TypeError: Object of type float32 is not JSON serializable

TypeError: Object of type float32 is not JSON serializable #

The Python "TypeError: Object of type float32 is not JSON serializable" occurs when we try to convert a numpy float32 object to a JSON string. To solve the error, convert the numpy float to a Python float before converting it to JSON, e.g. float(my_numpy_float).

Here is an example of how the error occurs.

main.py

Copied!
import json
import numpy as np

salary = np.power(100, 2.75, dtype=np.float32)

# ⛔️ TypeError: Object of type float32 is not JSON serializable
json_str = json.dumps({'salary': salary})

We tried passing a numpy float32 object to the json.dumps() method but the method doesn't handle numpy floats by default.

To solve the error, use the built-in float() (or str()) constructor to convert the numpy float to a native Python float before serializing it.

main.py

Copied!
import json
import numpy as np

salary = np.power(100, 2.75, dtype=np.float32)

print(salary)  # 👉️ 316227.78

# ✅ convert to float
json_str = json.dumps({'salary': float(salary)})

print(json_str)  # 👉️ '{"salary": 316227.78125}'
print(type(json_str))  # 👉️ <class 'str'>

We used the built-in float() constructor to convert the numpy float32 value to a native Python float.

Note that if you are worried about losing precision, you can use the str() constructor to convert the value to a string instead.

The default JSON encoder handles float and str values, so we can use a native Python float instead of a numpy float32 when serializing to JSON.

The json.dumps method converts a Python object to a JSON formatted string.

Alternatively, you can extend from the JSONEncoder class and handle the conversions in a default method.

main.py

Copied!
import json
import numpy as np


class NpEncoder(json.JSONEncoder):
    def default(self, obj):
        if isinstance(obj, np.integer):
            return int(obj)
        if isinstance(obj, np.floating):
            # 👇️ alternatively use str()
            return float(obj)
        if isinstance(obj, np.ndarray):
            return obj.tolist()
        return json.JSONEncoder.default(self, obj)


salary = np.power(100, 2.75, dtype=np.float32)

json_str = json.dumps({'salary': salary}, cls=NpEncoder)

print(json_str)  # 👉️ {"salary": 316227}
print(type(json_str))  # 👉️ <class 'str'>

We extended from the JSONEncoder class.

The JSONEncoder class supports the following objects and types by default.

Python	JSON
dict	object
list, tuple	array
str	string
int, float, int and float derived Enums	number
True	true
False	false
None	null

Notice that the JSONEncoder class doesn't support numpy float32 to JSON conversion by default.

We can handle this by extending from the class and implementing a default() method that returns a serializable object.

main.py

Copied!
import json
import numpy as np


class NpEncoder(json.JSONEncoder):
    def default(self, obj):
        if isinstance(obj, np.integer):
            return int(obj)
        if isinstance(obj, np.floating):
            return float(obj)
        if isinstance(obj, np.ndarray):
            return obj.tolist()
        return json.JSONEncoder.default(self, obj)

If the passed in object is an instance of np.integer, we convert the object to a Python int and return the result.

If the passed in object is an instance of np.floating, we convert it to a Python float and return the result.

Note that you can use the str() constructor instead of float() if you worry about losing precision.

If the object is an instance of np.ndarray, we convert it to a Python list and return the result.

In all other cases, we let the base classes' default method do the serialization.

To use a custom JSONEncoder, specify it with the cls keyword argument in your call to the json.dumps() method.

main.py

Copied!
import json
import numpy as np


class NpEncoder(json.JSONEncoder):
    def default(self, obj):
        if isinstance(obj, np.integer):
            return int(obj)
        if isinstance(obj, np.floating):
            return float(obj)
        if isinstance(obj, np.ndarray):
            return obj.tolist()
        return json.JSONEncoder.default(self, obj)


salary = np.power(100, 2.75, dtype=np.float32)

# ✅ pass cls kwarg
json_str = json.dumps({'salary': salary}, cls=NpEncoder)

print(json_str)  # 👉️ {"salary": 316227}
print(type(json_str))  # 👉️ <class 'str'>

If you don't provide the cls kwarg, the default JSONEncoder is used.

TypeError: Object of type float32 is not JSON serializable

TypeError: Object of type float32 is not JSON serializable #

Borislav Hadzhiev