如何在 pandas 中合并或加入堆叠的 dataframe？

Question

I cannot find this question answered elsewhere; I would like to do a SQL-like join in pandas but with the slight twist that one dataframe is stacked. I have created a dataframe A with a stacked column index from a csv file in pandas that looks as follows:

|           |      | 2013-01-04 | 2013-01-07 |
|----------:|-----:|-----------:|-----------:|
| Adj Close |  OWW | NaN        | NaN        |
|   Close   | OXLC | 4.155157   | 4.147217   |
|           |  OXM | 40.318089  | 42.988800  |
|           |  OXY | 50.416079  | 62.934800  |

The original csv had repeated what is in the 1st column for every entry like so:

|           |      | 2013-01-04 | 2013-01-07 |
|----------:|-----:|-----------:|-----------:|
| Adj Close |  OWW | NaN        | NaN        |
|   Close   | OXLC | 4.155157   | 4.147217   |
|   Close   |  OXM | 40.318089  | 42.988800  |
|   Close   |  OXY | 50.416079  | 62.934800  |

The original csv was the transposed version of this. Pandas chose to stack that when converting to dataframe. (I used this code: pd.read_csv(file, header = [0,1], index_col=0).T)

In another csv/dataframe BI have for all of those so-called ticker symbols another ID that I would rather like to use: CIK.

| CIK     | Ticker | Name                                           |
|---------|--------|------------------------------------------------|
| 1090872 | A      | Agilent Technologies Inc                       |
| 4281    | AA     | Alcoa Inc                                      |
| 1332552 | AAACU  | Asia Automotive Acquisition Corp               |
| 1287145 | AABB   | Asia Broadband Inc                             |
| 1024015 | AABC   | Access Anytime Bancorp Inc                     |
| 1099290 | AAC    | Sinocoking Coal & Coke Chemical Industries Inc |
| 1264707 | AACC   | Asset Acceptance Capital Corp                  |
| 849116  | AACE   | Ace Cash Express Inc                           |
| 1409430 | AAGC   | All American Gold Corp                         |
| 948846  | AAI    | Airtran Holdings Inc                           |

Desired output: I would like to have the CIK instead of the ticker in a new dataframe otherwise identical to A.

Now in SQL I could easily join on A.name_of_2nd_column = b.Ticker since the table would just have the entry in the 1st column repeated in every line (like the original csv) and the column would have a name but in pandas I cannot. I tried this code:

result = pd.merge(data, tix, how='left', left_on=[1] right_on=['Ticker'])

How do I tell pandas to use the 2nd column as the key and/or interpret the first column just as repeated values?

Answer 1

What you want is to transcode from one set of identifiers (tickers) to another (CIKs used in the SEC Edgar database I presume).

I would

1. Turn the index column(s) into ordinary columns especially if these are multi-indexes, possibly after renaming the index columns

A.index.names=('Data','Ticker')
A = A.reset_index()

2. Use the map method to transcode the tickers to CIKs

transco = B.set_index('Ticker').CIK
A['CIK'] = A.Ticker.map(transco)

3. Eventually reindex from what you want, dropping unused indices

A = A.drop('Ticker', axis=1).set_index(['Data','CIK'])

As a step 2.5, you might want to remove the entries for which you don't have any CIK, eg by doing:

A = A[A.CIK.notnull()]
A.CIK = A.CIK.astype(int)

You could also merge after doing a reset_index() , but I would avoid that, as you might end up with uselessly large dataframes, as the result of the merge will have a names column. This can grow if you have many different types of data (Adj Close, Close, etc.).

Answer 2

I was eventually able to do it the following way:

df = A
tix = B 

ticker_2_CIK = dict(zip(tix.Ticker,tix.CIK))  # create a dict

tmp = df.reset_index().assign(CIK=lambda x: x['ticker'].map(ticker_2_CIK)) # use dict to find the correct value for colum 

# data was unclean, some ticker symbols were created after the period my data is from 
# and data was incomplete with some tickers missing
solution = tmp.dropna(subset=['CIK']).astype({'CIK':int})