Python : save dictionaries through numpy.save

Question

I have a large data set (millions of rows) in memory, in the form of numpy arrays and dictionaries.

Once this data is constructed I want to store them into files;
so, later I can load these files into memory quickly, without reconstructing this data from the scratch once again.

np.save and np.load functions does the job smoothly for numpy arrays.
But I am facing problems with dict objects.

See below sample. d2 is the dictionary which was loaded from the file. See #out[28] it has been loaded into d2 as a numpy array, not as a dict. So further dict operations such as get are not working.

Is there a way to load the data from the file as dict (instead of numpy array) ?

In [25]: d1={'key1':[5,10], 'key2':[50,100]}

In [26]: np.save("d1.npy", d1)

In [27]: d2=np.load("d1.npy")

In [28]: d2
Out[28]: array({'key2': [50, 100], 'key1': [5, 10]}, dtype=object)

In [30]: d1.get('key1')  #original dict before saving into file
Out[30]: [5, 10]

In [31]: d2.get('key2')  #dictionary loaded from the file
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-31-23e02e45bf22> in <module>()
----> 1 d2.get('key2')

AttributeError: 'numpy.ndarray' object has no attribute 'get'

Asked By: ramu

||

Source

Answer 1

It’s a structured array. Use d2.item() to retrieve the actual dict object first:

import numpy as np

d1={'key1':[5,10], 'key2':[50,100]}
np.save("d1.npy", d1)
d2=np.load("d1.npy")
print d1.get('key1')
print d2.item().get('key2')

result:

[5, 10]
[50, 100]

Answered By: Kennet Celeste

Answer 2

pickle module can be used. Example code:

from six.moves import cPickle as pickle #for performance
from __future__ import print_function
import numpy as np

def save_dict(di_, filename_):
    with open(filename_, 'wb') as f:
        pickle.dump(di_, f)

def load_dict(filename_):
    with open(filename_, 'rb') as f:
        ret_di = pickle.load(f)
    return ret_di

if __name__ == '__main__':
    g_data = {
        'm':np.random.rand(4,4),
        'n':np.random.rand(2,2,2)
    }
    save_dict(g_data, './data.pkl')
    g_data2 = load_dict('./data.pkl')
    print(g_data['m'] == g_data2['m'])
    print(g_data['n'] == g_data2['n'])

You may also save multiple python objects in a single pickled file. Each pickle.load call will load a single object in that case.

Answered By: Kh40tiK

Python : save dictionaries through numpy.save

Question:

Answers: