The problem will not be LDA. It will be in the vector deserialization. I think you are 80% or better likelihood to be fine.
It should be easy to build a small experiment using two shards. Also, your use case *should* work, so if it doesn't it is reasonable to fix it. It should not be that difficult to fix. On Thu, Jul 16, 2009 at 10:50 AM, Florian Leibert <[email protected]> wrote: > run LDA over the vector outputs and > looking at the code of LDA I'm not sure if I can update this > incrementally... I'll have to take another look but I wanted to > clarify this before running a > large indexing job over my data... >
