Here is what has been done in January thanks to the funding of NumPyPy, I would like to thank all the donors and tell you that you can still donate :
- I have focused on implementing the object dtype this month, it is now possible to store objects inside ndarrays using the object dtype
- It is also possible to add an object ndarray to any other ndarray (implementing other operators is trivial)
The next things I plan on working on next are :
- Implementing the missing operations for object arrays
- Implementing garbage collection support for object arrays (currently, storing an object inside an ndarray doesn't keep the object alive)
- Packaging NumPyPy on PyPI
Cheers
Romain
Thanks for the post! This sounds pretty cool.
ReplyDeleteThe previous post suggested that there would be an update in regards to linalg. Does this mean linalg is working? Is having a working linalg what stands in the way of a working matplotlib? Thanks for answering what might be a naive question!
Linalg is basically usable with the usual caveats: use PyPy 2.5.0 or later, use pypy/numpy from the bitbucket repo, you can even use matplotlib from my fork at https://github.com/mattip/matplotlib but there is no gui backend available yet, so you can only save the plots to files. Watch this space for the promised blog post, hopefully next week.
ReplyDeleteGreat to hear there is some progress on numpy!
ReplyDeleteAbout matplotlib @mattip. Maybe a GSoC project for the GUI?
Regarding matplotlib, I whipped up a quick hack that can do at least very simple matplotlib stuff. Based on running a "slave" CPython using RpyC, as I recall was already done in 2011 or so demos.
ReplyDeleteSimple stuff can run unmodified, although can be of course slow if there's a lot or frequent data passing from PyPy to CPython.
Could be probably quite easily done in other direction to, ie running PyPy from CPython.
https://github.com/jampekka/cpyproxy