[SciPy-user] Data types

Travis Oliphant oliphant.travis at ieee.org
Mon Mar 6 01:44:35 CST 2006


Tony Mannucci wrote:
> Dear Scipy community,
>
> I cannot find a clear statement as to allowable data types, and how 
> these are specified. I am trying to read in an array using 
> read_array. One of the arguments of the function is "atype", to 
> specify the dtype of the output array. I cannot figure out what 
> format the tuple atype is in.
>
>   
While SciPy builds for NumPy, it has not been "fully" adapted in that it 
still uses a lot of Numeric idioms for data-types. 

The read_array code is still using the old concept of "typecode" which 
is a one-character string (i.e. what dtype.char gives now).  

There may also still be bugs still in SciPy.  
> If Numpy is like Numeric, then, according to "Python In A Nutshell", 
> the array objects contain a single type. So, I cannot use several 
> values, I would specify only one value for atype. This seems to 
> differ from the scipy online documentation, suggesting atype could be 
> a tuple. I realize the functionality is evolving.
>   
In this case, the functionality was there but might have become broken 
in the transition to NumPy (or the result of an old bug).  You should be 
able to specify a tuple for data-type.  Most SciPy functions that take 
data-types need to be made aware of the new concept of data type in NumPy.
> Along these lines, the following works:
> data = ARRIN.read_array('filename',columns=(0,2),atype='d')
>
> and appears to give a double precision array (as evidenced from 
> data.dtype.char = 'd'). However, using a few other strings for atype 
> does not work, e.g. 'I', or 'L', etc. 
What do you get as the error?

> Finally, the following does not seem to work:
> data = ARRIN.read_array('filename',columns=(2))
>
> so reading a single column appears to cause a problem. The following DOES work:
> data = ARRIN.read_array('filename',columns=(0,2))
>
>   
It would be easiest if you opened a ticket and attached an example 
file.  Open tickets at

http://projects.scipy.org/scipy/scipy/timeline

Thanks,

-Travis




More information about the SciPy-user mailing list