Hi, I've also packaged Numba [1, 2] for Debian Science [3] (group member).
Description: native machine code compiler for Python Numba compiles native machine code instructions from Python programs at runtime using the LLVM compiler infrastructure. It could be easily employed by decorating individual computation intensive functions in the Python code. Numba could significantly speed up the performance of computations, and optionally supports compilation to run on GPU processors through Nvidia's CUDA platform. It integrates well with the Python scientific software stack, and especially recognizes Numpy arrays. It's going into Experimental at first, where llvmlite currently is (it's written on the top of that). It needs some more finetuning, and the testsuite needs a fix for a subprocess import error, but runs all right (please see the examples in the doc package). Like Theano, I think it would belong into the Numerical Computation task, it's also build on the top of Numpy. Thanks, DS [1] http://numba.pydata.org/ [2] https://bugs.debian.org/743877 (ITP: numba -- NumPy-aware optimizing compiler for Python) [3] https://anonscm.debian.org/cgit/debian-science/packages/numba.git (not yet populated) -- 4096R/DF5182C8 http://www.danielstender.com/blog/
