Mailman 3 June 2005 - SciPy-User

Bug(?) in optimize.fmin_powell
by Nils Wagner 29 Jun '05

29 Jun '05

Hi all, I have used several optimization procedures to minimize the Rayleigh quotient. Here are the results of rayleigh.py (see attachment): Optimization terminated successfully. Current function value: 0.022338 Iterations: 648 Function evaluations: 924 fmin 924 Approximated smallest eigenvalue 0.0223383478603 Optimization terminated successfully. Current function value: 0.022338 Iterations: 7 Function evaluations: 569 Powell 790 Approximated smallest eigenvalue 0.0223384381901 Optimization terminated successfully. Current function value: 0.022338 Iterations: 17 Function evaluations: 420 Gradient evaluations: 0 fmin_cg 420 Approximated smallest eigenvalue 0.0223383476402 Optimization terminated successfully. Current function value: 0.022338 Iterations: 22 Function evaluations: 276 Gradient evaluations: 0 fmin_bfgs 276 Approximated smallest eigenvalue 0.0223383475942 smallest eigenvalue by linalg.eigvals 0.0223383475497 The number of function evaluations differs in case of fmin_powell, that is 790 is given by len(data) and 569 is given in the output of fmin_powell. Is that a bug in fmin_powell ? Moreover, the evolution of the Rayleigh quotient is quite different in case of fmin_powell. For what reason ? Nils

1 0

Re: SciPy-user Digest, Vol 22, Issue 3
by aurelien.gourrier＠free.fr 29 Jun '05

29 Jun '05

Hi all, Can anyone tell me if there is a possibility to use constraints on the leastsq method ? I need to do some bounded minimization of gaussians or lorentzians on 1-D data and couldn't find any way out so far... Aure

5 6

Getting started with optimize.fmin_l_bfgs_b
by Nils Wagner 29 Jun '05

29 Jun '05

Hi all, How can I solve the following LP problem with scipy ? Minimize f(x_1,x_2) = x_1-x_2 subject to -2 \le x_1 \le 2 -2 \le x_2 \le 2 The solution is x_1 = -2 , x_2 = 2 My implementation fails from scipy import * def func(x): return x[0]-x[1] guess = 1.2, 1.0 bounds = [(-2.0,2.0), (-2.0,2.0) ] best, val, d = optimize.fmin_l_bfgs_b(func, guess, bounds=bounds) Traceback (most recent call last): File "bfgs_b.py", line 8, in ? best, val, d = optimize.fmin_l_bfgs_b(func, guess, bounds=bounds) File "/usr/local/lib/python2.4/site-packages/scipy/optimize/lbfgsb.py", line 183, in fmin_l_bfgs_b f[0], g = func_and_grad(x) File "/usr/local/lib/python2.4/site-packages/scipy/optimize/lbfgsb.py", line 135, in func_and_grad f, g = func(x, *args) TypeError: unpack non-sequence How can I resolve the problem ? Nils

2 5

scipy tutorials
by Matt Fishburn 27 Jun '05

27 Jun '05

Hello all, I've written some basic scipy tutorials, located at: http://scipy.mit.edu/ The tutorials are meant to be a resource to people starting to work with scipy/matplotlib/ipython. A few more tutorials should be coming over the next month or so. Please let me have any feedback - if you find the tutorials useful, if you thought something wasn't explained well, etc. -Matt Fishburn

4 4

LiveDocs?
by Stephen Walton 24 Jun '05

24 Jun '05

I thought this was asked before but perhaps it was my imagination: when will the Scipy LiveDocs be back up?

2 1

Scalapack interface?
by rajb 24 Jun '05

24 Jun '05

Hi all I keep seeing references in the docs regarding a Python interface to Scalapack, but it does not seem to actually exist. Can anyone please tell me if there is one around or in development stage? I need such an interface for my research, and I was thinking of writing one myself, but it would be great if someone is already working on it. Please reply asap. Thanks a lot. Raj

2 1

ANN: PyUniversalLibrary version 20050623
by Andrew Straw 23 Jun '05

23 Jun '05

I would like to announce PyUniversalLibrary version 20050623. PyUniversalLibrary is a Python wrapper for Measurement Computing's Universal Library for data acquisition on Microsoft Windows operating systems. http://www.its.caltech.edu/~astraw/pyul.html Work in Progress ---------------- Currently PyUniversalLibrary is incomplete. I have only wrapped the functions which I have personally needed, primarily buffered analog input. If you need more functionality, you will hopefully find that the start I have made in wrapping this library useful. I welcome your changes and additions. I will include them with the next release with appropriate credit.

1 0

Bug in optimize.leastsq when Jacobian matrix is provided
by Abhiman Chatra 23 Jun '05

23 Jun '05

Hi, I am new to Python, SciPy and optimization. So excuse me if I have make mistakes. My Problem : ------------------- I want to use optimize.leastsq function of scipy to find a least square fit. But my function in the parametric form is quite complicated. Hence I want to provide the Jacobian matrix to the leastsquare optimizer to improve its performance. But when I provide the function for calculating the Jacobian matrix, I get a windows error saying "python.exe has generated errors and will be closed by Windows. You will need to restart the program. An error log is being created <OK>" Since this error cannot be caught I cannot trace the cause of this error. But my feeling is that this is some memory error. This error is not reproducible if I do not provide the function for calculating the Jacobian matrix. My Configutration : -------------------------- Machine : P4 OS : Windows 2000 Professional Python : Python 2.3.3 Numeric : Numeric-23.5.win32-py2.3 Scipy : SciPy_complete-0.3.2.win32-py2.3-num23.5 ( I have also tryed with the following configuration : Python : Python 2.2.3 Numeric : Numeric-23.0.win32-py2.2 SciPy : SciPy_complete-0.3.2.win32-py2.2-num22.0 -- Abhiman Chatra B, Technical Analyst, Trilogy India, Mobile # : 9845358165 ---------------------------------------------------- "If only your programs would be correct if you simply typed them three times"

2 1

Bug(?) in optimize.leastsq when Jacobian matrix is provided (Complete) <Ignore my previous mail>
by Abhiman Chatra 23 Jun '05

23 Jun '05

(Sorry my previous mail was incomplete) Hi, I am new to Python, SciPy and optimization. So excuse me if I have make mistakes. My Problem : ------------------- I want to use optimize.leastsq function of scipy to find a least square fit. But my function in the parametric form is quite complicated. Hence I want to provide the Jacobian matrix to the leastsquare optimizer to improve its performance. But when I provide the function for calculating the Jacobian matrix, I get a windows error saying "python.exe has generated errors and will be closed by Windows. You will need to restart the program. An error log is being created <OK>" Since this error cannot be caught I cannot trace the cause of this error. But my feeling is that this is some memory error. This error is not reproducible if I do not provide the function for calculating the Jacobian matrix. My Configutration : -------------------------- Machine : P4 OS : Windows 2000 Professional Python : Python 2.3.3 Numeric : Numeric-23.5.win32-py2.3 Scipy : SciPy_complete-0.3.2.win32-py2.3-num23.5 ( I have also tryed with the following configuration : Python : Python 2.2.3 Numeric : Numeric-23.0.win32-py2.2 SciPy : SciPy_complete-0.3.2.win32-py2.2-num22.0 - Again my code fails) Code : --------------------------------------------------------------------------------------------------------------------------------- #This is not my actual code but a simplified version for simple linear regression fit from scipy import * import random import pprint inputs = [ [ 1.0 , 2.0 ], [ 2.0 , 2.0 ], [ 3.0 , 1.0 ], [ 2.0 , 4.0 ]] answers = [ 4.05 , 6.1 , 6.9 , 8.0 ] def residual_values(betas, inp , ans) : ret = [0.0] * len(inp) for i in range(len(inp)) : ret[i] = 1.0 * betas[len(inp[i])] - ans[i] for j in range(len(inp[0])) : ret[i] += betas[j] * inp[i][j] print "In residual_values" #print "Betas : " #pprint.pprint(betas) #print "return value : " #pprint.pprint(ret) return ret def jacobian(betas, inp , ans) : print "In jacobian" J = [[0.0] * len(betas)] * len(inp) for i in range(len(inp)) : J[i][0] = inp[i][0] J[i][1] = inp[i][1] J[i][2] = 1.0 return J try: for i in range(100): print("---------------- %d -----------------" % (i)) initial = [0.0] * 3 #random.shuffle(initial) for i in range(3) : initial[i] = 1.0 opt = optimize.leastsq(residual_values, initial, args = (inputs, answers), Dfun=jacobian, maxfev=10000) #opt = optimize.leastsq(residual_values, initial, args = (inputs, answers), maxfev=10000) pprint.pprint(opt) except: print "Error regression fitting model..." pprint.pprint(locals()) raise --------------------------------------------------------------------------------------------------------------------------------- Can someone please let me know what is causing the crash? My feeling is that this is a bug in SciPy. Any help is greatly appreciated. Thanks, Abhiman -- Abhiman Chatra B, Technical Analyst, Trilogy India, Mobile # : 9845358165 ---------------------------------------------------- "If only your programs would be correct if you simply typed them three times"

3 4

Tiger and scipy
by Brian Granger 22 Jun '05

22 Jun '05

Hello all, I wanted to see what the status of scipy on Mac OS X, 10.4 was. I just installed the CVS version of scipy on Tiger, but I had to use gcc 3.3 (sudo gcc_select 3.3). I got the same error that others have seen when I used gcc4.0. Any word on how to resolve this problem with gcc4.0? Also, I have been playing with the bdist_mpkg command that comes with py2app (from the PyObjC project). This makes it nearly trivial to build a double clickable Mac installer for scipy. Has anyone else done this yet? Also, has anyone made installer packages for the other dependencies (g77 and fftw) yet? With a little work, it should be possible to have a single metapackage for scipy that make installation on a Mac trivial. Anyone else interested in helping with this? Cheers Brian

3 3