On Wednesday, November 20, 2013 11:13:32 AM UTC-8, Felix Breuer wrote:
>
> I don't think NumPy will help, as NumPy works with machine precision
> throughout, as far as I was able to figure out.
>
I think you can put arbitrary fixed length types in there, which would
include multiprecision intege
Hi Nils,
excellent point! Adding a special case for gcd = 1 help a *lot* on many
instances.
I'll definitely take the plunge and convert the code to lists or tuples
instead of vectors. (The convenience of writing things like "alpha * v + w"
is certainly not worth the performance penalty.)
I do
Hi Maarten,
a couple of other questions regarding Cython, based on the blog post you
linked:
* Why are MPIR integer operations slower (according to the post) than
Python's own big integer operations? What additional functionality does
MPIR provide that incurs this performance penalty?
* Most
On Wednesday, November 20, 2013 10:54:29 AM UTC-8, Nils Bruin wrote:
> return [c div g for c in v]
>
Sorry, that's spelled [c // g for c in v]
Incidentally, a typical vector of 20 integers has gcd 1 for its
coefficients. If that happens a lot in your data, you should shortcut on
gcd==1 t
On Wednesday, November 20, 2013 2:02:59 AM UTC-8, Felix Breuer wrote:
>
> Hi all!
>
> I have a large collection (~50,000) of integer vectors in low dimension
> (~20). For each of these vectors, I want to divide all entries by their
> gcd. In other words if
>
> def prim_v(v):
> d = abs(gcd(v))
Hi Simon, hi Maarten,
thank you for your answers!
@Simon:
Yes, I figured that calling vector(...) would impose these kinds of
overhead. This is why I am using apply_map in the code, in the hope of
saving some of this work.
Thanks to your suggestion, I tried specifying the base ring in the c
It seems like your code is mostly doing easy integer operations in tight
for loops (although in your case the for loop is hidden in v.apply_map). If
you care about performance in such cases, then you should not use python,
but cython. Because everything you do in python has a small overhead, and
Hi Felix,
On 2013-11-20, Felix Breuer wrote:
> If I am reading this correctly, this means that most of the time (~10
> seconds) is not spent doing the actual computation (using foo and bar) but
> simply creating vectors and read/writing values to/from vectors (in
> apply_map). (Do something li
Hi all!
I have a large collection (~50,000) of integer vectors in low dimension
(~20). For each of these vectors, I want to divide all entries by their
gcd. In other words if
def prim_v(v):
d = abs(gcd(v))
return v.apply_map(lambda vi: vi.divide_knowing_divisible_by(d))
then I want to
