Mailman 3 April 2017 - SciPy-Dev

Are you using Bento to build scipy?
by CJ Carey July 16, 2018

July 16, 2018

With the 1.0 release coming up, I figured this is a good time to take the pulse of the community regarding Bento. The project seems to have been abandoned before reaching stability, and it seems that the core problems it was trying to address have been at least partially fixed by the more official build systems. If you use Bento and would like to keep using it post 1.0, or if my previous sentence is wrong, please let me know!

10 15

Sparse multi-dimensional array library
by Matthew Rocklin April 26, 2017

April 26, 2017

Hello everyone, I would like to call attention to this small library for sparse multidimensional arrays: https://github.com/mrocklin/sparse . This is a small library that enables some computations with sparse arrays of dimension greater than two. It relies heavily on the numpy and scipy.sparse packages. More information about capabilities and intent is available on the project's readme available through the link above. Feedback is welcome. Also, while this library is strictly single-core and in-memory, work has been done to make it compatible to work smoothly with dask.arrays (in a development branch). Best, -matthew

1 0

Finding a fixed point on the simplex
by Erik Brinkman April 20, 2017

April 20, 2017

I recently wrote some code <https://github.com/egtaonline/gameanalysis/blob/master/gameanalysis/fixedpo…> that uses simplicial subdivision to compute an approximate fixed point for functions on a simplex. It's guaranteed to always return an approximate fixed point, but may take exponential time. It's particularly useful for functions which have fixed points but no basin of attraction, and so won't converge via iteration (see the code at at the bottom that's looking for a fixed point in a degenerate case of rock paper scissors). Does this warrant inclusion into scipy? Generally computing fixed points of continuous functions over compact sets seems useful to many communities, however there are a couple of issues I see with blindly putting this in scipy based off of the fact that a function for computing fixed points <https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.fixed_p…> already exists. - The proposed function works for cases where the existing function fails, which is an argument for inclusion. - The existing function is in optimize, but the proposed one is not an optimization algorithm, it's a search algorithm. - The proposed function only works for functions on the simplex. Most functions on compact sets should have homotopies that convert them into functions on the simplex, but the current fixed_point makes no such restriction. - The proposed function has different parameters than the current one. A few functions in optimize have signatures that take an `options` dictionary of parameters, but it would change the signature of fixed_point. - The algorithm is slightly more general than computing a fixed point of a simplex function. It actually finds fully labeled subsimplices for an arbitrary `proper` labeling function. From my very limited understanding, there are other labeling functions which produce approximate solutions to other problems. However, given this more general function, it's proper module seems less certain. In addition, the code as it stands is pure python, but it is relatively computationally intensive, and so potentially could be sped up by switching to a c or cython implementation. Conditioned on thinking that this would make a valid PR, is it worth looking into implementing it in a lower level language despite the fact that this makes a large number of calls to a supplied python function? Thanks, Erik Example ~~~~~~~ import numpy as np from scipy import optimize from gameanalysis import fixedpoint A = np.array([[[1, 0, 0], [2, 0, 0], [2, 0, 0]], [[0, 1, 0], [1, 0, 0], [1, 1, 0]], [[0, 0, 1], [1, 0, 1], [1, 0, 0]], [[0, 2, 0], [0, 1, 0], [0, 2, 0]], [[0, 1, 1], [0, 0, 1], [0, 1, 0]], [[0, 0, 2], [0, 0, 2], [0, 0, 1]]]) B = np.array([[ 0, 0, 0], [-2, 1, 0], [ 1, 0, -3], [ 0, 0, 0], [ 0, -3, 1], [ 0, 0, 0]]) FP = np.array([0.3129473 , 0.40598569, 0.28106701]) def fixed_point_func(x): x = np.maximum(x, 0) x /= x.sum() y = np.sum(np.prod(x ** A, -1) * B, 0) z = np.maximum(0, y - y.dot(x)) + x z /= z.sum() return z def main(): close = dict(rtol=1e-3, atol=1e-3) x0 = np.ones(3) / 3 # Verify fixed point assert np.allclose(FP, fixed_point_func(FP), **close) # Use guaranteed method fp = fixedpoint.fixed_point(fixed_point_func, x0, tol=1e-8) assert np.allclose(FP, fp, **close) # Expect error with scipy method try: optimize.fixed_point(fixed_point_func, x0) assert False, "shouldn't get this far" except RuntimeError: pass # Expected # Expect error with scipy method try: optimize.fixed_point(fixed_point_func, x0, method='iteration') assert False, "shouldn't get this far" except RuntimeError: pass # Expected if __name__ == '__main__': main()

1 0

new release of weave
by Ralf Gommers April 19, 2017

April 19, 2017

Hi all, In https://github.com/scipy/weave/issues/8 it was pointed out that the current release of Weave is incompatible with newer SciPy versions. So a new release seems in order. Does anyone see an issue with just making a new release from the current state of master? Or with making the version number 0.16.0 (no relation to scipy 0.16.0 but that seems okay)? Cheers, Ralf

2 2

FFTLog
by Dieter Werthmüller April 17, 2017

April 17, 2017

Dear SciPy-devs, I recently had the need for a logarithmic FFT routine, and did a quick f2py around Andrew Hamilton's FFTLog: http://casa.colorado.edu/~ajsh/FFTLog I put my f2py-setup and my pyf-file up on GitHub: https://github.com/prisae/fftlog I thought it might be a useful addition to the SciPy FFTPack. Specifically as three out of the five fortran-files, of which FFTLog consists, are already in scipy/fftpack. I have never contributed to SciPy and therefore do not know how much work it would involve to get it into SciPy. However, I thought I ask. If there is interest I expect that it would not take long for an experienced person, as it is a fairly small addition. Or someone could point me to the right direction on what to do to get it in. I append some comments on the involved files of FFTLog and some comments regarding their licenses. I believe it is compatible with SciPy. Thanks for all your good work! Dieter Files and Licenses/Permissions ------------------------------ Files of FFTLog [1] cdgamma.f [2] drfftb.f [3] drfftf.f [4] drffti.f [5] fftlog.f (plus a test routine, fftlogtest.f) -- [1] -- The original FFTLog states about this file: FFTLog uses [...] and a modified version of the complex Gamma function from the gamerf package at momonga.t.u-tokyo.ac.jp/~ooura/gamerf.html. The original gamerf copyright statement states: Copyright(C) 1996 Takuya OOURA (email: ooura(a)mmm.t.u-tokyo.ac.jp). You may use, copy, modify this code for any purpose and without fee. You may distribute this ORIGINAL package. Permission to distribute the modified gamma function code with the FFTLog package has been granted (email from Takuya Ooura to Andrew Hamilton dated 16 March 1999). Hence I think it is compatible with SciPy, or am I wrong? Alternatively it could be replaced with scipy.special.loggamma, I think, but I do not know if it is possible to mix Fortran and Python code with f2py. -- [2], [3], [4] -- They are from the NCAR suite of FFT routines (Swarztrauber 1979). As far as I can see they are already in SciPy, in scipy/fftpack/src/dfftpack/ (dfftb.f, dfftf.f, and dffti.f). So there is definitely not a licensing problem here. -- [5] -- This is the actual FFTLog routine from Hamilton. I asked Hamilton for permission before publishing his source code at github.com/prisae/fftlog. His email response was: -- start email from 28/09/2016 -- Dieter, You are welcome to use fftlog in any way you choose. Please note the credits commented in the code: c FFTLog uses the NCAR suite of FFT routines, c and a modified version of the complex Gamma function c from the gamerf package at c http://momonga.t.u-tokyo.ac.jp/~ooura/gamerf.html . c The original gamerf copyright statement states: c Copyright(C) 1996 Takuya OOURA (email: ooura(a)mmm.t.u-tokyo.ac.jp). c You may use, copy, modify this code for any purpose and c without fee. You may distribute this ORIGINAL package. c c Permission to distribute the modified gamma function code c with the FFTLog package has been granted c (email from Takuya Ooura to Andrew Hamilton dated 16 March 1999). Andrew -- end email from 28/09/2016 --

5 22

Problem to apply changes to scipy.special.eval_jacobi
by Evgeny Zhurko April 16, 2017

April 16, 2017

Hi all, I want to fix this issue https://github.com/scipy/scipy/issues/7001 I made the fix, but i can't include it in scipy. I changed eval_jacobi function in https://github.com/scipy/scipy/blob/master/scipy/special/orthogonal_eval.pxd and tried to reinstall scipy, but i don't get changes in eval_jacobi. I also tried to run generate_ufuncs.py once again and install again scipy, but i don't get fix in eval_jacobi. What am i doing wrong? Best regards, Evgeny Zhurko.

1 0

Matrix subspace angle
by Eric Larson April 11, 2017

April 11, 2017

I don't see a SciPy equivalent for the MATLAB/Octave `subspace` to compute the subspace angle between two matrices. Did I miss it somewhere? If we don't have it and people think it would be worthwhile, I could add one. A computational formula is given on page 2014 of this article: http://epubs.siam.org/doi/abs/10.1137/S1064827500377332 Eric

1 0

rv_scatter - companion to rv_histogram
by Andrew Nelson April 5, 2017

April 5, 2017

I've just put up a PR, 7257, that proposes to add rv_scatter, a companion to rv_histogram (6801). It also picks up some aspects of 6466. It allows one to specify a PDF from x/y scatter data. The PDF is piecewise linearly interpolated between the supplied data points. The rationale behind this is to be able to use experimentally determined data to create a statistical distribution. It can also be used to approximate other distributions. Comments are welcomed. A.

1 0

GSOC application : implement scipy.diff
by Yizheng Zhao April 4, 2017

April 4, 2017

Hello developers: I am Yizheng Zhao, a graduate students at Carnegie Mellon University majoring in Software Engineering. And I am interested for your diff project. I’ve been using spicy for 2 years for scientific computing and data science and I am excited that I have this opportunity to make my own contribution to the community. Why me? I majored in Math in college and I do a lot of project about numerical analysis. I also have strong coding skills. Here is my proposal, I am happy to receive any suggestions: ) https://github.com/YizhengZHAO/scipy/wiki/GSoC-2017-:-Implement-scipy.diff-… <https://github.com/YizhengZHAO/scipy/wiki/GSoC-2017-:-Implement-scipy.diff-…> Thanks, Yizheng Zhao

2 4

GSoC applications - next steps
by Ralf Gommers April 3, 2017

April 3, 2017

Hi prospective GSoC participants, The GSoC application deadline is just over a week away, so I thought it'd be useful to give a short overview of what will happen over the next few weeks. Until April 3 we'll do our best to help you write your proposals. Some of you have already shared a draft proposal; I strongly encourage the rest of you to do the same. If you have questions, please make them as concrete as possible (showing that you've done your own reading/research on the topics helps for getting answers). The week after proposal submission we'll interview you (over Skype or Google Hangouts) to help us pick the best applicants. If you are unavailable that week, please let me know now and we'll try to schedule an interview earlier (later is difficult, because we need to indicate how many slots we want to the PSF around the 11th). You'll find out on May 4th if your application was successful. If so, the "community bonding" period starts the next day - this is the time to get more familiar with the project and community: read the developer guide, start interacting on the mailing list, and ensure you're fully set up to hit the ground running on May 30th. April 3: student application deadline April 4-11: SciPy mentors interview students May 4: accepted students announced May 5-30: community bonding period If you have any questions about this, please do ask! Cheers, Ralf

1 1