Mailman 3 December 2005 - SciPy-User

fmin_bfgs
by LOPEZ GARCIA DE LOMANA, ADRIAN Dec. 21, 2005

Dec. 21, 2005

Hi all, I'm testing the routine fmin_bfgs from scipy.optimize and I have some problems. I do not understand why it gives me this warning. Warning: Desired error not necessarily achieved due to precision loss <---------------------------- Current function value: 152.114542 Iterations: 5 Function evaluations: 23 Gradient evaluations: 13 ¿Because of the machine (quite new, dual Intel XEON 3.0)? The python is version 2.4.1 The code I'm using is at: http://diana.imim.es/~alopez/questions/code.py Also I plot a graph showing the last value of the function against the norm of the gradient, as you can see at http://diana.imim.es/~alopez/questions/figure.png most of the times is stopping at very low gradient norm values, but other times is stopping while the norm gradient AND the function value is higher than the stopping conditions. Red crosses come from the optimizations given the gradient analytically and green ones from the same optimizations where the gradient has been calculated numerically . Does anyone face the same problem? Is there any problem with the fmin_bfgs routine? Why is stopping while the gradient norm is bigger than gtol? What does it mean that the precission was lost? Any help or commentary will be welcomed, thanks, Adrián.

2 1

As requested, a the OSX install odyssey
by Frank Dec. 21, 2005

Dec. 21, 2005

In response to Ed's request of me several days ago, this seems like a good time (recalling Felis' recent trouble with the scipy_core install on OSX) to relate my troubles with the scipy install on OSX. Before I go on, I should say that I ultimately had no trouble installing from the newly updated instructions at Chris' site after I deleted the directory (thank you, Robert): /System/Library/Frameworks/Python.framework/Versions/2.3/lib/ python2.3/site-packages/ and used the gcc 3.3 compiler. Incidentally, according to the instructions, the latter step seems optional - it is not. I have never gotten anything close to a correct build with gcc 4. Also, I think the former step should be explicitly included in the instructions. Something to the effect of "if you screw it up, then delete this directory." That was a source of pain for me, I had figured (foolishly) that a new build from SVN would just "fix" whatever was wrong in that directory. Of course, it did not. Building from SVN (from the OLDER OSX instructions) was painful and never fully worked. First, install Numeric, however, you see on the numpy sourceforge site a download of the scipy_core "Numeric replacement." That should be better, right? No, it doesn't install without distutils. Oh, there is no mention of those anywhere in the instructions. Back to installing Numeric. Oh, that doesn't work on Tiger without a modification to "customize.py." The readme claims that the installer will just "know" that it is an OSX system and thus "know" where the lapack (vecLib) directories are. That did not work for me so it took me awhile to figure out how just to get Numeric going. No mention of such troubles on the site. In the next step, I see (on the F2PY website, only) that F2PY requires scipy_distutils. The only link I ever found to get them was on the F2PY website. Ok, distutils and F2PY installed. I then followed the CVS link (which linked to SVN instructions) and dutifully downloaded the core and started my build/install. For whatever reason (possibly a bad link from installing some piece along the way via gcc4?), that never worked and I had to download the full scipy from the link on the OSX instruction site. Only after fighting through the many configuration changes (e.g., extra link args) was I able to get anything going. The worst bit was (again, probably due to crap from broken installs stuck in my site_packages directory) that after the core claimed to install, the scipy build/install never finished without errors. The only way I got anything going was to go into the CORE directory (yes, I wrote core; yes, I know it was wrong to do) and cd into the scipy directory and build/install scipy from there. Of course, this is wrong but please let me point out two key bits: 1) the build/install of scipy the correct way never completed (always citing missing core and/or distutils components) 2) I have read on forums (and even the archives of this mailing list) that the numeric replacement core and the core downloadable tarball (linked on the instruction page) are different. When I combined these bits, I arrived at "oh, they must have meant this" type of install of scipy. Viola! It actually worked! I tested out the only function I needed at the time linalg.eig() - success! Of course, this left me with the (now infamous?) installation without interpolate or integrate but a working linalg and fft. I hope I don't sound unappreciative. SciPy is truly great and I look forward to supporting its further development via spreading the word and contributing when I can. Sorry this message was so long, but hey, you (Ed) asked. Thanks, Frank

3 3

scipy.test() runs for >6hrs
by Alex Olivas Dec. 21, 2005

Dec. 21, 2005

I just installed SciPy on a Mac(10.4) lastnight. I'm looking for an alternative to ROOT and if anyone's familiar with that I'd appreciate any comparisons (i.e. what's better, what's worse?) In any case, after installing, the first thing I did was run the test 'scipy.test()' and went to sleep. When I woke up (6hrs later) the tests were still running. I'm assuming this isn't normal? Thanks, Alex.

3 4

Re: [SciPy-user] totally bogged down install on fedora core 4
by felis＠Safe-mail.net Dec. 20, 2005

Dec. 20, 2005

Hi Ed, thank you for pointing me in the right direction. I must say it is difficult when you start using this software to find which versions to use and what to do. Probably more common with packages in transition. I'll try and remove previous attempts and install this version, but i'm not sure where succesful installation of scipy_core will leave me, is it the SciPy, or is it a prerequisite to Scipy? On OSX installation of scipy_core went okay it seems. Thanks, now hopefully on to the GA examples ... felis -------- Original Message -------- From: Ed Schofield <schofield(a)ftw.at> Apparently from: scipy-user-bounces(a)scipy.net To: SciPy Users List <scipy-user(a)scipy.net> Subject: Re: [SciPy-user] totally bogged down install on fedora core 4 Date: Tue, 20 Dec 2005 17:05:24 +0000 > Joshua L. Adelman wrote: > > > What's the difference between the scipy_core version 0.8.4 that's > > available at sourceforge vs the one that's on the svn? According to > > Chris's installation instructions on the scipy website, he recommends > > getting both the core and the full scipy from svn. > > They're similar, since version 0.8.4 is a snapshot from last week. Both > versions should be fine, but there's a small chance of breakage with the > SVN tree if someone's in the middle of changes. > > -- Ed > > _______________________________________________ > SciPy-user mailing list > SciPy-user(a)scipy.net > http://www.scipy.net/mailman/listinfo/scipy-user

2 1

totally bogged down install on fedora core 4
by felis＠Safe-mail.net Dec. 20, 2005

Dec. 20, 2005

hi list, i have trouble getting scipy 0.32 to run on a fc4 installation, i also tried in vain an OSX installation. What ever i try satifying dependencies (python, ATLAS, lapack, djbfft etc.etc.) every avenue seems blocked. Is there a way to install scipy on fc4? I would love to use it but it's a science to get it running in the first place. Get anyone point me to some description of things to do to get it working on fc4 or OSX? I must be missing something. thanks!

3 4

Stats method in scipy.stats
by David Huard Dec. 19, 2005

Dec. 19, 2005

Hi, Can someone explain this ? >>> stats.gamma.stats(1) (nan, nan) >>> stats.norm.stats(loc = 0, scale =1) (nan, nan) and so on for all the distributions I've tried (although other methods such as pdf, rvs and fit work just fine. ) Thanks, David Huard Using scipy version 0.4.3.1478 and scipy core version 0.8.0.1586

1 0

why are the eigenvalues different
by Michael Blazej Dec. 19, 2005

Dec. 19, 2005

for the 3x3 matrix [-2,2,-3;2,1,-6;-1,-2,0] why does matlab give me -3,5,-3 and scipy give me -2,3,-7.5

1 1

what's wrong with radom.multivariate_normal
by jaonary＠free.fr Dec. 19, 2005

Dec. 19, 2005

Hi all, Here I am again with yet another scipy usage problem :-). I'd like to generate a sample of multivariate_normal random variable. Before, I used numeric with RandomArray. This worked well, but the resulting array was misformed. There was an extra comma at the end of each row something like [1,2,]. It's a problem for me because I need to put the result in a function that expect a scipy array as input. After, some google search I see that there's a funcition in scipy.random that sample different rv. When I try to use the one for the multinomial rv I get this error : data1 = random.multivariate_normal([0,0],[[0.5,0],[0,0.5]],[10]) Traceback (most recent call last): File "<pyshell#9>", line 1, in -toplevel- data1 = random.multivariate_normal([0,0],[[0.5,0],[0,0.5]],[10]) File "mtrand.pyx", line 837, in mtrand.RandomState.multivariate_normal AttributeError: 'module' object has no attribute 'singular_value_decomposition' What's up ? Thanks for your help Jaonary

10 16

Failed build (AGAIN)
by Frank Dec. 18, 2005

Dec. 18, 2005

Dear All, What am I doing wrong? Although I have never gotten an explicit build error, I have never been able to import the interpolate of integrate subpackages. After numerous failed attempts, I tried to build from scratch. I deleted the directory: /System/Library/Frameworks/Python.framework/Versions/2.3/lib/ python2.3/site-packages/scipy and then followed the instructions for building from SVN at Chris Fonnesbeck's "Building SciPy for Mac OS X" webpage to the letter. I still cannot import most subpackages such as integrate while others such as linalg and fft import just fine. Incidentally, I also do no have access to "top-level" functions such as derivative or factorial. I have attached a complete build log of the entire process. I would be very grateful for any help you could provide. Thank you so much, Frank

3 5

sequence emulation -/-> array access?
by Alan G Isaac Dec. 18, 2005

Dec. 18, 2005

Is a special array interface necessary for scipy.array to access even simple sequence emulators? E.g., suppose I have >>> class simple: ... def __init__(self,data): ... self.data = data ... where it is enforced that data is a list, and to class this I add sequence emulation http://docs.python.org/ref/sequence-types.html Can someone please point me to a *minimal* example where I can then do x = simple([1,2,3]) y = scipy.array(x) and end up with an array of the list values (just like when you do scipy.array([1,2,3])). Thank you, Alan Isaac

4 9