Pandas DataFrame is not JSON serializable

[s-celles]

Hello,

I'm trying to build a RESTful api with Flask-RESTful which will return Pandas DataFrame as JSON data.

from flask import Flask
from flask.ext import restful
from flask.ext.restful import Resource, Api
import pandas as pd
import click
import logging

app = Flask(__name__)
api = restful.Api(app)

class DataWebservice(Resource):
    def get(self, size):
        logging.info("get %d" % size)
        # get DB conn
        # df = pd.read_sql(...) # USE LIMIT
        df = pd.DataFrame({"col1": [1]*size, "col2": [2]*size})
        #return(df.to_json())
        return(df)

api.add_resource(DataWebservice, '/api/v1/solar/df/get/<int:size>')

@click.command()
@click.option('--host', default='127.0.0.1', \
    help="host (127.0.0.1 or 0.0.0.0 to accept all ip)")
@click.option('--debug/--no-debug', default=False, help="debug mode")
def main(debug, host):
    app.run(host=host, debug=debug)

if __name__ == '__main__':
    main()

I run server using
$ python server.py --debug

I run client using
$ curl http://127.0.0.1:5000/api/v1/solar/df/get/10

but I get the following error

TypeError:    col1  col2
0     1     2
1     1     2
...
8     1     2
9     1     2

[10 rows x 2 columns] is not JSON serializable

So it seems that Pandas DataFrame are not JSON serializable.

I try this using IPython

size = 10
df = pd.DataFrame({"col1": [1]*size, "col2": [2]*size})
json.dumps(df)

It raises same error.
I'm aware that DataFrame have method named to_json()
but it doesn't help me much as my server will return escaped strings such as

"{\"col1\":{\"0\":1,\"1\":1,\"2\":1,\"3\":1,\"4\":1,\"5\":1,\"6\":1,\"7\":1,\"8\":1,\"9\":1},\"col2\":{\"0\":2,\"1\":2,\"2\":2,\"3\":2,\"4\":2,\"5\":2,\"6\":2,\"7\":2,\"8\":2,\"9\":2}}"

see flask-restful/flask-restful#269

Kind regards

[jreback]

what are you asking here? pandas provides a read/write API to JSON: http://pandas.pydata.org/pandas-docs/stable/io.html#json, which should be deserializable by foreign JSON parsers. You must use .to_json() as a dataframe as many custom types.

[s-celles]

Hello,

Thanks I know to_json() method
I thought Pandas DataFrame could inherit an other class to become directly "JSON serializable".
So json.dumps(df) could return exactly the same result as df.to_json().
So in this Flask view we could directly return DataFrame (in fact jsonify(df)) instead of doing:

    resp = Response(response=df.to_json(),
        status=200,
        mimetype="application/json")
    return(resp)

But maybe I'm wrong and there is no way for json.dumps(df) to return a JSON string.
I thought that what I'm asking is more a syntactic sugar than a major improvement.

Kind regards

[jreback]

This is the typical way to extend the default json encoder

In [9]: class JSONEncoder(json.JSONEncoder):
   ...:     def default(self, obj):
   ...:         if hasattr(obj,'to_json'):
   ...:             return obj.to_json()
   ...:         return json.JSONEncoder.default(self, obj)
   ...:     

In [10]: json.dumps(df, cls=JSONEncoder)
Out[10]: '"{\\"col1\\":{\\"0\\":1,\\"1\\":1,\\"2\\":1,\\"3\\":1,\\"4\\":1,\\"5\\":1,\\"6\\":1,\\"7\\":1,\\"8\\":1,\\"9\\":1},\\"col2\\":{\\"0\\":2,\\"1\\":2,\\"2\\":2,\\"3\\":2,\\"4\\":2,\\"5\\":2,\\"6\\":2,\\"7\\":2,\\"8\\":2,\\"9\\":2}}"'

[s-celles]

So you think it's better to define our own JSONEncoder. The problem is that I don't think jsonify support cls argument for encoder... but that's an other problem

Moreover I noticed

In [20]: df.to_json()
Out[20]: '{"col1":{"0":1,"1":1,"2":1,"3":1,"4":1,"5":1I"6":1,"7":1,"8":1,"9":1},"col2":{"0":2,"1":2,"2":2,"3":2,"4":2,"5":2,"6":2,"7":2,"8":2,"9":2}}'

In [21]: json.dumps(df, cls=JSONEncoder)
Out[21]: '"{\"col1\":{\"0\":1,\"1\":1,\"2\":1,\"3\":1,\"4\":1,\"5\":1,\"6\":1,\"7\":1,\"8\":1,\"9\":1},\"col2\":{\"0\":2,\"1\":2,\"2\":2,\"3\":2,\"4\":2,\"5\":2,\"6\":2,\"7\":2,\"8\":2,\"9\":2}}"'

it's like a string inside a string...

What I want to encode is in fact

        d = {"success":1 , "return": df}

I know I can do

        d = "{\"success\":1, \"return\":%s}" % df.to_json()

but I feel that's not the right way of doing

[jreback]

HTH

their are lots of options for to_json(), see docs: http://pandas.pydata.org/pandas-docs/stable/io.html#json

not really sure what you are doing

closing this as its out of scope for pandas.

[s-celles]

Please try this:

d = {"success":1 , "return": df}

ser=json.dumps(d, cls=JSONEncoder)

unser=json.loads(ser)

type(unser["return"])

it returns unicode I could expect dict

[jreback]

you need to teach json how to do this by defining an object hook. http://www.yilmazhuseyin.com/blog/dev/advanced_json_manipulation_with_python/

In [13]: pd.read_json(json.loads(ser)['return'])
Out[13]: 
   col1  col2
0     1     2
1     1     2
2     1     2
3     1     2
4     1     2
5     1     2
6     1     2
7     1     2
8     1     2
9     1     2

[s-celles]

Thanks for this tutorial but there is a difference between the 2 serialized versions

In [74]: d = {"success":1 , "return": df}

In [75]: d
Out[75]:
{'return':    col1  col2
 0     1     2
 1     1     2
 2     1     2
 3     1     2
 4     1     2
 5     1     2
 6     1     2
 7     1     2
 8     1     2
 9     1     2

 [10 rows x 2 columns], 'success': 1}

In [76]: dat_json = "{\"success\":1, \"return\":%s}" % df.to_json()

In [77]: dat_json
Out[77]: '{"success":1, "return":{"col1":{"0":1,"1":1,"2":1,"3":1,"4":1,"5":1,"6":1,"7":1,"8":1,"9":1},"col2":{"0":2,"1":2,"2":2,"3":2,"4":2,"5":2,"6":2,"7":2,"8":2,"9":2}}}'

In [78]: dat_json2 = json.dumps(d, cls=JSONEncoder)

In [79]: dat_json2
Out[79]: '{"return": "{\\"col1\\":{\\"0\\":1,\\"1\\":1,\\"2\\":1,\\"3\\":1,\\"4\\":1,\\"5\\":1,\\"6\\":1,\\"7\\":1,\\"8\\":1,\\"9\\":1},\\"col2\\":{\\"0\\":2,\\"1\\":2,\\"2\\":2,\\"3\\":2,\\"4\\":2,\\"5\\":2,\\"6\\":2,\\"7\\":2,\\"8\\":2,\\"9\\":2}}", "success": 1}'

Is there a clean solution with a custom encoder for object with to_json method (like DataFrame) to output correctly JSON (without extra quotes)

[jreback]

this is maybe a question for SO, I don't use custom uncoders at all. It seems a whole lot simpler for you to simply call df.to_json() which returns a string no? (then since you know the structure, then just pd.read_json() when you need to.

[s-celles]

I understand your reply... but imagine you have several dataframes to output into the same JSON message.... doing things this way is not very clear. I think it's much more clear to have a dict structure which can contains several df (and other data) and after serialize it.

About deserialization... yes that's not a problem... I know structure and where are dataframes.

[jreback]

ok, as I said, you maybe want to do custom encoding/decoding like I showed above. You need to write that. a dataframe can be turned into json via to_json and read by read_json. when to do that is up 2 u.