[SciPy-user] large arrays ( > 2 gb) in numarray?
Francesc Altet
faltet at carabos.com
Thu Feb 10 06:04:53 EST 2005
Hi Lou,
If you just need to keep very large arrays and deal with them in
slices, you can use the EArray objects in PyTables:
from tables import *
from numarray import *
# creating the container
array_b = fileh.createEArray(root, 'array_b', Float64Atom(shape=(2,0,3)),
"Very large array",
Filters(complevel = 1)) # if you want compression
# adding chunks for the large array:
for i in range(100):
a = array(..., shape=(2,1000*1000*1000,3))
array_b.append(a)
# reading chunks of the very large array:
for i in range(100):
a = array_b[:, 1000*1000*1000*i:1000*1000*1000*(i+1), :]
# operations with chunk a after this...
You should be able to do this safely in PyTables as it supports 64-bit
addressing by default (although, to say the truth, I've never tried at
it, at least with EArrays).
This is probably not what you want, but just in case.
Cheers,
A Dimecres 09 Febrer 2005 06:53, Louis Wicker va escriure:
> Hello,
>
> I am new to the list. I have been building a python-fortran ensemble
> Kalman filter system for assimilating radar data into convective cloud
> models.
>
> I am running into a big problem that I thought I would ask about.
>
> I am designing this system for the SGI Altix machines, which is a
> Linux64 box with > 15 Gb of memory.
>
> I ran into the problem that Numeric or Numarray will not allow me to
> allocate an array larger than 2 gb. Is this a fundamental limitation
> of numarray/Numeric, or
> do I need to compile numarray/Numeric in some what to get it to use 64
> bit pointers. For example, here is the failure mode...
>
> >>> a = numarray.zeros((200,200,50,12,40), typecode='f')
>
> Traceback (most recent call last):
> File "<stdin>", line 1, in ?
> File "/usr/lib/python2.2/site-packages/numarray/numarraycore.py",
> line 1200, in zeros
> retarr = NumArray(shape=shape, type=type)
> libnumarray.error: NA_updateDataPtr: error getting read buffer data ptr
>
> any ideas? this is running python2.3 and numarray 1.1
>
> Thanks.
>
> Lou Wicker
> ------------------------------------------------------------------------
> ----
>
> | Dr. Louis J. Wicker
> | Research Scientist, National Severe Storms Lab
> | 1313 Halley Cir, Norman, OK 73069
> | E-mail: Louis.Wicker at nssl.noaa.gov
> | HTTP: www.nssl.noaa.gov/~lwicker
> | Phone: (405) 366-0416
> | Fax: (405) 366-0472
>
> ------------------------------------------------------------------------
> ----
>
> _______________________________________________
> SciPy-user mailing list
> SciPy-user at scipy.net
> http://www.scipy.net/mailman/listinfo/scipy-user
--
>qo< Francesc Altet http://www.carabos.com/
V V Cárabos Coop. V. Enjoy Data
""
More information about the SciPy-User
mailing list