On Monday, 7 January 2013 15:53:20 UTC+8, Konrad Hinsen wrote: > Mikera writes: > > > Initial tests seem to suggest that supporting N-dimensional arrays > > should be pretty easy in terms of the API itself. We could also > > provide fast-paths for 1D and 2D arrays. > > > > Of course, actual support for N-dimensional arrays would depend on > > the underlying implementation. > > There are a few array implementations that support N-d arrays, for example > the ones in JHDF5 (https://wiki-bsse.ethz.ch/display/JHDF5) and in netCDF > (http://www.unidata.ucar.edu/software/netcdf-java/). > > > Yep, the idea is to be flexible enough to support many different > implementations. > > The real problem I see in making all this practically usable is > efficient conversion between implementations. The fragmentation of the > scientific computing ecosystem in Java makes it nearly impossible to > get any real work done without some conversion between array > libraries. If I want to read a matrix from an HDF5 file and then > calculate its eigenvalues, I need to convert from HDF5 arrays to Colt > (or whatever) arrays. Some libraries provide a way to construct an N-d > array from a 1-d data storage area plus a shape vector, without > copying the data. I think it's important to leverage such > functionality in a Clojure API in order to let conversion happen as > much as possible behind the scenes. >
I've actually got this working to an extent via a "coercion" protocol that can be extended to different implementations. It's not perfect yet but it works, e.g. you can do (+ m [[2 0] [0 2]]) And it is smart enough to figure out how to coerce the nested persistent vectors into something that the implementation for matrix m understands. There's a lot of scope to make this much more efficient of course.... > > BTW, I started my own attempt at something similar a while ago, but > never found the time to get it to a usable state: > > https://code.google.com/p/clj-multiarray/ > > Konrad. > Very interesting, thanks! I like the use of protocols to construct the specialised array types in particular. Would you be willing and interested to work on merging this into the "core.matrix" code base? I had started on a simple NDArray implementation myself but your code is clearly much further along than mine..... -- You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com Note that posts from new members are moderated - please be patient with your first post. To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en
