On Nov 28, 2008, at 4:31 PM, Ronan Paixão wrote: > Em Ter, 2008-11-25 às 08:07 -0800, Mike Hansen escreveu: >> Hi, >> >> On Tue, Nov 25, 2008 at 6:22 AM, Stan Schymanski >> <[EMAIL PROTECTED]> wrote: >>> >>> Hi Mike, >>> >>> This is pretty cool, thanks! Is there something equivalent for >>> passing >>> a function f to python or numpy? >> >> I'm not exactly sure what you mean by this so I'll take a guess. >> Given, your f=a*x^2+b, do you want to be able to get an object >> that is >> / acts like the Python >> >> def f(a,b,x): >> return a*x**2 + b >> >> ? >> >> For numpy, if you had an expression like f = sin(x) + 2, you'd want >> something like, >> >> def f(x): >> return numpy.sin(x)+2 >> >> so that it'd work well with numpy arrays? >> >> If so, then none of this is currently possible :-) But, it's >> primarily not possible since no one has seriously thought about doing >> this before. I think it would definitely make Sage's symbolic stuff >> much more useful to a wider range of people. It's also probably not >> too far off with the Pynac stuff. >> >> I'm sending this to sage-devel to get comments / feedback from >> people there. >> >> --Mike > > Hi, > Actually, I have been thinking about it and discussing that in irc for > some time now, since I currently do my projects using pure numpy > (without sage). > > Also, I have been trying to do some stuff in sage in a way that I > wanted > exactly that to work, since (I presume) doing all the calculations > of a > function in sage, then simplifying, and only then applying to a big > numerical array "should" be more efficient than just doing all ops > directly to the array. > > So far, the results I have are that most ways to do that have a lot of > type conversions, because applying a sage function to a each > element of > the array will return an array of sage objects. I'm doing > sage: array(map(f, my_data)) > Which will return an array of sage objects. To get it back to numpy > types I use the astype() function, but I think is a terrible waste on > performance and memory. OBS: this also only works for 1-D arrays, > so one > also has to do some reshaping magic to work on n-dimension arrays. > > Also, don't forget that for doing operations on arrays you sometimes > have to coerce numbers to raw ones, otherwise they will pass sage > types > to numpy and generate arrays of sage objects. > > So far, I'm doing everything in pure python + numpy, for performance > reasons, though it would be nice to have some sage powers in the > future. > > Well, that's my 2 cents.
Note that if f is a symbolic function, calling it will have a *huge* amount of overhead. Hopefully with Pynac going in, this will change. But with numpy the loops happen in the C level, rather than the Python level, so would be much faster. Using NumPy will (nearly) always be faster than trying to do things one at a time with a Python loop. - Robert --~--~---------~--~----~------------~-------~--~----~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~----------~----~----~----~------~----~------~--~---
