I was looking to build a n-gram prediction model with tidymodels and came across textrecipes which looks very helpful.
I set up a few steps like step_tokenize() and step_ngram(), but I'm stuck wondering if there's any way I can define the recipe to build the data set to be features [first word, second word], then the predicted class being the third word. I think I may need to build this more manually but wanted to see if anyone has any suggestions on if this is currently possible. Thanks for any advice.