Setting "allow_pickle=True" by default

Question

I am encountering the error message

ValueError: Object arrays cannot be loaded when allow_pickle=False

while running a tool designed by somebody else (see DUDes). The solutions proposed in the related thread suggest setting allow_pickle=True directly in the np.load function. (My understanding is that the code in question was created while this used to be the default behavior of np.load, which has changes since then.) I would like to resolved this issue without tinkering with the original code. The obvious solution would be reverting to the earlier version of numpy. However, I am wondering whether, it is possible to simply change the default behavior for np.load in the (whatever) current version of numpy/python?

@juanpa.arrivillaga could you give more details? This sounds cryptic to me... — Roger Vadim, Jul 26 '22 at 09:32
@juanpa.arrivillaga but this should be done in the original code, which I do not want to modify? — Roger Vadim, Jul 26 '22 at 09:33
@juanpa.arrivillaga the code in question is launched from the command line. I do launch it from a python script though, via `sp.call` - is this what you mean? — Roger Vadim, Jul 26 '22 at 09:38
I do not know what `sp.call` is. It really isn't clear to me what you mean. — juanpa.arrivillaga, Jul 26 '22 at 09:40
@juanpa.arrivillaga ok, Suppose I simply run it from the command line. Could you provide instructions on how to monkey-patch the existing code? — Roger Vadim, Jul 26 '22 at 09:41
You can't really do much from the command line. It's going to require at least a little python code. — mousetail, Jul 26 '22 at 10:00
@mousetail it is okay tow rite some python code, but I need a more specific prescription of what this code is, where it is placed and when it is run. I have looked up a bit about [monkey-patching](https://stackoverflow.com/questions/5626193/what-is-monkey-patching), and in this context it seems to suggest recoding the np.load function. — Roger Vadim, Jul 26 '22 at 10:03

score 0 · Answer 1 · answered Jul 26 '22 at 10:08

The solution that has worked for me so far is downgrading numpy to a version with default setting allow_pickle=True:

conda install -c conda-forge python=3.7 numpy=1.16.2

Monkey-patching was suggested as a possible solution in the comments, but it is still not clear (to me) how to implement it here.

Setting "allow_pickle=True" by default

1 Answers1