Mailman 3 May 2009 - NumPy-Discussion

Re: [Numpy-discussion] linalg.svd not working?
by Muhammad Alkarouri 08 May '09

08 May '09

> Date: Tue, 5 May 2009 09:24:53 -0600 > From: Charles R Harris <charlesr.harris(a)gmail.com> ... > This is almost always an ATLAS problem. Where did your > ATLAS come from and > what distro are you running? You are probably right. I compiled and installed ATLAS from source. The distro is Redhat Enterprise Linux 4. I had to because the ones from the distro are compiled targetting 64-bit architecture. I largely followed the instructions at http://www.scipy.org/Installing_SciPy/Linux#head-eecf834fad12bf7a6257525285… . Built lapack 3.1.1, then copied the library to ATLAS and compiled per instructions. The compilers are CC='gcc -m32' version 3.4.4 (enforcing 32 bit compilation), g77 3.4.4. At various points I needed to define flags to ensure 32 bits, though not in numpy compilation. I have gfortran on the system but I didn't use it. liblapack.so and other .so files are linked to libg2c.so not libgfortran. ATLAS (3.8.3) was configured with the additional -32 flag as well. make check works nicely. What should I check in order to find the error with ATLAS configuration and/or installation? Or is there a 32 bit version binary I can download/ use (even if only for testing)? Regards, Muhammad Alkarouri

2 5

Installation NumPy v.1.2.1 on Linux
by Ilya A. Kozyreff 08 May '09

08 May '09

From: Ilya A. Kozyreff [mailto:ia@jscc.ru] Sent: Friday, May 08, 2009 11:16 AM To: 'numpy-discussion(a)scipy.org' Subject: Installation NumPy v.1.2.1 on Linux Hi, all! I try to install NumPy v.1.2.1 on Red Hat Linux as user like this: $ python setup.py install --prefix=/nethome/ia/usr/ $ python -c 'import numpy; numpy.test()' Traceback (most recent call last): File "<string>", line 1, in ? ImportError: No module named numpy What I do wrong? Best regards, Ilya Kozyrev

2 1

replacing Nan's in a string array converted from a float array
by Brennan Williams 08 May '09

08 May '09

I've created an array of strings using something like.... stringarray=self.karray.astype("|S8") If the array value is a Nan I get "1.#QNAN" in my string array. For cosmetic reasons I'd like to change this to something else, e.g. "invalid" or "inactive". My string array can be up to 100,000+ values. Is there a fast way to do this? Thanks Brennan

3 3

element wise help
by Chris Colbert 07 May '09

07 May '09

suppose i have two arrays: n and t, both are 1-D arrays. for each value in t, I need to use it to perform an element wise scalar operation on every value in n and then sum the results into a single scalar to be stored in the output array. Is there any way to do this without the for loop like below: for val in t_array: out = (n / val).sum() # not the actual function being done, but you get the idea Thanks, Chris

3 12

FYI: Numpy and Unladen Swallow
by Bruce Southey 07 May '09

07 May '09

Hi, LWN.net had a article on the development of Unladen Swallow that aims to speed up CPython. http://lwn.net/SubscriberLink/332038/675304c610f0e34a/ The project link is: http://code.google.com/p/unladen-swallow/ At least on my Linux system, Numpy does run without any test failures. It appears slightly faster than my distros Python2.5 (the installation overwrote my local copy of Python2.6 that I built myself so I can not say if is faster than that). Bruce

1 0

optimization when there are mssing values
by Mathew Yeates 07 May '09

07 May '09

What is the canonical method for dealing with missing values? Suppose f(x,y) returns None for some (x,y) pairs (unknown until evaluation). I don't like the idea of setting the return to some small value as this may create local maxima in the solution space. So any of the scipy packages deal with this? Mathew

1 0

hairy optimization problem
by Mathew Yeates 07 May '09

07 May '09

I have a function f(x,y) which produces N values [v1,v2,v3 .... vN] where some of the values are None (only found after evaluation) each evaluation of "f" is expensive and N is large. I want N x,y pairs which produce the optimal value in each column. A brute force approach would be to generate [v11,v12,v13,v14 ....] [v21,v22,v23 ....... ] etc then locate the maximum of each column. This is far too slow ......Any other ideas?

4 6

Recurrence relationships
by Talbot, Gerry 06 May '09

06 May '09

Does anyone know how to efficiently implement a recurrence relationship in numpy such as: y[n] = A*x[n] + B*y[n-1] Thanks, Gerry

6 8

ValueError: dimensions too large.
by natachai wongchavalidkul 06 May '09

06 May '09

Hello alls, I currently have a problem with creating a multi-dimensional array in numpy. The following is what I am trying to do and the error message. >>> test = zeros((3,3,3,3,3,3,10,4,6,2,18,10,11,4,2,2), dtype=float); Traceback (most recent call last): File "<pyshell#39>", line 1, in <module> test = zeros((3,3,3,3,3,3,10,4,6,2,18,10,11,4,2,2), dtype=float); ValueError: dimensions too large. I haven't sure if they should be something to do with the memory or any other suggestions for the way to solve this problem. Anyway, comments or suggestions will be really appreciate though. Thank you _________________________________________________________________ Hotmail® has a new way to see what's up with your friends. http://windowslive.com/Tutorial/Hotmail/WhatsNew?ocid=TXT_TAGLM_WL_HM_Tutor…

2 1

bitwise view on numpy array
by Vincent Schut 06 May '09

06 May '09

Hi, I'm gonna have large (e.g. 2400x2400) arrays of 16 and 32 bit bitfields. I've been searching in vain for an efficient and convenient way to represent these array's individual bit's (or, even better, configureable bitfields of 1-4 bits each). Of course I know I can 'split' the array in its separate bitfields using bitwise operators and shifts, but this will greatly increase the memory usage because it'll create one byte array for each bitfield. So I was looking for a way to create a bitwise view on the original array's data. I've been looking at recarray's, but the smallest element these can use are bytes, correct?. I've been looking at ctypes arrays of Structure subclasses, which can define bitfields. However, these will give me an object array of elements with the Structure class subclass, and only allow me to access the bits per array element instead of for the entire array (or a subset), e.g. data[:].bit17-19 or someting like that. After searching the net in vain for some hours, the list is my last resort :-) Anyone having ideas of how to get both memory-efficient and convenient access to single bits of a numpy array? On a slightly related note, during my search I found some comments saying that numpy.bool arrays use an entire byte for each element. Could someone confirm (or, better, negate) that? Thanks, Vincent.

2 1