A Divendres 09 Març 2007 18:56, Francesc Altet escrigué: > A Divendres 09 Març 2007 18:40, Sebastian Haase escrigué: > > Which dtypes are supported by numexpr ? > > Well, numexpr does support any dtype that is homogeneous, except 'uint64'. > This is because internally all the unsigned types are upcasted to the > smallest *signed* integer that can fit the info for it. As it happens > that 'uint64' doesn't have a standard signed type above that is able to > keep its info: this is why it is unsupported. > > Besides, there is the limitation that Win32 doesn't have such a 'uint64', > and computations in numpy or python are normally done by converting them to > python long integers (correct me if I'm wrong here), whose arithmetic is > very slow compared with the types supported by the compiler. So, the best > approach is to avoid 'uint64' types in general.
The info above is somewhat inexact. I was talking about the enhanced numexpr version included in PyTables 2.0 (see [1]). The original version of numexpr (see [2]) doesn't have support for int64 on 32-bit platforms and also neither does for strings. Sorry for the confusion. [1] http://www.pytables.org/trac/browser/trunk/tables/numexpr [2] http://projects.scipy.org/scipy/scipy/browser/trunk/Lib/sandbox/numexpr > > We are very interested in numexpr ! > > Where is the latest / most-up-to-date documentation ? > > At the moment, I think documentation is reduced at some docstrings. If you > want more, you will have to look at the sources. Ups, I spoke too fast. David Cooke kindly added this page: http://www.scipy.org/SciPyPackages/NumExpr Actually, numexpr doesn't need many more info that this as it is pretty straighforward to use already. Cheers, -- >0,0< Francesc Altet http://www.carabos.com/ V V Cárabos Coop. V. Enjoy Data "-" _______________________________________________ Numpy-discussion mailing list Numpy-discussion@scipy.org http://projects.scipy.org/mailman/listinfo/numpy-discussion
