将Pandas DataFrame转换为Python列表

Question

I have the following dataframe:

In [137]: counts
Out[137]: 
SourceColumnID                    3029903181  3029903182  3029903183  3029903184  ResponseCount
ColID      QuestionID RowID                                                                    
3029903193 316923119  3029903189         773         788         778         803           3142
3029903194 316923119  3029903189         766         799         782         773           3120

[2 rows x 5 columns]

that works well for what I want when I access it via iloc:

In [138]: counts.iloc[0][3029903181]
Out[138]: 773

but when I convert this to a dict it formats it in a way that isn't accessible in the same way anymore:

In [139]: counts.to_dict()
Out[139]: 
{3029903181: {(3029903193, 316923119, 3029903189): 773,
  (3029903194, 316923119, 3029903189): 766},
 3029903182: {(3029903193, 316923119, 3029903189): 788,
  (3029903194, 316923119, 3029903189): 799},
 3029903183: {(3029903193, 316923119, 3029903189): 778,
  (3029903194, 316923119, 3029903189): 782},
 3029903184: {(3029903193, 316923119, 3029903189): 803,
  (3029903194, 316923119, 3029903189): 773},
 'ResponseCount': {(3029903193, 316923119, 3029903189): 3142,
  (3029903194, 316923119, 3029903189): 3120}}

In [140]: counts.to_dict('list')
Out[140]: 
{3029903181: [773, 766],
 3029903182: [788, 799],
 3029903183: [778, 782],
 3029903184: [803, 773],
 'ResponseCount': [3142, 3120]}

I need to convert this datastructure to a standard python object to return for an API to consume it.

Should I have created the table in a different format?

I started with this DataFrame:

In [141]: df
Out[141]: 
        ColID  QuestionID  ResponseCount       RowID  SourceColumnID
0  3029903193   316923119            773  3029903189      3029903181
1  3029903193   316923119            788  3029903189      3029903182
2  3029903193   316923119            778  3029903189      3029903183
3  3029903193   316923119            803  3029903189      3029903184
4  3029903194   316923119            766  3029903189      3029903181
5  3029903194   316923119            799  3029903189      3029903182
6  3029903194   316923119            782  3029903189      3029903183
7  3029903194   316923119            773  3029903189      3029903184

[8 rows x 5 columns]

and converted it to a pivot table like this:

counts = df.pivot_table(values='ResponseCount', rows=['ColID', 'QuestionID', 'RowID'], cols='SourceColumnID', aggfunc='sum')

I'm really looking for the datastructure to come out looking like this:

[
  {
    'QuestionID': 316923119, 
    'RowID': 3029903189, 
    'ColID': 3029903193, 
    '3029903181': 773,
    '3029903182': 788,
    '3029903183': 778,
    '3029903184': 803,
    'ResponseCount': 3142
  },
  {
    'QuestionID': 316923119, 
    'RowID': 3029903189, 
    'ColID': 3029903194, 
    '3029903181': 766,
    '3029903182': 799,
    '3029903183': 782,
    '3029903184': 773,
    'ResponseCount': 3120
  },
]

Answer 1

I believe you want counts.reset_index().to_dict('records') .

Using 'records' with to_dict makes it give you a list of dicts, one dict per row, which is what you want. You need to use reset_index() to get the index information in as columns (because 'records' throws away the index). Conceptually, the dicts you say you want don't distinguish between what's in the index of your pivot table and what's in the columns (you just want all index and column labels as keys in the dict), so you need to reset_index to remove the index/column distinction.