An implementation of "semantle" (see https://semantle.novalis.org/) for the Toki Pona conlang (see https://tokipona.org/).
Usage:
$ python3
>>> exec(open('semantemusi.py').read())
>>> guess('your_guess')
Returns a number between 0-1 for similarity.
The target word is stored in a variable secret_word, so you can cheat and look at it, or poke around with other model queries, like model.most_similar(secret_word).
Corpus sources:
- "mdm" (Matthew Dean Martin) files: https://github.com/matthewdeanmartin/tokipona.parser/tree/master/TokiPonaTools/TokiPona/corpus
- "pu": https://gist.github.com/increpare/10d94b6d7f223dd0f07c73b10dadc020
If you want to (re)train on these corpora or your own, you can use the (included) train-models.py file, or save your own trained models in the models/ folder (in word2vec KeyedVectors format).