Finalization of changes in conjunction with rOpenSci release.
See the 2.0b release notes for the most important changes, including the cool new syntax (vectorset %>% closest_to(~"king" - "man" + "woman")
.
New functions.
Two new functions.
distend
generalizes reject
and project
to allow partial warping along a vector.
improve_vectorspace
implements some suggestions from this artice (Jiaqi Mu, Suma Bhat, Pramod Viswanath. All-but-the-Top: Simple and Effective Postprocessing for Word Representations. https://arxiv.org/abs/1702.01417)
Faster prep_word2vec
Much faster prep_word2vec
makes it vaguely reasonable.
Name changes
closest_to
is now the default ease-of-use function, with nearest_to
a wrapper for back compatibility. The difference is essentially that closest_to returns a data.frame
with nice naming for integrating into the tidyverse, while nearest_to
returns a named vector. Any code written to work on the 2.0 beta will need to be changed, but all code written for any versions under 1.0 should run as is.
Now exports magrittr's pipe.