From: Karol L. <kar...@kn...> - 2007-01-26 07:36:55
|
On Thursday 25 of January 2007 20:25, Adam Tenderholt wrote: > I briefly looked into the conversion from Numeric to numpy, and it > looks like there is a script that does all of the conversion for you > since there is an oldnumeric module of numpy. Granted, the script > probably isn't flawless. Also, it may be that we can detect whether > numpy is available and if so, import the numpy.oldnumeric interface > as Numeric (that's actually what the aforementioned script does); > otherwise, just import Numeric. Yes, there is a script, but I've read some people complaining about it (don't know if justified). > I do agree that we should wait until after the next release as this > release already had numerous changes. I don't know if numpy compiles > on Gentoo---since I bought my mac, I don't develop on Gentoo much > these days. I checked on a machine that runs an up-to-date Gentoo system, and numpy is installed system-wide there from portage, version 0.9.8 with python 2.4.3. Seems to work fine when played with, but running numpy.test() ends in a segmentation fault. That's probably caused by a problem with system libraries, becasue I also have a local install of numpy 1.0 with python 2.5 on that machine, and the tests finish fine in that case - the local installation uses my local ATLAS compilation. Here is a list of attributes cclib uses in Numeric: ['Numeric.matrixmultiply', 'Numeric.resize', 'Numeric.multiply', 'Numeric.divide', 'Numeric.array', 'Numeric.swapaxes', 'Numeric.innerproduct', 'Numeric.alltrue', 'Numeric.transpose', 'Numeric.arange', 'Numeric.add', 'Numeric.__version__', 'Numeric.reshape', 'Numeric.zeros'] As you can see, it isn't much, so there really shouldn't be much work in this regard. There's also the slicing and other references to arrays, but that hasn't really changed in numpy. What has changed in the above functions: 1. matrixmultiply is no more (use numpy.dot) 2. innerproduct is no more (I don't know what to use instead) Also, there is a new matrix object (matrix multiplication for a*b), but that's not too useful for cclib. Karol -- written by Karol Langner Fri Jan 26 07:51:39 CET 2007 |