Re: [Numpy-discussion] Buffer interface PEP

>> Is this saying that either NULL or a pointer to "B" can be supplied >> by getbufferproc to indicate to the caller that the array is unsigned >> bytes? If so, is there a specific reason to put the (minor) >> complexity of handling this case in the caller's hands, instead of >> dealing with it int

[Numpy-discussion] Further matrix oddities: no inner product

In mathematics, and inner product is a sesquilinear form on pairs of vectors, so at the least it should return a scalar. In numpy inner is a sum over the last indices. OK, so we have In [10]: inner(ones(2),ones(2)) Out[10]: 2.0 This doesn't work as an inner product for column vectors, which woul

Re: [Numpy-discussion] matrix indexing question

>> On Tue, 27 Mar 2007, Robert Kern apparently wrote: >>> Gram-Schmidt orthogonalization > Alan G Isaac wrote: >> I take it from context that you consider it desirable >> to end up with a list of matrices? Robert wrote: > Honestly, I don't care. You asked about iteration, and > I gave y

Re: [Numpy-discussion] use of concatenate with subclass of ndarray

On Tuesday 27 March 2007 20:08:04 Bryce Hendrix wrote: > We have a class which is a subclass of ndarray which defines > __array_finalize__ to add an attribute. The class looks something like > this: Ahah, we ran into this problem a few months ago: You should not initialize your "units" attribute i

[Numpy-discussion] use of concatenate with subclass of ndarray

We have a class which is a subclass of ndarray which defines __array_finalize__ to add an attribute. The class looks something like this: class UnitArray(ndarray): # ... def __new__(cls, data, dtype=None, copy=True, units=None): # ... arr = array(data, dtype=dtype, copy=copy) res = nd

Re: [Numpy-discussion] Complex 'where' expression

You need to use numpy.logical_and. The and and or operators do not work elementwise on arrays to the best of my understanding. I think it is a limitation on the way logical operators are applied in python or maybe it is an intentional numpy limitation. I'm sure others on the list could explain be

Re: [Numpy-discussion] Buffer interface PEP

Zachary Pincus wrote: > Hi, > > > Is this saying that either NULL or a pointer to "B" can be supplied > by getbufferproc to indicate to the caller that the array is unsigned > bytes? If so, is there a specific reason to put the (minor) > complexity of handling this case in the caller's han

Re: [Numpy-discussion] masked array with mask an instance of numpy.memmap

> Is there something deficient about numpy.core.ma? numpy.core.ma objects are not ndarrays. Therefore, their mask disappear when used as the input of something like numpy.array(data, subok=True), which is quickly problematic. Moreover, they are not especially subclassing-friendly, which was my

Re: [Numpy-discussion] masked array with mask an instance of numpy.memmap

On Tue, Mar 27, 2007 at 05:47:24PM -0500, Glen W. Mabey wrote: > Is there something deficient about numpy.core.ma? Okay, I just found sandbox/maskedarray/CHANGELOG . Glen ___ Numpy-discussion mailing list Numpy-discussion@scipy.org http://projects.scipy

Re: [Numpy-discussion] masked array with mask an instance of numpy.memmap

On Tue, Mar 27, 2007 at 06:35:51PM -0400, Pierre GM wrote: > > > Whoa. You didn't mean numpy SVN, I take it. So, will numpy.median not > > ever take into account the mask? > > Mmh, that should be part of numpy.core.ma instead: let's keep the masked > arrays a bit on the side. > > > And, how re

Re: [Numpy-discussion] masked array with mask an instance of numpy.memmap

> Whoa. You didn't mean numpy SVN, I take it. So, will numpy.median not > ever take into account the mask? Mmh, that should be part of numpy.core.ma instead: let's keep the masked arrays a bit on the side. > And, how recently did you upload them? I updated scipy today and I > still get: Ah,

Re: [Numpy-discussion] matrix indexing question

On Tue, 27 Mar 2007, Fernando Perez wrote: > it's probably worth mentioning, as a data point, that the > python language has a well established precedent for > 'iteration over a FOO where a FOO is yielded': > In [1]: s='abc' Yes. And I recall being bitten by this long ago. Now it feels natura

Re: [Numpy-discussion] A unique function...

Zach, > There's unique and unique1d, but these don't output the number of > occurences. > There's also bincount, which outputs the number of each element, but > includes zeros for non-present elements and so could be problematic > for certain data. Well, I found it anyway. As I was pretty sure it

Re: [Numpy-discussion] masked array with mask an instance of numpy.memmap

On Tue, Mar 27, 2007 at 05:54:05PM -0400, Pierre GM wrote: > On Tuesday 27 March 2007 17:42:18 Glen W. Mabey wrote: > > However, using a memmap for the mask value seems to load the entire > > object into memory, even if it is already of type bool. > > > > How hard would it be to change that behavio

Re: [Numpy-discussion] masked array with mask an instance of numpy.memmap

On Tuesday 27 March 2007 17:42:18 Glen W. Mabey wrote: > However, using a memmap for the mask value seems to load the entire > object into memory, even if it is already of type bool. > > How hard would it be to change that behavior? Oh, I have no idea about that... > Also, it seems that median()

[Numpy-discussion] masked array with mask an instance of numpy.memmap

I've been using numpy.memmap to access large arrays very nicely, but now I desire to instantiate a masked array that uses memmap objects. Using a memmap instance for the data works just fine and as expected. However, using a memmap for the mask value seems to load the entire object into memory,

Re: [Numpy-discussion] matrix indexing question

Alan G Isaac wrote: > On Tue, 27 Mar 2007, Robert Kern apparently wrote: >> Gram-Schmidt orthogonalization > > I take it from context that you consider it desirable > to end up with a list of matrices? Honestly, I don't care. You asked about iteration, and I gave you an example where it was imp

Re: [Numpy-discussion] A unique function...

Hello, There's unique and unique1d, but these don't output the number of occurences. There's also bincount, which outputs the number of each element, but includes zeros for non-present elements and so could be problematic for certain data. Zach On Mar 27, 2007, at 2:28 PM, Pierre GM wrot

[Numpy-discussion] A unique function...

All, I could swear that I ran once into a numpy (or scipy) function that output the unique values of a 1d ndarray along with the number of occurences of each element. If I'm not completely mistaken, it 's a private function initially in Fortran. Does this ring a bell to anyone ? Where could I fi

Re: [Numpy-discussion] Buffer interface PEP

Hi, I have a specific question and then a general question, and some minor issues for clarification. Specifically, regarding the arguments to getbufferproc: > 166 format > 167address of a format-string (following extended struct > 168syntax) indicating what is in each element of > 169

Re: [Numpy-discussion] matrix indexing question

On 3/27/07, Zachary Pincus <[EMAIL PROTECTED]> wrote: So we are left with the current behavior: (a) M[i,:] is a matrix. (b) M[i] is a matrix. (c) Iteration over M yields matrices. I find it useful if M[i] indexes the matrix interpreted as a column-vector (vec(M), or M(:) in Matlab notation),

Re: [Numpy-discussion] matrix indexing question

On 3/27/07, Alan G Isaac <[EMAIL PROTECTED]> wrote: > Hi Zach, > > The use case I requested was for iteration over a > matrix where it is desirable that matrices are yielded. I'm personally not really partial to any side of this discussion, given how I don't use matrices at all (I'm content with a

Re: [Numpy-discussion] matrix indexing question

> The use case I requested was for iteration over a > matrix where it is desirable that matrices are yielded. > That is not what you offered. No, I offered a thorough discussion of why the design of numpy, as I imperfectly understand it, make the trade-offs to achieve your desired goal unpalat

Re: [Numpy-discussion] matrix indexing question

On Tue, 27 Mar 2007, Robert Kern apparently wrote: > Gram-Schmidt orthogonalization I take it from context that you consider it desirable to end up with a list of matrices? I guess I would find it more natural to work with the arrays, but perhaps that starts just being taste. Thank you, Alan I

Re: [Numpy-discussion] matrix indexing question

Alan G Isaac wrote: > Hi Zach, > > The use case I requested was for iteration over a > matrix where it is desirable that matrices are yielded. > That is not what you offered. > > The context for this request is my own experience: > whenever I have needed to iterate over matrices, > I have always

Re: [Numpy-discussion] matrix indexing question

Hi Zach, The use case I requested was for iteration over a matrix where it is desirable that matrices are yielded. That is not what you offered. The context for this request is my own experience: whenever I have needed to iterate over matrices, I have always wanted the arrays. So I am simply int

Re: [Numpy-discussion] matrix indexing question

Hello all, I suspect my previous email did not contain the full chain of my reasoning, because I thought that some parts were basically obvious. My point was merely that given some pretty basic fundamental tenants of numpy, Alan's suggestions quickly lead to *serious issues* far worse than

[Numpy-discussion] Buffer interface PEP

Hi all, I'm having a good discussion with Carl Banks and Greg Ewing over on python-dev about the buffer interface. We are converging on a pretty good design, IMHO. If anybody wants to participate in the discussion, please read the PEP (there are a few things that still need to change) and a

Re: [Numpy-discussion] matrix indexing question

It seems to me that using shapes, (m,1) versus (1,n), to determine whether a vector is column- or row-oriented is a hack (or at least feels like one). If we know we have a vector, then we want to use a single index to obtain a scalar, and that extra "0," or ",0" shouldn't be necessary. It

Re: [Numpy-discussion] matrix indexing question

Alan Isaac wrote: > On Mon, 26 Mar 2007, Travis Oliphant wrote: > >> It actually has been offered. You just don't accept it. >> Matrices are containers of matrices. >> If M is an (mxn) matrix then M[0] is a (1xn) matrix. >> Viewing this 1xn matrix as a 1-d array loses it's row-vectornes

Re: [Numpy-discussion] Complex 'where' expression

Ludwig wrote: > In pseudo code: > > indexes = where(image1 > 0 and image2 > 0 and sigma < quality) > result[indexes] = i # scalar > > > When I run this numpy tells me that the that the truth value of the array > comparison is ambiguous -- but I do not want to compare the whole arrays here, > bu

[Numpy-discussion] Complex 'where' expression

A bit new to numpy, trying to move over some IDL code. I have a complex array selection expression, where I need an array of indexes as a result, so I can use this as a selection expression to apply changes to a different array. I have two images, image1 and image2, plus an array sigma, the resul

Re: [Numpy-discussion] matrix indexing question

Alan G Isaac wrote: > On Mon, 26 Mar 2007, "Colin J. Williams" apparently wrote: >> One would expect the iteration over A to return row >> vectors, represented by (1, n) matrices. > > This is again simple assertion. > **Why** would "one" expect this? > Some people clearly do not. > > One perso

Re: [Numpy-discussion] matrix indexing question

I would like to hear your opinion on developing an explicit sparse/dense 2D matrix class with indexing similar to Matlab, and without significant differences between sparse and dense matrices from the user's perspective... I know that it's not one of Numpy/Scipy's goals to clone Matlab, but I th

Re: [Numpy-discussion] Tuning sparse stuff in NumPy

David Koch wrote: > On 3/27/07, Robert Cimrman <[EMAIL PROTECTED]> wrote: >> >> >> ok. now which version of scipy (scipy.__version__) do you use (you may >> have posted it, but I missed it)? Not so long ago, there was an effort >> by Nathan Bell and others reimplementing sparsetools + scipy.sparse

Re: [Numpy-discussion] defmatrix.py

Charles R Harris wrote: > > > On 3/26/07, *Travis Oliphant* <[EMAIL PROTECTED] > > wrote: > > Charles R Harris wrote: > > > > > > > The rule 1-d is always a *row* vector only applies when > converting > > to a > > matrix. >

Re: [Numpy-discussion] defmatrix.py

Charles R Harris wrote: > > > On 3/26/07, *Travis Oliphant* <[EMAIL PROTECTED] > > wrote: > > > > I think that might be the simplest thing, dot overrides subtypes. > BTW, > > here is another ambiguity > > > > In [6]: dot(array([[1]]),ones(2)) >

Re: [Numpy-discussion] missing use case

Hi Zachary, I think your response highlights very well the apparent design flaw. Here is your response to my request for a use case where iteration over a matrix should yield matrices: do not iterate! Does some part of you not find that just a little bizarre?! As for offering as a use ca

Re: [Numpy-discussion] matrix indexing question

Zachary Pincus wrote: > rest of linear algebra -- e.g. that m[0] yields a matrix if m is a > matrix-- it almost certainly would violate the principle of least > surprise for iteration over m (intuitively understood to be choosing m > [0] then m[1] and so forth) to yield anything other than a m

Re: [Numpy-discussion] Tuning sparse stuff in NumPy

On 3/27/07, Robert Cimrman <[EMAIL PROTECTED]> wrote: ok. now which version of scipy (scipy.__version__) do you use (you may have posted it, but I missed it)? Not so long ago, there was an effort by Nathan Bell and others reimplementing sparsetools + scipy.sparse to get better usability and per

Re: [Numpy-discussion] Tuning sparse stuff in NumPy

David Koch wrote: > Ok, > > I did and the results are: > csc * csc: 372.601957083 > csc * csc: 3.90811300278 a typo here? which one is csr? > csr * csc: 15.3202679157 > csr * csr: 3.84498214722 > > Mhm, quite insightful. Note, that in an operation X.transpose() * X, > where X > is csc_matrix,

Re: [Numpy-discussion] matrix indexing question

> Exactly: that was one other thing I found artificial. > Surely the points will then be wanted as arrays. > > So my view is that we still do not have a use case > for wanting matrices yielded when iterating across > rows of a matrix. It's pretty clear from my perspective: 1-D slices of matrices *

Re: [Numpy-discussion] ATLAS problems

The last time I had this problem i ended up running python in gdb and setting breakpoints at the appropriate dgemm calls. That way i could see exactly what numpy was using to do dgemm. Simon. On Tue, 27 Mar 2007 10:20:41 +1000 "Justin Bedo" <[EMAIL PROTECTED]> wrote: > Hi All, > > I'm trying t

Re: [Numpy-discussion] matrix indexing question

On Tue, 27 Mar 2007, Bill Baxter apparently wrote: > xformedPt = someComplicatedNonLinearThing(pt) > I do stuff like the above quite frequently in my code, > although with arrays rather than matrices. Exactly: that was one other thing I found artificial. Surely the points will then be wanted

Re: [Numpy-discussion] defmatrix.py

Travis Oliphant schrieb: > O.K. So, the problem is that when I defined matrix multiplication, I > mistakenly believed that dot always interpreted 1-d arrays as row > vectors which it is now clear it doesn't. > > I think this led to the issues. > > So, making dot always return arrays seems li