In their recent KDD article [1], Etsy used 200 SVD dimensions. For those interested in trying this out in Python: * `gensim` contains stochastic SVD for large

In their recent KDD article [1], Etsy used 200 SVD dimensions. For those interested in trying this out in Python: * `gensim` contains stochastic SVD for large data (fast online model training) [2] * I wrote a benchmark of (approximate) nearest neighbour libraries in Python [3] [1] https://dl.dropboxusercontent.com/u/2143857/papers/topics.pdf https://dl.dropboxusercontent.com/u/2143857/papers/topics.pd... [2] https://github.com/piskvorky/gensim/ https://github.com/piskvorky/gensim/ [3] http://radimrehurek.com/2013/12/performance-shootout-of-nearest-neighbours-contestants/ http://radimrehurek.com/2013/12/performance-shootout-of-near...