GPU acceleration? (feature) about redm

[ha0ye]

Distributed computing is one solution. A development version of rEDM used C++11 threading, but I commented out the code because Rtools at the time did not support that feature. I haven't checked since to see if that situation has changed. There are probably ways of using a distributed network for computation, but I'm not familiar with the technology for doing so, and it's outside the scope of my research at the moment.

Another option is algorithmic improvements. We currently do the naive "calculate all distances and find the nearest neighbors" method, but k-d trees are a standard alternative (at least for low dimensionality, I think?). In higher dimensions, it's less effective (one of the corollaries to curse of dimensionality), but I would guess different distance metrics might be preferred, and then you'd want a custom algo. anyway.

Regardless, I encourage you to fork the package and have a go if you like (esp. if you just want to play around with the C++11 multi-threading code).