erikbern / ann-presentation

Various gfx for a presentation at NYC ML meetup

Home Page:http://www.slideshare.net/erikbern/approximate-nearest-neighbor-methods-and-vector-models-nyc-ml-meetup

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Question about building the trees.

beyondguo opened this issue · comments

Hi @erikbern ,
Thank you so much for your clear illustration in the blog and repo about Annoy, which I learned a lot about ANN.

I have a simple question:
After I randomly pick two points (A & B) in the space and decide a hyperplane to split the space into two parts, how to allocate all the other points into these two parts?
Need I compute the distance between A/B with all the other points in the space to decide which part each point should belong to? Like a brute forece search? That will be so slow.
If not, how does Annoy implement it?

I'm not quite familier with C++, so could you please expalin it breifly to me? Or could you pls tell me the exact position of code in the Annoy repo?

I would appreciate it if you could reply! Thanks!

Need I compute the distance between A/B with all the other points in the space to decide which part each point should belong to? Like a brute forece search? That will be so slow.

Yes. It's not slow – it's just a linear scan