Mailman 3 August 2010 - NumPy-Discussion

Changing a matrix element into a scalar
by Wayne Watson Aug. 3, 2010

Aug. 3, 2010

How do I access 1.2 in such a way as to end up with a float? I keep getting a matrix. from numpy import matrix m = matrix([[1.2],[2.3]]) -- Wayne Watson (Watson Adventures, Prop., Nevada City, CA) (121.015 Deg. W, 39.262 Deg. N) GMT-8 hr std. time) Obz Site: 39° 15' 7" N, 121° 2' 32" W, 2700 feet "Republicans are always complaining that government is out of control. If they get into power, they will prove it." -- R. J. Rourke Web Page:<www.speckledwithstars.net/>

6 8

floating point arithmetic issue
by Guillaume Chérel Aug. 3, 2010

Aug. 3, 2010

Hello, I ran into a difficulty with floating point arithmetic in python. Namely that: >>> 0.001 + 1 - 1 0.00099999999999988987 And, as a consequence, in python: >>> 0.001 + 1 - 1 == 0.001 False In more details, my problem is that I have a fonction which needs to compute (a + b - c) % a. And for b == c, you would expect the result to be 0 whatever the value of a. But it isn't... >>> (0.001 + 1 - 1) % 0.001 0.00099999999999988987 Is there any way to solve this? Many thanks, Guillaume

9 15

numpy.vectorize fails, howto avoid hardcoding parameters?
by Alex Kraus Aug. 2, 2010

Aug. 2, 2010

Hi, I am trying to create a function that calculates the integral of another function. The integral function should later be used in scipy.optimize.leastsq(f, ...), so ideally it should have the format: def f(x, *param) so that it works for a variable number of parameters. While my code works for a fixed number of parameters I cannot get it to work with a variable number of parameters. It seems that numpy.vectorize fails here. Is there a different/better way to do this? from scipy.integrate import quad import numpy as np def integrand_function(x, a, b, c): result = a*x**2 + np.exp(b*x) + np.cos(a*c) return result def define_integral(f, lower, upper): assert(lower < upper) def function(a, b, c): result = quad(f, lower, upper, args=(a, b, c))[0] return result return np.vectorize(function) def integrand_function_param(x, *param): a, b, c = param result = a*x**2 + np.exp(b*x) + np.cos(a*c) return result def define_integral_param(f, lower, upper): assert(lower < upper) def function(a, *param): print(param) result = quad(f, lower, upper, args=(a, param))[0] return result return np.vectorize(function) a = np.array([1,2,3,4]) print(integrand_function(1,2,3,4)) # 21.9400368894 f = define_integral(integrand_function, 0.0, 2.0) print(f(a, 1,2)) # [ 8.22342909 10.41510219 16.30939667 16.7647227 ] print(integrand_function_param(1,2,3,4)) # 21.9400368894 fp = define_integral_param(integrand_function_param, 0.0, 2.0) print(fp(a, 1,2)) # ValueError: mismatch between python function inputs and received arguments # fp should later be used in scipy.optimize.leastsq(fp, ... Any help is very appreciated! Alexander

2 1

accumulated sum-of-squared-differences
by Gregory, Matthew Aug. 2, 2010

Aug. 2, 2010

Hi all, I'm trying to calculate accumulated sum-of-squared-differences for an array in the following manner: import numpy as np a = np.array([1, 2, 3, 49., 50, 51, 98, 99, 100], dtype=np.float32) # Calculate accumulated means over all elements means = np.add.accumulate(a) / (np.arange(a.size) + 1) # Create a matrix of squared differences (elements minus means) diff_sqr = np.asarray(np.matrix(a) - np.matrix(means).T) ** 2 # Sum the lower triangular elements ssd = np.tril(diff_sqr).sum(axis=1) Is there an easier or more terse way to do this? Ideally, I'd like to accumulate over all slices in 'a' as well, ie. not just [0:n], but also [1:n], [2:n], etc. I see some functions within scipy.stats.stats, but I'm struggling on figuring out how to put it together. thanks, matt

1 0

[ANN] ALGOPY 0.21, algorithmic differentiation in Python
by Sebastian Walter Aug. 2, 2010

Aug. 2, 2010

I'm happy to announce the first official release of ALGOPY in version 0.2.1. Rationale: ~~~~~~~~ The purpose of ALGOPY is the evaluation of higher-order derivatives in the forward and reverse mode of Algorithmic Differentiation (AD) using univariate Taylor polynomial arithmetic. Particular focus are functions that contain numerical linear algebra functions (e.g. inv, dot, eigh, qr, cholesky,...) as they often appear in statistically motivated functions. Download: ~~~~~~~~~ http://pypi.python.org/pypi/algopy/0.2.1 or bleeding edge versions from from http://github.com/b45ch1/algopy Documentation: ~~~~~~~~~~~~ available at http://packages.python.org/algopy/ OS Support: ~~~~~~~~~~ Linux, Windows (tested with pythonxy), should also work on Mac Software Dependencies: ~~~~~~~~~~~~~~~~~~~~ for the core: numpy, scipy for testing: nose Exampe Session: ~~~~~~~~~~~~~ Consider the contrived example where it is the goal to compute the directional derivative df/dx_1 : >>> import numpy; from numpy import log, exp, sin, cos >>> import algopy; from algopy import UTPM, dot, inv, zeros >>> >>> def f(x): ... A = zeros((2,2),dtype=x) ... A[0,0] = numpy.log(x[0]*x[1]) ... A[0,1] = numpy.log(x[1]) + exp(x[0]) ... A[1,0] = sin(x[1])**2 + cos(x[0])**3.1 ... A[1,1] = x[0]**cos(x[1]) ... return log( dot(x.T, dot( inv(A), x))) ... >>> >>> x = UTPM(zeros((2,1,2),dtype=float)) >>> x.data[0,0] = [1,2] >>> x.data[1,0] = [1,0] >>> y = f(x) >>> >>> print 'normal function evaluation f(x) = ',y.data[0,0] normal function evaluation f(x) = 0.641250189986 >>> print 'directional derivative df/dx1 = ',y.data[1,0] directional derivative df/dx1 = 1.62982340133

3 3

ANN: Numexpr 1.4 released
by Francesc Alted Aug. 1, 2010

Aug. 1, 2010

Hi, After the Theano talk in last EuroSciPy I suddenly realized that it would not be too difficult to implement a multi-threaded version of Numexpr. Well, as usual I was terribly wrong and it took me a *long* week to do the job :-/ Anyway the thing is done now, so... enjoy! Note for PyTables users: Numexpr does not include changes in the API/ABI, so the upgrade from 1.3.1 is recommended. ======================== Announcing Numexpr 1.4 ======================== Numexpr is a fast numerical expression evaluator for NumPy. With it, expressions that operate on arrays (like "3*a+4*b") are accelerated and use less memory than doing the same calculation in Python. What's new ========== The main improvement in this version is the support for multi-threading in pure C. Threading in C provides the best performance in nowadays multi-core machines. In addition, this avoids the GIL that hampers performance in many Python apps. Just to wet your appetite, look into this page where the implementation is briefly described and where some benchmarks are shown: http://code.google.com/p/numexpr/wiki/MultiThreadVM In case you want to know more in detail what has changed in this version, see: http://code.google.com/p/numexpr/wiki/ReleaseNotes or have a look at RELEASE_NOTES.txt in the tarball. Where I can find Numexpr? ========================= The project is hosted at Google code in: http://code.google.com/p/numexpr/ And you can get the packages from PyPI as well: http://pypi.python.org/pypi Share your experience ===================== Let us know of any bugs, suggestions, gripes, kudos, etc. you may have. Enjoy! -- Francesc Alted

3 4

Mathematical language for reasoning about numpy like arrays?
by John Salvatier Aug. 1, 2010

Aug. 1, 2010

Is anyone aware of a good mathematical language to describe and reason about how numpy arrays work (broadcastable ndarrays)? I am particularly interested in reasoning about linear maps from one ndarray to another. Clearly ndarrays can be thought of as regular vectors and you can reason about them using linear algebra, but their ndimensionality and broadcasting also gives them more structure which gives them some interesting properties which I would like to reason about. John

1 0

Numpy 1.4.1 fails to build on (Debian) alpha and powepc
by Sandro Tosi Aug. 1, 2010

Aug. 1, 2010

Hello, I finally found the time to update numpy in Debian. But, there is a problem... As you probably know, we support several architectures and we need to have any package available on each of them. After the upload I noticed numpy has problem building on alpha [1] (ieee754.c compilation error, probably related to the conversion from .src) and powerpc [2] (long_double identification) [1] https://buildd.debian.org/fetch.cgi?pkg=python-numpy&arch=alpha&ver=1%3A1.4… [2] https://buildd.debian.org/fetch.cgi?pkg=python-numpy&arch=powerpc&ver=1%3A1… They are quite specific pieces of code, so I'm here asking your help in order to be able to build numpy on all Debian supported architectures and so be released in the upcoming stable release :) Of course, feel free to ask me any kind of support, also if you need to test/execute something on those archs, just tell me and I'll do. Regards, -- Sandro Tosi (aka morph, morpheus, matrixhasu) My website: http://matrixhasu.altervista.org/ Me at Debian: http://wiki.debian.org/SandroTosi

3 14

str == int puzzlement
by Matthew Brett Aug. 1, 2010

Aug. 1, 2010

Hi, Please forgive me if this is obvious, but this surprised me: In [15]: x = np.array(['a', 'b']) In [16]: x == 'a' # this was what I expected Out[16]: array([ True, False], dtype=bool) In [17]: x == 1 # this was strange to me Out[17]: False Is it easy to explain why this is? Thanks a lot, Matthew

5 9