Exporting Pandas series as JSON - error with numpy types

Question

I want to export some rows from a Pandas DataFrame to JSON. However, when exporting a columns I got an error:

TypeError: False is not JSON serializable

or

TypeError: 0 is not JSON serializable

I looked up in my data and the problem occurs for numpy.int64 and numpy.bool_ (numpy.float64 works fine).

For example, the problem appears for the following:

import pandas as pd
import simplejson as json

df = pd.DataFrame([[False,0],[True,1]], columns=['a','b'])
json.dumps(df.ix[0].to_dict())

(The same thing happens for dict(df.ix[0])).

Is there a simple workaround to export Pandas Series to JSON?

Or at least, a function that coerce any numpy type to the closest type compatible with JSON?

Viktor Kerkez · Accepted Answer · 2013-09-03 18:40:45Z

7

DataFrame has a method to export itself to json string:

>>> df.to_json()
'{"a":{"0":false,"1":true},"b":{"0":0,"1":1}}'

You can also export it directly to a file:

>>> df.to_json(filename)

answered Sep 3, 2013 at 18:40

Viktor Kerkez

46.8k13 gold badges109 silver badges88 bronze badges

Sign up to request clarification or add additional context in comments.

5 Comments

It works with DataFrames, thanks. Does it work also with Series? I am getting OverflowError: Maximum recursion level reached.

@PiotrMigdal It does work with Series. Can you post a Series object that reproduces the stack overflow?

@PhillipCloud For example the one in the question, i.e. df.ix[0].to_json() for df = pd.DataFrame([[False,0],[True,1]], columns=['a','b']).

I get '{"a":false,"b":0}', as expected. That's on pandas git master, which may have this fix in it (and not in the 0.12 release). Let me see.

@PhillipCloud Yup, it's a bug in 0.12, but it's fixed on master.