Mailman 3 August 2015 - SciPy-Dev

RBF epsilon (changes 0.15 => 0.16)
by Ariel Rokem 31 Aug '15

31 Aug '15

Hi everyone, In our work on dipy, we've run into a small issue moving from 0.15 to 0.16: https://github.com/nipy/dipy/issues/688 The gist of this is that we are using RBF to interpolate a signal measured on the unit sphere. To do this, we have designed a custom 'norm' function based on the angle between measurements (rather than the default Euclidean). Code is here: https://github.com/nipy/dipy/blob/master/dipy/core/sphere.py#L500 This seemed to work fine with 0.15, but the changes to the default epsilon introduced here: https://github.com/scipy/scipy/pull/4643 seem to be throwing us off. I've taken a look at the discussion on that PR and the issue that led to this work, and it does make sense, but I wanted to get some feedback to see how we should proceed. Setting a larger default epsilon in our code (effectively undoing this change) would be one way to go, but I want to make sure that it wouldn't be an obvious blunder, and that there isn't some issue on scipy that we should be thinking about. Thanks! Ariel

2 2

Need help on scipy central header + logo
by Surya 30 Aug '15

30 Aug '15

Hello, I'm finalizing on scipy central logo now. Thus, looking forward for any suggestions/opinions. I found the theme described in https://github.com/tonysyu/SciPy-Central-Logo interesting, and used it partially. The below are the two variants (I'd prefer the latter - for the reason that the green part isn't visible anyway). Thanks Surya [image: Inline image 1] [image: Inline image 2]

6 6

scipy.sparse.linalg.eigsh routine crashing for negative values of the parameter.
by MANUEL HUMBERTO MUNOZ ARIAS 26 Aug '15

26 Aug '15

Good Day scipy-dev members, Recently I have been working with the diagonalization routines of scipy, I am doing an study if the eigenvalues of a certain matrix, in particular I have interest in the differences between these eigenvalues. In order to do that I am doing the diagonalization process of each of the values of a parameter in a linespace, the diagonalization it has been carried out with the scipy.sparse.linalg.eigsh routine for few eigenvalues, however it seems to crash for negative values of the parameter. Why is this happening? In order to make my point more clear I have attached a plot of the difference between the first two eigenvalues. Thanks in advance, Manuel.

1 0

Python extensions for Python 3.5 - useful info...
by Fernando Perez 26 Aug '15

26 Aug '15

Just an FYI for the upcoming Python release, a very detailed post from Steve Dower, the Microsoft developer who is now in charge of the Windows releases for Python, on how the build process will change in 3.5 regarding extensions: http://stevedower.id.au/blog/building-for-python-3-5/ Cheers, f -- Fernando Perez (@fperez_org; http://fperez.org) fperez.net-at-gmail: mailing lists only (I ignore this when swamped!) fernando.perez-at-berkeley: contact me here for any direct mail

1 0

offline for ~6 weeks
by Ralf Gommers 19 Aug '15

19 Aug '15

Hi all, Just a quick note to say that I'll be offline for pretty much the whole period 25 Aug - 9 Oct. Regarding a possible 0.16.1 release: I don't see a reason to cut a release right now, but if a quick release is required for whatever reason then please go ahead without me. Cheers, Ralf

1 0

removing the Lena image
by Ralf Gommers 16 Aug '15

16 Aug '15

Hi all, The Lena image that we ship is a standard image in the image processing literature, however it's also controversial (see https://en.wikipedia.org/wiki/Lenna). That was the reason for: - adding 2 new images in scipy 0.12.0 ( https://github.com/scipy/scipy/pull/351) - recommending to not using Lena in the documentation in scipy 0.16.0 ( https://github.com/scipy/scipy/pull/4785) Now it looks like Debian was decided that the licensing of this image is NOK and that it needs to be removed from all of Debian. Which hasn't affected Scipy yet but has affected scikit-image (it's "marked for auto-removal"): https://groups.google.com/forum/#!topic/scikit-image/q36bajdvWew https://groups.google.com/forum/#!topic/scikit-image/8d15Egg7qAY I know opinions on the image itself and on Debian's action will vary, but I'd like to be pragmatic here and prevent issues with Scipy packaging in Debian. Therefore I propose to do the same as what scikit-image will do: remove Lena for the next release and raise an error from the scipy.misc.lena() function. Ralf

12 15

Fit method for discrete distributions in scipy.stats?
by Graham Clenaghan 16 Aug '15

16 Aug '15

Hi all, Recently I needed to fit a Zipf distribution to some data for an application and noticed that while scipy has a fit method for the continuous distributions, it has none for any discrete distributions. Is there any reason for this, or has it just not been done yet? There aren't that many discrete distributions and I'd be happy to try to put together a PR to add this, but I'd just like to make sure that there is no specific reason it's not already there before I start writing any code. One possible issue I can think of is that some distributions have discrete shape parameters (i.e. binomial and the number of trials): an MLE estimate would probably return a non-integer (the distributions accept non-integers just fine). Thoughts? Thanks, Graham Clenaghan

2 1

Re: [SciPy-Dev] removing the Lena image
by Cantwell Carson 16 Aug '15

16 Aug '15

My only concern is that existing code might rely on being able to invoke an image with this command. Then, rather than depreciate and remove, why don't we replace the image. After some discussion here, I think the best candidate would be an image of a cat. There are enough pictures of cats available to select one with identical dimensions and similar color characteristics. On Aug 15, 2015 8:40 AM, "Matthew Brett" <matthew.brett(a)gmail.com> wrote:

3 3

solve_banded is very slow as a tridiagonal solver
by Marcel Oliver 15 Aug '15

15 Aug '15

Hello, I noticed that using solve_banded to solve a tridiagonal linear system is at least an order of magnitude slower than it should be based on a straightforward operation count of the Thomas algorithm. In the python source code, solve_banded is pretty much a straightforward call into gbsv from lapack, but I did not dare to follow up on what is going on in there. Maybe it's doing pivoting, or there is some inefficiency due to being coded for more general band-diagonal problems. In any event, in a lot of textbook applications of tridiagonal solvers, pivoting is not necessary (see, e.g., Higham, Accuracy and Stability of Numerical Algorithms, SIAM 1996, Section 9.5) and the Thomas algorithm does not vectorize efficiently. So it would be very good if either solve_banded could be fixed if it does something stupid, or otherwise if a plain Thomas algorithm tridiagonal solver could be added as this would cover a lot of run-off-the-mill problems while solve_banded could handle the rest. In any event, just to demonstrate that the performance gap is really there, I coded up a Thomas algorithm in cython and here is the output for a tridiagonal system of size 10^6: CPU time Thomas algorithm in python: 3.47022 CPU time solve_banded: 0.428928 CPU time Thomas algorithm in cython: 0.032048 Speedup python/cython: 108.281952072 Speedup solve_banded/cython: 13.3839241138 The code is attached, but it's quickly put together and I am not claiming that it's well. In fact, it's the first time I try cython which was easy enough, except that @cython.boundscheck(False) makes the code run slower, not faster... Regards, Marcel

2 2

Weekly Summary 2015/08/12
by Abraham Escalante 13 Aug '15

13 Aug '15

Hello all, Here is last last week's progress and this week's plan for `scipy.stats` improvements. *Week 10 topics:* - NaN policy ready to be merged. - `f_oneway` NIST test cases ready to be merged. *Week 11 topics:* - `mstats` consistency with `stats`: WIP. - `linregress`: same treatment as `f_oneway`. - `shapiro`: WIP - `kendalltau`: standby - `pointbiserialr`: standby For your reference: - Open Pull Requests <https://github.com/scipy/scipy/pulls/aeklant> - Open StatisticsCleanup issues <https://github.com/scipy/scipy/issues?q=is%3Aopen+is%3Aissue+milestone%3ASt…> Regards, Abraham.

1 0