Extending a series of nonuniform netcdf data in a numpy array

Question

I am new to python, apologies if this has been asked already.

Using python and numpy, I am trying to gather data across many netcdf files into a single array by iteratively calling append().

Naively, I am trying to do something like this:

from numpy import *
from pupynere import netcdf_file

x = array([])
y = [...some list of files...]

for file in y:
    ncfile = netcdf_file(file,'r')
    xFragment = ncfile.variables["varname"][:]
    ncfile.close()
    x = append(x, xFragment)

I know that under normal circumstances this is a bad idea, since it reallocates new memory on each append() call. But two things discourage preallocation of x:

1) The files are not necessarily the same size along axis 0 (but should be the same size along subsequent axes), so I would need to read the array sizes from each file beforehand to precalculate the final size of x.

However...

2) From what I can tell, pupynere (and other netcdf modules) load the entire file into memory upon opening the file, rather than just a reference (such as many netcdf modules in other enviroments). So to preallocate, I'd have to open the files twice.

There are many (>100) large (>1GB) files, so overallocating and reshaping is not practical, from what I can tell.

My first question is whether I am missing some intelligent way to preallocate.

My second question is more serious. The above snippet works for a single-dimension array. But if I try to load in a matrix, then initialisation becomes a problem. I can append a one-dimensional array to an empty array:

append( array([]), array([1, 2, 3]) )

but I cannot append an empty array to a matrix:

append( array([]), array([ [1, 2], [3, 4] ]), axis=0)

Something like x.extend(xFragment) would work, I believe, but I don't think numpy arrays have this functionality. I could also avoid the initialisation problem by treating the first file as a special case, but I'd prefer to avoid that if there's a better way to do it.

If anyone can offer help or a suggestion, or can identify a problem with my approach, then I'd be grateful. Thanks

Ofri Raviv · Accepted Answer · 2010-04-15 06:35:11Z

1

You can solve the two problems by first loading the arrays from the files files into a list of arrays, and then using concatenate to join all the arrays. Something like this:

x = [] # a normal python list, not np.array
y = [...some list of files...]

for file in y:
    ncfile = netcdf_file(file,'r')
    xFragment = ncfile.variables["varname"][:]
    ncfile.close()
    x.append(xFragment)

combined_array = concatenate(x, axis=0)

answered Apr 15, 2010 at 6:35

Ofri Raviv

24.9k3 gold badges59 silver badges55 bronze badges

Sign up to request clarification or add additional context in comments.

4 Comments

marshall.ward Over a year ago

The netcdf data comes in as numpy arrays, so this will create lists of arrays like this: [array(...), array(...), array(...) ... ] for each append statement. Maybe if I convert the numpy arrays to regular lists first, then back to a numpy array at the end, this will work?

marshall.ward Over a year ago

After trying this idea, I decided that it's probably not what I'm looking for. I'd have to call xFragment.tolist() each iteration, which doesn't seem to be an improvement over setting x to xFragment on the first iteration and using x.append(xFragment) on subsequent iterations.

Ofri Raviv Over a year ago

x would be a list of NumPy arrays. that's exactly what concatenate expects to get.

marshall.ward Over a year ago

You're right, I must have not been using concatenate properly when I first tried your suggestion. It seems to work for some smaller examples. Unfortunately it seems that concatenate is too slow, or possibly not working at all, on these larger data sets (Maybe it uses a lot of memory?). I may be doing something wrong, but your suggestion appears to be working in principle. Thanks for your help.

Collectives™ on Stack Overflow

Extending a series of nonuniform netcdf data in a numpy array

1 Answer 1

4 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

4 Comments

Your Answer

Sign up or log in

Post as a guest

Related