Problem with get_preds on large test sets

I’m getting an OSError: [Errno 24] Too many open files when running get_preds on a large test set like so: learn.get_preds(ds_type=DatasetType.Test). I’m using this for collaborative filtering so the test set is a combination of all possible users and items (north of 5 million in some cases). The error occurs whenever I reach a 1000 batches (regardless of batch size).

I tried changing ulimit but this is not a reliable solution. What I’m trying now is to loop over the test set and get predictions for a batch at a time with pred_batch.

Is this a good way to solve this? Did anyone else encounter a similar problem and has a solution?


Any solutions to this problem? I am having the same problem when i try to open the full mnist dataset and append it to a list.For some reason when i try to append to the list it gives me this error.

Here is the code:

from import *
path = untar_data(URLs.MNIST)

def DatasetOpen(path,number):
numbers = []
for i in range(0,10):
temp = (path/‘training’/str(i)).ls().sorted()
numbers.append(([ for o in temp])