Numpy array is not JSON serializable

python

numpy

json-serialization

data-serialization

byAnton Shumikhin·Oct 12, 2024

Easily convert a NumPy array to JSON by transforming it into a list with tolist() and serializing with json.dumps():

import json
import numpy as np

np_array = np.array([1, 2, 3])

# Convert and serialize like a charm!
json_data = json.dumps(np_array.tolist())

Custom encoding for complex NumPy types

Some NumPy types don't play nice with JSON. Fear not! Craft a custom encoder to handle these troublesome types.

import json
import numpy as np

# Custom encoder at your service
class NumpyEncoder(json.JSONEncoder):
    def default(self, obj):
        if isinstance(obj, np.ndarray):
            return obj.tolist()
        return json.JSONEncoder.default(self, obj)

np_array = np.array([1, 2, 3])

# Let the custom encoder do its magic
json_data = json.dumps(np_array, cls=NumpyEncoder)

The custom encoder ensures your JSON stays friendly and consistent.

How to get your data back

Rescue your data from JSON format back to a NumPy array using good old np.array():

import json
import numpy as np

json_data = '[1, 2, 3]'
list_data = json.loads(json_data)

# Welcome back to the cool kids club, data!
np_array = np.array(list_data)

Your data is now back as a high-performance NumPy array. High fives all round!

Storing data like a boss

For big data, file storage methods are key. Use codecs.open() for the perfect save:

import codecs
import json
import numpy as np

np_array = np.array([1, 2, 3])

# codecs - because saving should be sassy!
with codecs.open('datafile.json', 'w', encoding='utf-8') as f:
    json.dump(np_array.tolist(), f)

Your data is now safe and sound. Cheers to neuropreservation!

Manage multi-dimensional data elegantly

For those brain-melting multi-dimensional arrays, use Pandas and make your life simple:

import pandas as pd
import numpy as np

df = pd.DataFrame({'col1': [1, 2], 'col2': [3, 4]})
json_data = df.to_json(orient='values')

# Magic spell to bring our data back as a NumPy array
np_array = np.array(json.loads(json_data))

Even the most complex datasets are no match for Pandas!

Conquering the beasts - numpy-specific types

Taming specialist NumPy data types for JSON serialization is as easy as pie:

import json
import numpy as np

# Tame the beast!
def numpy_serializer(obj):
    if isinstance(obj, np.generic):
        return obj.item()
    raise TypeError("Type not serializable")

np_array = np.array([1, 2, 3], dtype=np.int64)

# JSON - meet your new friend, np.int64
json_data = json.dumps(np_array.tolist(), default=numpy_serializer)