Object arrays cannot be loaded when allow_pickle=False

Borislav Hadzhiev

Last updated: Apr 11, 2024

Reading time·4 min

**The Python "ValueError: Object arrays cannot be loaded when allow_pickle=False
" occurs when you try to load an object array with the allow_pickle argument
set to False.**

**To solve the error, set the allow_pickle argument to True or change the
type of the array.**

Here is an example of how the error occurs.

main.py

`import numpy as np np.save('data.npy', np.array([[1, 2, 3], [4, 5, 6]], dtype=object)) arr = np.load('data.npy', allow_pickle=False) # ⛔️ ValueError: Object arrays cannot be loaded when allow_pickle=False print(arr)`

The code for this article is available on GitHub

We used the
numpy.save()
method to save an array to a binary file in NumPy (`.npy`

) format.

The first argument the method takes is the file path.

The second argument is the array that should be saved and the third is the data type.

Notice that we set the
dtype to
`object`

.

The
numpy.load()
method loads arrays or pickled objects from `.npy`

, `.npz`

or pickled files.

The first argument the method takes is the file path.

The

`allow_pickle`

keyword argument determines whether it is allowed to load pickled object arrays that are stored in `.npy`

files.**If the allow_pickle argument is set to False, loading object arrays causes
the error.**

By default, the `allow_pickle`

argument is set to `False`

.

`allow_pickle`

argument to `True`

to solve the errorOne way to solve the error is to set the `allow_pickle`

argument to `True`

.

main.py

`import numpy as np np.save('data.npy', np.array([[1, 2, 3], [4, 5, 6]], dtype=object)) arr = np.load('data.npy', allow_pickle=True) # [[1 2 3] # [4 5 6]] print(arr)`

The code for this article is available on GitHub

When the `allow_pickle`

keyword argument is set to `True`

, loading object arrays
is allowed.

Note that the default value of the `allow_pickle`

argument is `False`

, so it has
to be explicitly specified.

The `allow_pickle`

argument can also be passed to the `numpy.save()`

method.

main.py

`import numpy as np np.save( 'data.npy', np.array([[1, 2, 3], [4, 5, 6]], dtype=object), allow_pickle=True ) arr = np.load('data.npy', allow_pickle=True) # [[1 2 3] # [4 5 6]] print(arr)`

However, passing the `allow_pickle`

argument to `numpy.save()`

is not necessary
as it defaults to `True`

.

You can also set the `allow_pickle`

argument to `True`

when using the `with`

statement.

Replace this:

main.py

`# ⛔️ Incorrect import numpy as np with np.load('data.npy') as arr: print(arr)`

With this:

main.py

`# ✅ Correct import numpy as np with np.load('data.npy', allow_pickle=True) as arr: print(arr)`

You can also set a different data type for the NumPy array to solve the error.

For the error to occur, 2 conditions have to be met:

- The
`allow_pickle`

argument has to be set to`False`

. - The data type of the array has to be set to
`object`

.

You can set the data type of the array to `int`

or `float`

to solve the error.

main.py

`import numpy as np np.save( 'data.npy', np.array([[1, 2, 3], [4, 5, 6]], dtype=int), ) arr = np.load('data.npy') # [[1 2 3] # [4 5 6]] print(arr)`

The code for this article is available on GitHub

We set the `dtype`

(data type) of the array to `int`

instead of `object`

.

Notice that we didn't have to explicitly set the `allow_pickle`

keyword
argument.

`numpy`

to version `1.16.1`

If you use `keras`

, you can also try to downgrade your version of the `numpy`

module to `1.16.1`

as the error occurs in `numpy`

version `1.16.4`

and more
recent.

Open your terminal in your project's root directory and run the following command.

shell

`pip install numpy==1.16.1 # or with pip3 pip3 install numpy==1.16.1 # or in Jupyter notebook !pip install numpy==1.16.1 # or with Anaconda conda install numpy==1.16.1`

If you have a `requirements.txt`

file, you can add the following line.

requirements.txt

`numpy==1.16.1`

`keras`

If you got the error when using the `load_data()`

method in `keras`

, try to set
the default value of the `allow_pickle`

keyword argument to `True`

.

main.py

`import numpy as np from tensorflow import keras np.load.__defaults__ = (None, True, True, 'ASCII') np.save('data.npy', np.array([[1, 2, 3], [4, 5, 6]], dtype=object)) (x_train, y_train), (x_test, y_test) = keras.datasets.mnist.load_data() np.load.__defaults__ = (None, False, True, 'ASCII')`

The code for this article is available on GitHub

We updated the `__defaults__`

tuple of the `numpy.load()`

method, setting
`allow_pickle`

to `True`

.

`allow_pickle`

is the second argument in the tuple.

Once we're done, we set the default value of the argument back to `False`

.

You can learn more about the related topics by checking out the following tutorials:

- Cannot import name 'pad_sequences' from 'keras.preprocessing.sequence'
- ModuleNotFoundError: No module named 'tensorflow' in Python
- ValueError: cannot reshape array of size X into shape Y
- lbfgs failed to converge (status=1): STOP: TOTAL NO. of ITERATIONS REACHED LIMIT
- How to draw empty circles on a Scatter Plot in Matplotlib
- TypeError: Image data cannot be converted to float [Solved]
- OverflowError: Python int too large to convert to C long
- How to use numpy.argsort in Descending order in Python
- TypeError: cannot pickle '_thread.lock' object [Solved]
- ValueError: Columns must be same length as key [Solved]
- ufunc 'add' did not contain loop with signature matching types
- ValueError: Found array with dim 3. Estimator expected 2
- Input contains infinity or value too large for dtype(float64)
- ValueError: unsupported pickle protocol: 5 [Solved]
- ValueError: columns overlap but no suffix specified [Solved]
- Get the first Row of each Group in a Pandas DataFrame
- How to Multiply two or more Columns in Pandas
- Add columns of a different Length to a DataFrame in Pandas