How to initialize LDA with a set of seed words using gensim package

I have read several papers in which they actually initialize parameteres using a seed set of words for LDA. does anyone know how is this possible in gensim pac开发者_开发知识库kage?

---

For the sake of completeness, copy&paste from the gensim mailing list reply:

Seeding with existing documents was a part of gensim some versions back, in the code that was directly ported from LDA-C. It was meant to improve convergence (although the final, converged result was the same).

In recent versions, that code was replaced by a more efficient algorithm which doesn't use seeding anymore. It is an online (mini-batch) algorithm, so you could say it does "seeding" automatically, in a more principled manner.