Hi everyone,
It's been a little while since we wrote about NumPy on PyPy, so we wanted to give everyone an update on what we've been up to, and what's up next for us.
We would also like to note that we're launching a funding campaign for NumPy support in PyPy. Details can be found on the donation page.
Some of the things that have happened since last we wrote are:
- We added dtype support, meaning you can now create arrays of a bunch of different types, including bools, ints of a various sizes, and floats.
- More array methods and ufuncs, including things like comparison methods (==, >, etc.)
- Support for more and more argument types, for example you can index by a tuple now (only works with tuples of length one, since we only have single-dimension arrays thus far).
Some of the things we're working on at the moment:
- More dtypes, including complex values and user-defined dtypes.
- Subscripting arrays by other array as indices, and by bool arrays as masks.
- Starting to reuse Python code from the original numpy.
Some of the things on the near horizon are:
- Better support for scalar data, for example did you know that numpy.array([True], dtype=bool)[0] doesn't return a bool object? Instead it returns a numpy.bool_.
- Multi-dimensional array support.
If you're interested in helping out, we always love more contributors, Alex, Maciej, Justin, and the whole PyPy team
7 comments:
What is the best way to contact people about this? Our company has some interest in sponsporing this work, but it wasn't clear from this or the donations page how to actually talk to anyone about it. Maybe I'm missing the obvious.
Anonymous: The address to contact is "pypy at sfconservancy.org". Thanks!
Yay! Time to put my money where my mouth is. :)
What does it mean "Starting to reuse Python code from the original numpy"? If it is copy-paste and something will be changed in numpy git trunc, will it be automatically taken into account by your numpy for PyPy?
This is off topic but, congratulations! You already achieved Unladen Swallow's performance goal of 5x faster than cpython on average.
http://code.google.com/p/unladen-swallow/wiki/ProjectPlan#Performance
http://speed.pypy.org/
You probably have already seen that, but there is an interesting comment from Travis Oliphant about the porting of numpy to pypy :
http://technicaldiscovery.blogspot.com/2011/10/thoughts-on-porting-numpy-to-pypy.html
You haven't answered my question about reuse numpy code for 3 days, I guess because you don't know it overall. I'm not 100% agree with neither Travis opinion nor Stefan M comment from http://morepypy.blogspot.com/2011/09/py3k-for-pypy-fundraiser.html , but in answer to Stefan M you say "Since this is open source, people either work on what they like, because it's fun or scratches their itch" and "Improving the [C extensions] support is boring and frustrating". Guys, AFAIK you received FP7 support for developing some soft for users, not for fun. You should spend some efforts for boring yet important work toward the mentioned things, if you would like to obtain further increase of users number and finance support. Also, clarification about reusing CPython numpy code is also highly appreciated.
Post a Comment