Thanks a lot for your answer Francesc, Cyril.
Francesc Alted a écrit : > Cyril, > > A Tuesday 28 April 2009, cyril giraudon escrigué: > >> Hi, >> >> I use HDF5 to store complex numbers and I see two ways of defining a >> complex number (since I believe there is no official manner): >> > > Yes, as far as I know there is not an blessed way to declare complex > types in HDF5. In fact, PyTables chose the compound approach back in > 2004 (in fact, contributed by Tom Hedley) and frankly, as it seems to > work quite well, we have never looked back (in fact, Octave follows the > same strategy when writes to HDF5 files, except that the compound > fields are called 'real' and 'imag' instead of the PyTables 'r' and 'i' > convention, but this is supported by PyTables for reading too). > > In addition, lately the h5py project seems to have chosen the same > approach than PyTables to represent complex numbers. So, even though > it is true that there is not an 'official' way to specify complex > numbers, there is an a certain tradition in doing it the PyTables' way > (let's call it this way). > > >> 1. A compound datatype : one real for the real part and one real for >> the imaginery part >> 2. A 2 element array datatype : two reals in an array A, A(0) is the >> real part and A(1) is the imaginary part. >> >> Pytables read directly the compound datatype as a complex number but >> read "logically" the array datatype as an array. >> Is it possible pytables read directly an array datatype as a complex >> number without any conversion ? >> >> Why this question ? It is very simpler to manipulate an array >> datatype from C, fortran ... than a compound datatype and the new >> lite API allows easily the creation of such a structure. >> > > Yes, this should be possible, but provided that it already exists a way, > I think there is not much point in changing it. Perhaps you may want > to create your complex arrays out of your HDF5 data by operating with > NumPy, like for example: > > In [66]: r = np.array([1,2,3], dtype='f8') > > In [67]: i = np.array([4,5,6], dtype='f8') > > In [68]: cplx = r+i*1.j > > In [69]: cplx > Out[69]: array([ 1.+4.j, 2.+5.j, 3.+6.j]) > > which should be fast enough for most of applications. > > HTH, > > ------------------------------------------------------------------------------ Register Now & Save for Velocity, the Web Performance & Operations Conference from O'Reilly Media. Velocity features a full day of expert-led, hands-on workshops and two days of sessions from industry leaders in dedicated Performance & Operations tracks. Use code vel09scf and Save an extra 15% before 5/3. http://p.sf.net/sfu/velocityconf _______________________________________________ Pytables-users mailing list Pytables-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/pytables-users
