Handling of nested JSON records

[wesm]

Is there a simple way of grabbing nested keys when constructing a Pandas Dataframe from JSON. Using the example JSON from below, how would I build a Dataframe that uses this column_header = ['id_str', 'text', 'user.screen_name'], (i.e. how do I get the 'screen_name' from the 'user' key without flattening the JSON).

Thanks,

Vishal

{   u'_id': ObjectId('4f8b95e8a504d022e2000000'),
    u'contributors': None,
    u'coordinates': None,
    u'created_at': u'Mon Apr 16 03:45:44 +0000 2012',
    u'entities': {   u'hashtags': [],
                     u'urls': [   {   u'display_url': u'stks.co/3M4X',
                                      u'expanded_url': u'http://stks.co/3M4X',
                                      u'indices': [72, 92],
                                      u'url': u'http://t.co/mGdTYusF'}],
                     u'user_mentions': []},
    u'favorited': False,
    u'geo': None,
    u'id': 191734090783916032L,
    u'id_str': u'191734090783916032',
    u'in_reply_to_screen_name': None,
    u'in_reply_to_status_id': None,
    u'in_reply_to_status_id_str': None,
    u'in_reply_to_user_id': None,
    u'in_reply_to_user_id_str': None,
    u'place': None,
    u'possibly_sensitive': False,
    u'possibly_sensitive_editable': True,
    u'processed_metadata': {   u'created_date': datetime.datetime(2012, 4, 16, 3, 45, 44, tzinfo=<bson.tz_util.FixedOffset object at 0x104d63790>),
                               u'search_queries': [   u'$AAPL',
                                                      u'$MSFT',
                                                      u'$GOOG'],
                               u'source': u'Twitter Streaming API'},
    u'retweet_count': 0,
    u'retweeted': False,
    u'source': u'<a href="http://stocktwits.com" rel="nofollow">StockTwits Web</a>',
    u'text': u'Interesting infographic on the internet and evolution of social media \u2794 http://t.co/mGdTYusF $FB $GOOG $TWIT $LNKD $AOL',
    u'truncated': False,
    u'user': {   u'_id': u'speculatethemkt',
                 u'contributors_enabled': False,
                 u'created_at': u'Tue Nov 30 02:28:20 +0000 2010',
                 u'default_profile': False,
                 u'default_profile_image': False,
                 u'description': u"I'm a 22-year old full-time forex trader living the location independent lifestyle. Author of The Trading Elite \u2794 http://amzn.to/I0nacY",
                 u'favourites_count': 1,
                 u'follow_request_sent': None,
                 u'followers_count': 19658,
                 u'following': None,
                 u'friends_count': 596,
                 u'geo_enabled': False,
                 u'id': 221226895,
                 u'id_str': u'221226895',
                 u'is_translator': False,
                 u'lang': u'en',
                 u'listed_count': 6,
                 u'location': u'Portland, OR',
                 u'name': u'Jared M.',
                 u'notifications': None,
                 u'processed_metadata': {   u'created_date': datetime.datetime(2012, 4, 16, 3, 45, 44, tzinfo=<bson.tz_util.FixedOffset object at 0x104d63790>),
                                            u'search_queries': [   u'$AAPL',
                                                                   u'$MSFT',
                                                                   u'$GOOG'],
                                            u'source': u'Twitter Streaming API'},
                 u'profile_background_color': u'4f4f4f',
                 u'profile_background_image_url': u'http://a0.twimg.com/profile_background_images/438353849/TWITTER-BG.jpg',
                 u'profile_background_image_url_https': u'https://si0.twimg.com/profile_background_images/438353849/TWITTER-BG.jpg',
                 u'profile_background_tile': False,
                 u'profile_image_url': u'http://a0.twimg.com/profile_images/2039368182/twitpic1_normal.png',
                 u'profile_image_url_https': u'https://si0.twimg.com/profile_images/2039368182/twitpic1_normal.png',
                 u'profile_link_color': u'bd0000',
                 u'profile_sidebar_border_color': u'eeeeee',
                 u'profile_sidebar_fill_color': u'efefef',
                 u'profile_text_color': u'333333',
                 u'profile_use_background_image': True,
                 u'protected': False,
                 u'screen_name': u'speculatethemkt',
                 u'show_all_inline_media': True,
                 u'statuses_count': 492,
                 u'time_zone': u'Pacific Time (US & Canada)',
                 u'url': u'http://www.speculatethemarkets.com',
                 u'utc_offset': -28800,
                 u'verified': False}}

[hayd]

related: http://stackoverflow.com/questions/16444797/which-is-the-most-efficient-way-of-loading-a-json-dataset-into-pandas-dataframes

[jreback]

this is invalid JSON (according to jsonlint), but generalized inference is IMHO too complicated, but #3804 should be able to do some of this. close this issue?

[hayd]

Is it feasible to grab the user section, (actually this example from the other thread is better):

Convert the (data, posts) section to DataFrame

s = r'''{
    "status": "success",
    "data": {
        "posts": [
            {
                "id": 1,
                "title": "A blog post",
                "body": "Some useful content"
            },
            {
                "id": 2,
                "title": "Another blog post",
                "body": "More content"
            }
        ]
    }
}'''

read_json(s, grab_nest=(data, posts)) # some better argument name
                  body  id              title
0  Some useful content   1        A blog post
1         More content   2  Another blog post

[jreback]

This is probably getting too cute.....

In [52]: def extract(df, l):
   ....:     for e in l:
   ....:         df = df[e]
   ....:     return df
   ....: 

In [54]: DataFrame.extract = extract

In [56]: DataFrame(pd.read_json(s).extract(['data','posts']))
Out[56]: 
                  body  id              title
0  Some useful content   1        A blog post
1         More content   2  Another blog post

[hayd]

Ha! Perhaps less overhead to do pd.DataFrame(extract(pd.io.json.loads(s), ('data', 'posts'))), but in either case we would lose the datetime parsing atm.

[hayd]

This could be a reasonably ok solution... tricky with orient (?), then parse_dates or whatever?

...other choice is just to loads/dumps/parse? :s

[hayd]

here's a bigish nested json: https://github.com/hayd/sf-city-lots-json ~200mb

I think I'd want to extract ['features'].

I'm on an incredibly old macbook air, hence slow timings:

In [9]: %time with open('citylots.json', 'r') as f: pd.read_json(pd.io.json.dumps(extract(pd.io.json.loads(f.read()), ['features'])))
CPU times: user 29.13 s, sys: 28.27 s, total: 57.40 s
Wall time: 304.71 s

In [10]: %time with open('citylots.json', 'r') as f: pd.DataFrame(extract(pd.io.json.loads(f.read()), ['features']))
CPU times: user 11.96 s, sys: 11.79 s, total: 23.75 s
Wall time: 136.50 s

In [11]: %time with open('citylots.json', 'r') as f: pd.read_json(f.read())CPU times: user 13.47 s, sys: 10.41 s, total: 23.88 s
Wall time: 77.47 s

What is an extreme for reading in json?

[jreback]

After I figured out all I needed to do was clone the repository! (these also include full dtype conversions) FYI

In [3]: %time with open('citylots.json', 'r') as f: pd.read_json(pd.io.json.dumps(extract(pd.io.json.loads(f.read()), ['features'])))
CPU times: user 13.03 s, sys: 0.50 s, total: 13.53 s
Wall time: 15.12 s

In [6]: %time with open('citylots.json', 'r') as f: pd.DataFrame(extract(pd.io.json.loads(f.read()), ['features']))
CPU times: user 6.03 s, sys: 0.08 s, total: 6.11 s
Wall time: 6.13 s

In [7]: %time with open('citylots.json', 'r') as f: pd.read_json(f.read())
CPU times: user 6.27 s, sys: 0.16 s, total: 6.44 s
Wall time: 6.45 s

[jreback]

see #3876

[hayd]

Wow, I should never do any data analysis on that laptop... (sorry I forgot that you had to clone it).

But what I mean is, you'd lose the control from the read_json arguments. Will be interesting to see if this use case comes up a lot "in the wild".

... really this really this kind of stuff should be done the other end, e.g. with http://www.elasticsearch.org/guide/reference/api/get/ (getting the _source directly).

[wesm]

I have a JSON normalization function I can clean up and make a PR before before anyone goes crazy writing one to save you some time. It would be nice to have a higher performance one at some point though

[nehalecky]

Hey @wesm and @hayd! I've been keeping my eye on this thread for a few days now—really impressive all the work that built up to this, thank you. Anyways, I though you might know that I could go crazy writing some JSON normalization soon. Perhaps I should wait? :)

Thanks for all.

[hayd]

Related: https://groups.google.com/forum/#!topic/pydata/XkiWtZKT698 (json is a list of nested dictionaries...)

Hey @nehalecky , I think @wesm says he has something in the works already, so perhaps if you can hold off til he's pushed, then you could hack that? :)

[ImNick23]

Is there a simple way of grabbing nested keys when constructing a Pandas Dataframe from JSON. Using the example JSON from below, how would I build a Dataframe that uses this column_header = ['id_str', 'text', 'user.screen_name'], (i.e. how do I get the 'screen_name' from the 'user' key without flattening the JSON).

Thanks,

Vishal

{   u'_id': ObjectId('4f8b95e8a504d022e2000000'),
    u'contributors': None,
    u'coordinates': None,
    u'created_at': u'Mon Apr 16 03:45:44 +0000 2012',
    u'entities': {   u'hashtags': [],
                     u'urls': [   {   u'display_url': u'stks.co/3M4X',
                                      u'expanded_url': u'http://stks.co/3M4X',
                                      u'indices': [72, 92],
                                      u'url': u'http://t.co/mGdTYusF'}],
                     u'user_mentions': []},
    u'favorited': False,
    u'geo': None,
    u'id': 191734090783916032L,
    u'id_str': u'191734090783916032',
    u'in_reply_to_screen_name': None,
    u'in_reply_to_status_id': None,
    u'in_reply_to_status_id_str': None,
    u'in_reply_to_user_id': None,
    u'in_reply_to_user_id_str': None,
    u'place': None,
    u'possibly_sensitive': False,
    u'possibly_sensitive_editable': True,
    u'processed_metadata': {   u'created_date': datetime.datetime(2012, 4, 16, 3, 45, 44, tzinfo=<bson.tz_util.FixedOffset object at 0x104d63790>),
                               u'search_queries': [   u'$AAPL',
                                                      u'$MSFT',
                                                      u'$GOOG'],
                               u'source': u'Twitter Streaming API'},
    u'retweet_count': 0,
    u'retweeted': False,
    u'source': u'<a href="http://stocktwits.com" rel="nofollow">StockTwits Web</a>',
    u'text': u'Interesting infographic on the internet and evolution of social media \u2794 http://t.co/mGdTYusF $FB $GOOG $TWIT $LNKD $AOL',
    u'truncated': False,
    u'user': {   u'_id': u'speculatethemkt',
                 u'contributors_enabled': False,
                 u'created_at': u'Tue Nov 30 02:28:20 +0000 2010',
                 u'default_profile': False,
                 u'default_profile_image': False,
                 u'description': u"I'm a 22-year old full-time forex trader living the location independent lifestyle. Author of The Trading Elite \u2794 http://amzn.to/I0nacY",
                 u'favourites_count': 1,
                 u'follow_request_sent': None,
                 u'followers_count': 19658,
                 u'following': None,
                 u'friends_count': 596,
                 u'geo_enabled': False,
                 u'id': 221226895,
                 u'id_str': u'221226895',
                 u'is_translator': False,
                 u'lang': u'en',
                 u'listed_count': 6,
                 u'location': u'Portland, OR',
                 u'name': u'Jared M.',
                 u'notifications': None,
                 u'processed_metadata': {   u'created_date': datetime.datetime(2012, 4, 16, 3, 45, 44, tzinfo=<bson.tz_util.FixedOffset object at 0x104d63790>),
                                            u'search_queries': [   u'$AAPL',
                                                                   u'$MSFT',
                                                                   u'$GOOG'],
                                            u'source': u'Twitter Streaming API'},
                 u'profile_background_color': u'4f4f4f',
                 u'profile_background_image_url': u'http://a0.twimg.com/profile_background_images/438353849/TWITTER-BG.jpg',
                 u'profile_background_image_url_https': u'https://si0.twimg.com/profile_background_images/438353849/TWITTER-BG.jpg',
                 u'profile_background_tile': False,
                 u'profile_image_url': u'http://a0.twimg.com/profile_images/2039368182/twitpic1_normal.png',
                 u'profile_image_url_https': u'https://si0.twimg.com/profile_images/2039368182/twitpic1_normal.png',
                 u'profile_link_color': u'bd0000',
                 u'profile_sidebar_border_color': u'eeeeee',
                 u'profile_sidebar_fill_color': u'efefef',
                 u'profile_text_color': u'333333',
                 u'profile_use_background_image': True,
                 u'protected': False,
                 u'screen_name': u'speculatethemkt',
                 u'show_all_inline_media': True,
                 u'statuses_count': 492,
                 u'time_zone': u'Pacific Time (US & Canada)',
                 u'url': u'http://www.speculatethemarkets.com',
                 u'utc_offset': -28800,
                 u'verified': False}}

I am trying convert this kind of 3/4 levels of nested json into python dataframe with every attribute present in it. I am able to extract up to 2 levels. How can i do for rest?