Question

I have a list of csv files ("file1", "file2", ...") that have two columns but do not have header labels. I’d like to assign them header labels and them as a DataFrame which is indexed by file and then indexed by those column labels. For example, I tried:

import pandas

mydict = {}
labels = ["col1", "col2"]
for myfile in ["file1", "file2"]:
  my_df = pandas.read_table(myfile, names=labels)
  # build dictionary of dataframe records
  mydict[myfile] = my_df

test = pandas.DataFrame(mydict)

this produces a DataFrame, test, indexed by "myfile1", "myfile2"... however, I’d like each of those to be indexed by "col1" and "col2" as well. My questions are:

1. how can I make it so the first index is file, and second index are columns I assigned (in the variable labels)? So that I can write:

   test["myfile1"]["col1"]

right now, test["myfile1"] only gives me a series of records.

1. also, how can I then reindex things so that the first indices are the column labels of each file and the second is the filename? So that I can write:

   test["col1"]["myfile1"]

or print test["col1"] and then see the value of "col1" shown for myfile1, myfile2, etc.

Answer 1

If you’re using pandas >= 0.7.0 (currently only available in the GitHub repository, though I’ll be making a release imminently!), you can concatenate your dict of DataFrames:

http://pandas.sourceforge.net/merging.html#more-concatenating-with-group-keys

In [6]: data
Out[6]: 
{'file1.csv':    A       B     
0  1.0914 -1.3538
1  0.5775 -0.2392
2 -0.2157 -0.2253
3 -2.4924  1.0896
4  0.6910  0.8992
5 -1.6196  0.3009
6 -1.5500  0.1360
7 -0.2156  0.4530
8  1.7018  1.1169
9 -1.7378 -0.3373,
 'file2.csv':    A       B      
0 -0.4948 -0.15551
1  0.6987  0.85838
2 -1.3949  0.25995
3  1.5314  1.25364
4  1.8582  0.09912
5 -1.1717 -0.21276
6 -0.2603 -1.78605
7 -3.3247  1.26865
8  0.7741 -2.25362
9 -0.6956  1.08774}


In [10]: cdf = concat(data, axis=1)

In [11]: cdf
O    ut[11]: 
   file1.csv          file2.csv         
   A          B       A          B      
0  1.0914    -1.3538 -0.4948    -0.15551
1  0.5775    -0.2392  0.6987     0.85838
2 -0.2157    -0.2253 -1.3949     0.25995
3 -2.4924     1.0896  1.5314     1.25364
4      0.6910     0.8992  1.8582     0.09912
5 -1.6196     0.3009 -1.1717    -0.21276
6 -1.5500     0.1360 -0.2603    -1.78605
7 -0.2156     0.4530 -3.3247     1.26865
8  1.7018     1.1169  0.7741    -2.25362
9 -1.7378    -0.3373 -0.6956     1.08774

Then if you wish to switch the order of the column indexes, you can do:

In [14]: cdf.swaplevel(0, 1, axis=1)
Out[14]: 
   A          B          A          B        
   file1.csv  file1.csv  file2.csv  file2.csv
0  1.0914    -1.3538    -0.4948    -0.15551  
1  0.5775    -0.2392     0.6987     0.85838  
2 -0.2157    -0.2253    -1.3949     0.25995  
3 -2.4924     1.0896     1.5314     1.25364  
4  0.6910     0.8992     1.8582     0.09912  
5 -1.6196     0.3009    -1.1717    -0.21276  
6 -1.5500     0.1360    -0.2603    -1.78605  
7 -0.2156     0.4530    -3.3247     1.26865  
8  1.7018     1.1169     0.7741    -2.25362  
9 -1.7378    -0.3373    -0.6956     1.08774  

Alternately, and perhaps a bit straightforwardly, you can use a Panel:

In [16]: p = Panel(data)

In [17]: p
Out[17]: 
<class 'pandas.core.panel.Panel'>
Dimensions: 2 (items) x 10 (major) x 2 (minor)
Items: file1.csv to file2.csv
Major axis: 0 to 9
Minor axis: A to B

In [18]: p = p.swapaxes(0, 2)

In [19]: p
Out[19]: 
<class 'pandas.core.panel.Panel'>
Dimensions: 2 (items) x 10 (major) x 2 (minor)
Items: A to B
Major axis: 0 to 9
Minor axis: file1.csv to file2.csv

In [20]: p['A']
Out[20]: 
   file1.csv  file2.csv
0  1.0914    -0.4948   
1  0.5775     0.6987   
2 -0.2157    -1.3949   
3 -2.4924     1.5314   
4  0.6910     1.8582   
5 -1.6196    -1.1717   
6 -1.5500    -0.2603   
7 -0.2156    -3.3247   
8  1.7018     0.7741   
9 -1.7378    -0.6956