Prodigy) doesn’t drop articles (“the”, “a”, etc), unlike the function in sense2vec’s merge_text. So that’s a change you might want to make.

1 Like