Hi Jon Olav, A Saturday 13 June 2009 13:25:13 Jon Olav Vik escrigué: > If I understand the question correctly, this has confused me too. The short > answer may be: > a = tbl[:] > a = tbl[:].view(numpy.recarray) > or for single rows: > a = tbl[i] > a = tbl[i:i+1].view(numpy.recarray) > > The various options all have the same a.dtype and allow named field access > with a["label"]. However, "dotting" (a.label) only works with recarrays > (numpy.recarray aka numpy.core.records.recarray). The biggest gotcha for me > was that casting tbl[i] to recarray has no effect: You need tbl[i:i+1] to > get dotted access to fields of a single record. > > It is somewhat confusing that > type(tbl[i]) is numpy.void > type(tbl[i:i+1]) is numpy.ndarray > and that none of them are numpy.recarray.
This is intended and mimics the behaviour of NumPy: In [12]: type(r[0]) Out[12]: <type 'numpy.void'> In [13]: type(r[0:1]) Out[13]: <type 'numpy.ndarray'> > > It is also confusing that the term "record array" is used both for > numpy.recarray and a numpy.ndarray with a compound dtype (i.e. nonempty > a.dtype.names). > http://docs.scipy.org/doc/numpy/user/basics.rec.html Yeah, this confusion has been mainly due to the existence of two objects that are similar in NumPy, but with some differences. The first one, a ndarray with a compound type, is implemented at C level, and is generally faster to build and deal with. And the other, numpy.recarray, is a Python class that inherit from ndarray and adds more functionality to it, most specially the capability to access fields as members of the array, using ``recarr.x`` and ``recarr.y``. However, it is more expensive to build. Lately, it seems that the NumPy crew has stuck with the term ``structured arrays`` for naming the ndarray with a compound type, and ``record array`` (or ``recarray``, for short) to the Python ``recarray`` class. For what is worth, in general, PyTables always returns ``structured arrays`` as they are cheaper to build. For people preferring the functionality of ``record arrays`` instead, they can always build a view, as you are doing. Cheers, -- Francesc Alted ------------------------------------------------------------------------------ Crystal Reports - New Free Runtime and 30 Day Trial Check out the new simplified licensing option that enables unlimited royalty-free distribution of the report engine for externally facing server and web deployment. http://p.sf.net/sfu/businessobjects _______________________________________________ Pytables-users mailing list Pytables-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/pytables-users
