search

plasticityai / magnitude

A fast, efficient universal vector embedding utility package.
MIT License
1.63k stars 120 forks source link

Feature Request: Option to update individual embeddings #58

Closed Santosh-Gupta closed 5 years ago

Santosh-Gupta commented 5 years ago

As mentioned in this thread https://github.com/plasticityai/magnitude/issues/32

The ability to quickly and easily update individual vectors in a Magnitude object can make this library a powerful tool for sparse training of embeddings. This can be implemented very easily in Pytorch, and fairly straight forward in Keras.

AjayP13 commented 5 years ago

We cannot do this unfortunately due to the other features of the package, namely the most_similar spatial index created by Annoy. That index would have to be updated as well, which is not easy to do, so this is out-of-scope for now.