numpy array flags information is not saved when dumping to pickle

Question

I'm writing a class that has an attribute of numpy array type. Since I'd like it to be read-only, I set its WRITABLE flag to be false:

import numpy as np
class MyClass:
    def __init__(self):
        self.my_array = np.zeros(5)
        self.my_array.setflags(write=False)

After doing some other stuff, I dump MyClass into a pickle file:

pickle.dump(self, writer)

Later, I load it using x = pickle.load(reader), but then the WRITABLE flag is true. How can I make the pickle dump to preserve the numpy array WRITABLE flag?

Your constructor only runs when you create an instance of your object. How did you find out that it didn't set the flag? — Mazdak, Aug 11 '16 at 16:54
It's not surprising that the flags aren't preserved, considering that most of them represent properties that wouldn't be preserved by the pickling/unpickling process. For example, C_CONTIGUOUS and OWNDATA. — user2357112, Aug 11 '16 at 17:01

score 0 · Answer 1 · edited May 23 '17 at 12:00

For pickling arrays, numpy uses the np.save function. Details of this save are in np.lib.format file. This format saves a header and a byte representation of the data buffer. The content of header is a dictionary.

In [1212]: np.lib.format.header_data_from_array_1_0(x)
Out[1212]: {'descr': '<i4', 'fortran_order': False, 'shape': (2, 3)}
In [1213]: np.lib.format.header_data_from_array_1_0(u)
Out[1213]: {'descr': '<c16', 'fortran_order': False, 'shape': (4,)}

As you can see it does not save the whole FLAGS attribute, just info on the order.

Class pickling can be customized with __reduce__ and __setstate__ methods.

numpy array flags information is not saved when dumping to pickle

1 Answers1