George Nurser wrote:
> running config_fc
> unifing config_fc, config, build_clib, build_ext, build commands
> --fcompiler options
> running build_clib
> customize UnixCCompiler
> customize UnixCCompiler using build_clib
> building 'npymath' library
> compiling C sources
> C compiler: gcc -arch ppc
You're right, that's a little inconsistent. I would also prefer to get an
overflow for divide by 0 rather than casting to zero.
- ralf
On Sun, Jun 7, 2009 at 12:22 AM, wrote:
> On Sat, Jun 6, 2009 at 11:49 PM, Ralf Gommers
> wrote:
> > Hi,
> >
> > I expect `reciprocal(x)` to calculate 1/x, an
On Sat, Jun 6, 2009 at 5:07 PM, George Nurser wrote:
> Hi,
> the current svn version 7039 isn't compiling for me.
> Clean checkout, old numpy directories removed from site-packages..
> Same command did work for svn r 6329
>
> [george-nursers-macbook-pro-15:~/src/numpy] agn% python setup.py
> confi
On Sat, Jun 6, 2009 at 11:49 PM, Ralf Gommers
wrote:
> Hi,
>
> I expect `reciprocal(x)` to calculate 1/x, and for input 0 to either follow
> the python rules or give the np.divide(1, 0) result. However the result
> returned (with numpy trunk) is:
>
np.reciprocal(0)
> -2147483648
>
np.div
Hi,
On Jun 6, 2009 3:11pm, Chris Colbert wrote:
it definately found your threaded atlas libraries. How do you know
it's numpy is using lapack_lite?
I don't, actually. But it is importing it. With python -v, this is the
error I get if I don't set LD_LIBRARY_PATH to my scipy_build director
Hi,
On Jun 6, 2009 3:11pm, Chris Colbert wrote:
it definately found your threaded atlas libraries. How do you know
it's numpy is using lapack_lite?
I don't, actually. But it is importing it. With python -v, this is the
error I get if I don't set LD_LIBRARY_PATH to my scipy_build director
Hi,
I expect `reciprocal(x)` to calculate 1/x, and for input 0 to either follow
the python rules or give the np.divide(1, 0) result. However the result
returned (with numpy trunk) is:
>>> np.reciprocal(0)
-2147483648
>>> np.divide(1, 0)
0
>>> 1/0
Traceback (most recent call last):
File "", lin
On Fri, Jun 5, 2009 at 06:09, David Cournapeau wrote:
> Please test it ! I am particularly interested in results for scipy
> binaries on mac os x (do they work on ppc).
Test suite passes on Intel Mac OS X (10.5.7) built from source:
OK (KNOWNFAIL=6, SKIP=21)
Cheers
Adam
__
Fernando Perez wrote:
>
> On Sat, Jun 6, 2009 at 11:03 AM, Charles R
> Harris wrote:
>
>> I don't think we can change the current matrix class, to do so would
>> break
>> too much code. It would be nice to extend it with an explicit inner
>> product,
>> but I can't think of any simple notation
On Sun, Jun 7, 2009 at 12:53 AM, Robin wrote:
> I haven't seen this before - is it something wrong with my build or
> the current svn state? I am using macports python 2.5.4 on os x 10.5.7
Hmmm... after rebuilding from the same version the problem seems to
have gone away... sorry for the noise...
Hi,
I just updated to latest numpy svn:
In [10]: numpy.__version__
Out[10]: '1.4.0.dev7039'
It seemed to build fine, but I am getting a lot of errors testing it:
--
Ran 178 tests in 0.655s
FAILED (errors=138)
Out[8]:
Almost al
On 6/6/2009 6:39 PM Robert Kern apparently wrote:
> Ah, that's the beauty of .flat; it takes care of that for you. .flat
> is not a view onto the memory directly. It is a not-quite-a-view onto
> what the memory *would* be if the array were contiguous and the memory
> directly reflected the layout a
Hi,
the current svn version 7039 isn't compiling for me.
Clean checkout, old numpy directories removed from site-packages..
Same command did work for svn r 6329
[george-nursers-macbook-pro-15:~/src/numpy] agn% python setup.py
config_fc --fcompiler=gnu95 build_clib --fcompiler=gnu95 build_ext
--fco
On Sat, Jun 6, 2009 at 17:34, Alan G Isaac wrote:
> On 6/6/2009 6:02 PM Keith Goodman apparently wrote:
>>> def fill_diag(arr, value):
>> if arr.ndim != 2:
>> raise ValueError, "Input must be 2-d."
>> if arr.shape[0] != arr.shape[1]:
>> raise ValueError, 'Input must be squar
On 6/6/2009 6:02 PM Keith Goodman apparently wrote:
>> def fill_diag(arr, value):
> if arr.ndim != 2:
> raise ValueError, "Input must be 2-d."
> if arr.shape[0] != arr.shape[1]:
> raise ValueError, 'Input must be square.'
> arr.flat[::arr.shape[1]+1] = value
You might
On Sat, Jun 6, 2009 at 2:01 PM, Robert Kern wrote:
> There is a neat trick for accessing the diagonal of an existing array
> (a.flat[::a.shape[1]+1]), but it won't work to implement
> diag_indices().
Perfect. That's 3x faster.
def fill_diag(arr, value):
if arr.ndim != 2:
raise Value
On Sat, Jun 6, 2009 at 3:00 PM, Alan G Isaac wrote:
> On 6/6/2009 4:30 PM Robert Kern apparently wrote:
> > The old idea of introducing RowVector and ColumnVector would help
> > here. If x were a ColumnVector and A a Matrix, then you can introduce
> > the following rules:
> >
> > x.T is a RowVect
> On Sat, Jun 6, 2009 at 1:59 PM, Alan G Isaac For sure GAUSS does. The result of x' * A * x
> is a "matrix" (it has one row and one column) but
> it functions like a scalar (and even more,
> since right multiplication by it is also allowed).
On 6/6/2009 4:32 PM Charles R Harr
On Sat, Jun 6, 2009 at 16:00, Alan G Isaac wrote:
> On 6/6/2009 4:30 PM Robert Kern apparently wrote:
>> The old idea of introducing RowVector and ColumnVector would help
>> here. If x were a ColumnVector and A a Matrix, then you can introduce
>> the following rules:
>>
>> x.T is a RowVector
>> Ro
On Sat, Jun 6, 2009 at 13:30, Fernando Perez wrote:
> On Sat, Jun 6, 2009 at 12:09 AM, Robert Kern wrote:
>
>> +1
>
> OK, thanks. I'll try to get it ready.
>
>> diag_indices() can be made more efficient, but these are fine.
>
> Suggestion? Right now it's not obvious to me...
Oops! Never mind. I
On 6/6/2009 4:30 PM Robert Kern apparently wrote:
> The old idea of introducing RowVector and ColumnVector would help
> here. If x were a ColumnVector and A a Matrix, then you can introduce
> the following rules:
>
> x.T is a RowVector
> RowVector * ColumnVector is a scalar
> RowVector * Matrix is
On Sat, Jun 6, 2009 at 2:30 PM, Robert Kern wrote:
> On Sat, Jun 6, 2009 at 14:59, Alan G Isaac wrote:
> > On 6/6/2009 2:58 PM Charles R Harris apparently wrote:
> >> How about the common expression
> >> exp((v.t*A*v)/2)
> >> do you expect a matrix exponential here?
> >
> >
> > I take your point
On Sat, Jun 6, 2009 at 15:31, Bruce Southey wrote:
> While not trying to be negative, this raises important questions that
> need to be covered because the user should not have to do trial and
> error to find what actually works and what that does not. While
> certain features can be fixed within
Thanks for the summary! I'm +1 on points 1, 2 and 3.
+0 for points 4 and 5 (assume_unique keyword and renaming arraysetops).
Neil
PS. I think you mean deprecate, not depreciate :)
___
Numpy-discussion mailing list
Numpy-discussion@scipy.org
http://ma
On Sat, Jun 6, 2009 at 1:59 PM, Alan G Isaac wrote:
> On 6/6/2009 2:58 PM Charles R Harris apparently wrote:
> > How about the common expression
> > exp((v.t*A*v)/2)
> > do you expect a matrix exponential here?
>
>
> I take your point that there are conveniences
> to treating a 1 by 1 matrix as a
On Sat, Jun 6, 2009 at 2:01 AM, Fernando Perez wrote:
[snip]
>
>
> def mask_indices(n,mask_func,k=0):
> """Return the indices for an array, given a masking function like
> tri{u,l}."""
> m = np.ones((n,n),int)
> a = mask_func(m,k)
> return np.where(a != 0)
>
>
> def diag_indices(n,
On Sat, Jun 6, 2009 at 14:59, Alan G Isaac wrote:
> On 6/6/2009 2:58 PM Charles R Harris apparently wrote:
>> How about the common expression
>> exp((v.t*A*v)/2)
>> do you expect a matrix exponential here?
>
>
> I take your point that there are conveniences
> to treating a 1 by 1 matrix as a scala
On 6/6/2009 2:58 PM Charles R Harris apparently wrote:
> How about the common expression
> exp((v.t*A*v)/2)
> do you expect a matrix exponential here?
I take your point that there are conveniences
to treating a 1 by 1 matrix as a scalar.
Most matrix programming languages do this, I think.
For sur
Thanks for this excellent recipe.
I have not tried it out myself yet, but I will follow the instruction on
clean Ubuntu 9.04 64-bit.
Best,
Minjae
On Sat, Jun 6, 2009 at 11:59 AM, Chris Colbert wrote:
> since there is demand, and someone already emailed me, I'll put what I
> did in this post. I
i need the full link to pastebin.com in order to view your post.
It will be something like http://pastebin.com/m6b09f05c
chris
On Sat, Jun 6, 2009 at 2:32 PM, Richard Llewellyn wrote:
> I posted the setup.py build output to pastebin.com, though missed the
> uninteresting stderr (forgot tcsh co
On Sat, Jun 6, 2009 at 12:34 PM, Olivier Verdier wrote:
> I took that very seriously when you said that matrices were important to
> you. Far from me the idea of forbidding numpy users to use matrices.
> My point was the fact that newcomers are confused by the presence of both
> matrices and arra
On Sat, Jun 6, 2009 at 11:35 AM, Gael
Varoquaux wrote:
> I think they need examples. Right now, it is not clear at all to me what
> they do.
Cheers,
f
# With doctests, set to be repeatable by seeding the rng.
def structured_rand_arr(size, sample_func=np.random.random,
lt
On 6/6/2009 2:03 PM Charles R Harris apparently wrote:
> So is eye(3)*(v.T*v) valid? If (v.T*v) is 1x1 you have incompatible
> dimensions for the multiplication
Exactly. So it is not valid. As you point out, to make it valid
implies a loss of the associativity of matrix multiplication.
Not a goo
On Sat, Jun 6, 2009 at 11:46 AM, Keith Goodman wrote:
> On Sat, Jun 6, 2009 at 11:30 AM, Fernando Perez wrote:
>> On Sat, Jun 6, 2009 at 12:09 AM, Robert Kern wrote:
>>> diag_indices() can be made more efficient, but these are fine.
>>
>> Suggestion? Right now it's not obvious to me...
>
> I'm i
On Sat, Jun 6, 2009 at 11:30 AM, Fernando Perez wrote:
> On Sat, Jun 6, 2009 at 12:09 AM, Robert Kern wrote:
>> diag_indices() can be made more efficient, but these are fine.
>
> Suggestion? Right now it's not obvious to me...
I'm interested in a more efficient way too. Here's how I plan to adap
On Sat, Jun 6, 2009 at 11:03 AM, Charles R
Harris wrote:
> I don't think we can change the current matrix class, to do so would break
> too much code. It would be nice to extend it with an explicit inner product,
> but I can't think of any simple notation for it that python would parse.
Maybe it'
On Sat, Jun 06, 2009 at 11:30:37AM -0700, Fernando Perez wrote:
> On Sat, Jun 6, 2009 at 12:09 AM, Robert Kern wrote:
> - Any interest in also having the stuff below? I'm needing to build
> structured random arrays a lot (symmetric, anti-symmetric, symmetric
> with a particular diagonal, etc), an
I took that very seriously when you said that matrices were important to
you. Far from me the idea of forbidding numpy users to use matrices.
My point was the fact that newcomers are confused by the presence of both
matrices and arrays. I think that there should be only one
matrix/vector/tensor obj
I posted the setup.py build output to pastebin.com, though missed the
uninteresting stderr (forgot tcsh command to redirect both).
Also, used setup.py build --fcompiler=gnu95.
To be clear, I am not certain that my ATLAS libraries are not found. But
during the build starting at line 95 (pastebin.c
On Sat, Jun 6, 2009 at 8:27 AM, Keith Goodman wrote:
> What do you think of passing in the array a instead of n and ndim
> (diag_indices_list_2 below)?
Yes, I thought of that too. I see use cases for both though. Would
people prefer both, or rather a flexible interface that tries to
introspect t
On Sat, Jun 6, 2009 at 12:09 AM, Robert Kern wrote:
> +1
OK, thanks. I'll try to get it ready.
> diag_indices() can be made more efficient, but these are fine.
Suggestion? Right now it's not obvious to me...
A few more questions:
- Are doctests considered enough testing for numpy, or are se
On Sat, Jun 6, 2009 at 9:29 AM, Alan G Isaac wrote:
> On 6/6/2009 12:41 AM Charles R Harris apparently wrote:
> > Well, one could argue that. The x.T is a member of the dual, hence maps
> > vectors to the reals. Usually the reals aren't represented by 1x1
> > matrices. Just my [.02] cents.
>
> Of
and where exactly are you seeing atlas not found? during the build
process, are when import numpy in python?
if its the latter, you need to add a .conf file in /etc/ld.so.conf.d/
with the line /usr/local/rich/src/scipy_build/lib and then run sudo
ldconfig
Chris
On Sat, Jun 6, 2009 at 1:42 P
can you run this and post the build.log to pastebin.com:
assuming your numpy build directory is /home/numpy-1.3.0:
cd /home/numpy-1.3.0
rm -rf build
python setup.py build &&> build.log
Chris
On Sat, Jun 6, 2009 at 1:37 PM, Richard Llewellyn wrote:
> Hi Chris,
> thanks much for posting those
Hi Chris,
thanks much for posting those installation instructions. Seems similar to
what I pieced together.
I gather ATLAS not found. Oops, drank that beer too early.
I copied Atlas libs to /usr/local/rich/src/scipy_build/lib.
This is my site.cfg. Out of desperation I tried search_static_fir
when you build numpy, did you use site.cfg to tell it where to find
your atlas libs?
On Sat, Jun 6, 2009 at 1:02 PM, Richard Llewellyn wrote:
> Hello,
>
> I've managed a build of lapack and atlas on Fedora 10 on a quad core, 64,
> and now (...) have a numpy I can import that runs tests ok. :] I
Hello,
I've managed a build of lapack and atlas on Fedora 10 on a quad core, 64,
and now (...) have a numpy I can import that runs tests ok. :]I am
puzzled, however, that numpy builds and imports lapack_lite. Does this mean
I have a problem with the build(s)?
Upon building numpy, I see the tr
since there is demand, and someone already emailed me, I'll put what I
did in this post. It pretty much follows whats on the scipy website,
with a couple other things I gleaned from reading the ATLAS install
guide:
and here it goes, this is valid for Ubuntu 9.04 64-bit (# starts a
comment when wo
On 6/6/2009 12:41 AM Charles R Harris apparently wrote:
> Well, one could argue that. The x.T is a member of the dual, hence maps
> vectors to the reals. Usually the reals aren't represented by 1x1
> matrices. Just my [.02] cents.
Of course that same perspective could
lead you to argue that a M
On Sat, Jun 6, 2009 at 12:01 AM, Fernando Perez wrote:
> def diag_indices(n,ndim=2):
> """Return the indices to index into a diagonal.
>
> Examples
>
> >>> a = np.array([[1,2,3,4],[5,6,7,8],[9,10,11,12],[13,14,15,16]])
> >>> a
> array([[ 1, 2, 3, 4],
> [ 5,
On Fri, Jun 5, 2009 at 2:37 PM, Chris Colbert wrote:
> I'll caution anyone from using Atlas from the repos in Ubuntu 9.04 as the
> package is broken:
>
> https://bugs.launchpad.net/ubuntu/+source/atlas/+bug/363510
>
>
> just build Atlas yourself, you get better performance AND threading.
> Buildi
On Sat, Jun 6, 2009 at 4:42 AM, Neil Crighton wrote:
> Robert Cimrman ntc.zcu.cz> writes:
>
>> Anne Archibald wrote:
>>
>> > 1. add a keyword argument to intersect1d "assume_unique"; if it is not
>> > present, check for uniqueness and emit a warning if not unique
>> > 2. change the warning to an
Robert Cimrman ntc.zcu.cz> writes:
> Anne Archibald wrote:
>
> > 1. add a keyword argument to intersect1d "assume_unique"; if it is not
> > present, check for uniqueness and emit a warning if not unique
> > 2. change the warning to an exception
> > Optionally:
> > 3. change the meaning of the fun
On Fri, Jun 05, 2009 at 06:02:09PM -0400, Alan G Isaac wrote:
> I think something close to this would be possible:
> add dot as an array method.
> A .dot(B) .dot(C)
> is not as pretty as
> A * B * C
> but it is much better than
> np.dot(np.dot(A,B),C)
> In fact it is so much bett
On Sat, Jun 6, 2009 at 02:01, Fernando Perez wrote:
> Howdy,
>
> I'm finding myself often having to index *into* arrays to set values.
> As best I can see, in numpy functions/methods like diag or tri{u,l}
> provide for the extraction of values from arrays, but I haven't found
> their counterparts
Howdy,
I'm finding myself often having to index *into* arrays to set values.
As best I can see, in numpy functions/methods like diag or tri{u,l}
provide for the extraction of values from arrays, but I haven't found
their counterparts for generating the equivalent indices.
Their implementations ar
56 matches
Mail list logo
