A proposal about Numpypy missing functions
Hello, I have seen that there are some missing functions in numpypy that could be implemented, at least naïvely, very easily. An example of this is zeros_like, as we have zeros and shape. Another example is the random module, easy to port using list comprehension and the random module from the standard library. In any case, based on the benchmarks I have seen, I think the pypy version would be even faster than the pure CPython+Numpy. Would you be interested in a sub-optimal port of those functions? In my opinion, it is better to have something not great (that could be improved) that nothing at all. If so, what should I do? Regards, David.
On 16.06.2012 16:03, Daπid wrote:
Another example is the random module, easy to port using list comprehension and the random module from the standard library. In any case, based on the benchmarks I have seen, I think the pypy version would be even faster than the pure CPython+Numpy.
Would you be interested in a sub-optimal port of those functions? In my opinion, it is better to have something not great (that could be improved) that nothing at all. If so, what should I do?
Hello Daπid, I remember having written a random module for numpy based on stdlibs random module, but somehow it didn't make it into numpypy. Does anyone know where it went? Writing that is a bit problematic, though, unless you know how to make all those distributions work based on a stream of random bits. I didn't at that time.
Regards,
David.
Regards, - Timo
On Sat, Jun 16, 2012 at 4:03 PM, Daπid <davidmenhur@gmail.com> wrote:
Hello,
I have seen that there are some missing functions in numpypy that could be implemented, at least naïvely, very easily. An example of this is zeros_like, as we have zeros and shape. Another example is the random module, easy to port using list comprehension and the random module from the standard library. In any case, based on the benchmarks I have seen, I think the pypy version would be even faster than the pure CPython+Numpy.
Would you be interested in a sub-optimal port of those functions? In my opinion, it is better to have something not great (that could be improved) that nothing at all. If so, what should I do?
The only requirement is that they come with tests (including weird corner cases) Cheers, fijal
Ok, I will start working on that. Hopefully, you will hear of me again. On Mon, Jun 18, 2012 at 11:01 AM, Maciej Fijalkowski <fijall@gmail.com> wrote:
On Sat, Jun 16, 2012 at 4:03 PM, Daπid <davidmenhur@gmail.com> wrote:
Hello,
I have seen that there are some missing functions in numpypy that could be implemented, at least naïvely, very easily. An example of this is zeros_like, as we have zeros and shape. Another example is the random module, easy to port using list comprehension and the random module from the standard library. In any case, based on the benchmarks I have seen, I think the pypy version would be even faster than the pure CPython+Numpy.
Would you be interested in a sub-optimal port of those functions? In my opinion, it is better to have something not great (that could be improved) that nothing at all. If so, what should I do?
The only requirement is that they come with tests (including weird corner cases)
Cheers, fijal
participants (3)
-
Daπid -
Maciej Fijalkowski -
Timo Paulssen