I wonder, how to save and load numpy.array data properly. Currently I’m using the numpy.savetxt() method. For example, if I got an array markers, which looks like this:

I try to save it by the use of:

numpy.savetxt('markers.txt', markers)

In other script I try to open previously saved file:

markers = np.fromfile("markers.txt")

And that’s what I get…

Saved data first looks like this:

0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00
0.000000000000000000e+00

But when I save just loaded data by the use of the same method, ie. numpy.savetxt() it looks like this:

1.398043286095131769e-76
1.398043286095288860e-76
1.396426376485745879e-76
1.398043286055061908e-76
1.398043286095288860e-76
1.182950697433698368e-76
1.398043275797188953e-76
1.398043286095288860e-76
1.210894289234927752e-99
1.398040649781712473e-76

What am I doing wrong? PS there are no other “backstage” operation which I perform. Just saving and loading, and that’s what I get. Thank you in advance.

The most reliable way I have found to do this is to use np.savetxt with np.loadtxt and not np.fromfile which is better suited to binary files written with tofile. The np.fromfile and np.tofile methods write and read binary files whereas np.savetxt writes a text file. So, for example:

In [1]: a = np.array([1, 2, 3, 4])
In [2]: np.savetxt('test1.txt', a, fmt='%d')
In [3]: b = np.loadtxt('test1.txt', dtype=int)
In [4]: a == b
Out[4]: array([ True,  True,  True,  True], dtype=bool)

Or:

In [5]: a.tofile('test2.dat')
In [6]: c = np.fromfile('test2.dat', dtype=int)
In [7]: c == a
Out[7]: array([ True,  True,  True,  True], dtype=bool)

I use the former method even if it is slower and creates bigger files (sometimes): the binary format can be platform dependent (for example, the file format depends on the endianness of your system).

There is a platform independent format for NumPy arrays, which can be saved and read with np.save and np.load:

In  [8]: np.save('test3.npy', a)    # .npy extension is added if not given
In  [9]: d = np.load('test3.npy')
In [10]: a == d
Out[10]: array([ True,  True,  True,  True], dtype=bool)

np.save('data.npy', num_arr) # save
new_num_arr = np.load('data.npy') # load

np.fromfile() has a sep= keyword argument:

Separator between items if file is a text file. Empty (“”) separator means the file should be treated as binary. Spaces (” ”) in the separator match zero or more whitespace characters. A separator consisting only of spaces must match at least one whitespace.

The default value of sep="" means that np.fromfile() tries to read it as a binary file rather than a space-separated text file, so you get nonsense values back. If you use np.fromfile('markers.txt', sep=" ") you will get the result you are looking for.

However, as others have pointed out, np.loadtxt() is the preferred way to convert text files to numpy arrays, and unless the file needs to be human-readable it is usually better to use binary formats instead (e.g. np.load()/np.save()).

For a short answer you should use np.save and np.load. The advantages of these is that they are made by developers of the numpy library and they already work (plus are likely already optimized nicely) e.g.

import numpy as np
from pathlib import Path

path = Path('~/data/tmp/').expanduser()
path.mkdir(parents=True, exist_ok=True)

lb,ub = -1,1
num_samples = 5
x = np.random.uniform(low=lb,high=ub,size=(1,num_samples))
y = x**2 + x + 2

np.save(path/'x', x)
np.save(path/'y', y)

x_loaded = np.load(path/'x.npy')
y_load = np.load(path/'y.npy')

print(x is x_loaded) # False
print(x == x_loaded) # [[ True  True  True  True  True]]

Expanded answer:

In the end it really depends in your needs because you can also save it human readable format (see this Dump a NumPy array into a csv file) or even with other libraries if your files are extremely large (see this best way to preserve numpy arrays on disk for an expanded discussion).

However, (making an expansion since you use the word “properly” in your question) I still think using the numpy function out of the box (and most code!) most likely satisfy most user needs. The most important reason is that it already works. Trying to use something else for any other reason might take you on an unexpectedly LONG rabbit hole to figure out why it doesn’t work and force it work.

Take for example trying to save it with pickle. I tried that just for fun and it took me at least 30 minutes to realize that pickle wouldn’t save my stuff unless I opened & read the file in bytes mode with wb. Took time to google, try thing, understand the error message etc… Small detail but the fact that it already required me to open a file complicated things in unexpected ways. To add that it required me to re-read this (which btw is sort of confusing) Difference between modes a, a+, w, w+, and r+ in built-in open function?.

So if there is an interface that meets your needs use it unless you have a (very) good reason (e.g. compatibility with matlab or for some reason your really want to read the file and printing in python really doesn’t meet your needs, which might be questionable). Furthermore, most likely if you need to optimize it you’ll find out later down the line (rather than spend ages debugging useless stuff like opening a simple numpy file).

So use the interface/numpy provide. It might not be perfect it’s most likely fine, especially for a library that’s been around as long as numpy.

I already spent the saving and loading data with numpy in a bunch of way so have fun with it, hope it helps!

import numpy as np
import pickle
from pathlib import Path

path = Path('~/data/tmp/').expanduser()
path.mkdir(parents=True, exist_ok=True)

lb,ub = -1,1
num_samples = 5
x = np.random.uniform(low=lb,high=ub,size=(1,num_samples))
y = x**2 + x + 2

# using save (to npy), savez (to npz)
np.save(path/'x', x)
np.save(path/'y', y)
np.savez(path/'db', x=x, y=y)
with open(path/'db.pkl', 'wb') as db_file:
    pickle.dump(obj={'x':x, 'y':y}, file=db_file)

## using loading npy, npz files
x_loaded = np.load(path/'x.npy')
y_load = np.load(path/'y.npy')
db = np.load(path/'db.npz')
with open(path/'db.pkl', 'rb') as db_file:
    db_pkl = pickle.load(db_file)

print(x is x_loaded)
print(x == x_loaded)
print(x == db['x'])
print(x == db_pkl['x'])
print('done')

Some comments on what I learned:

• np.save as expected, this already compresses it well (see https://stackoverflow.com/a/55750128/1601580), works out of the box without any file opening. Clean. Easy. Efficient. Use it.
• np.savez uses a uncompressed format (see docs) Save several arrays into a single file in uncompressed .npz format. If you decide to use this (you were warned to go away from the standard solution so expect bugs!) you might discover that you need to use argument names to save it, unless you want to use the default names. So don’t use this if the first already works (or any works use that!)
• Pickle also allows for arbitrary code execution. Some people might not want to use this for security reasons.
• human readable files are expensive to make etc. Probably not worth it.
• there is something called hdf5 for large files. Cool! https://stackoverflow.com/a/9619713/1601580

Note this is not an exhaustive answer. But for other resources check this:

• For pickle (guess the top answer is don’t use pickle us np.save): Save Numpy Array using Pickle
• For large files (great answer! compares storage size, loading save and more!): https://stackoverflow.com/a/41425878/1601580
• For matlab (we have to accept matlab has some freakin’ nice plots!): “Converting” Numpy arrays to Matlab and vice versa
• For saving in human readable format: Dump a NumPy array into a csv file