Shift ragged rows of Pandas DataFrame to clean data with partial string search

Question

After importing data from a .csv file, I have some data that looks similar to this (albeit order hundreds of columns and thousands of rows):

         4        5        6      7       8       9        10     11    12      13         14         15   16
0   302255Z  09005KT        1  1/4SM      BR     CLR  M00/M00  A3044   RMK    AO2A     SLP311  T10021002   $;
1   302232Z  08003KT        1    1/4      BR     CLR  M00/M00  A3044   RMK    AO2A     SLP310         $;  NaN
2   302225Z  09005KT        1  1/2SM      BR     CLR  M00/M00  A3044   RMK    AO2A     SLP309         $;  NaN
3   302155Z  08003KT        2  1/2SM      BR     CLR  M00/M00  A3043   RMK    AO2A     SLP306  T10001000   $;
4   302055Z  09004KT      3SM     BR     CLR   00/00    A3042    RMK  AO2A  SLP304  T00020002      56001   $;
5   301955Z  00000KT      3SM     BR     CLR   01/01    A3042    RMK  AO2A  SLP304  T00080008         $;  NaN
6   301855Z  09006KT      3SM     BR  FEW055   01/01    A3042    RMK  AO2A  SLP303  T00110011         $;  NaN
7   301655Z  10004KT        2  1/2SM      BR  FEW050  M00/M00  A3041   RMK    AO2A     SLP301  T10031003   $;
8   301610Z  09004KT        2  1/2SM      BR     CLR    00/00  A3041   RMK    AO2A     SLP301         $;  NaN
9   301555Z     AUTO  08005KT   4800      BR     CLR    01/01  A3041   RMK     AO2     SLP300  T00070007   $;
10  301509Z     AUTO  06003KT   4800      BR     CLR    01/01  A3041   RMK     AO2     SLP300         $;  NaN
11  301449Z     AUTO  10003KT   4000      BR     CLR    01/01  A3041   RMK     AO2     SLP300         $;  NaN
12  301355Z     AUTO  07004KT   6000      BR     CLR    02/02  A3041   RMK     AO2     SLP300  T00230023   $;
13  301255Z     AUTO  07003KT   6000      BR     CLR    02/02  A3041   RMK     AO2     SLP299  T00200020   $;
14  301055Z     AUTO  00000KT   9000      BR     CLR    04/04  A3040   RMK     AO2     SLP298  T00360036   $;

I abandoned trying to shift everything to match up correctly. Instead, I'm trying to create a new column that combines entries from columns 5 and 6 for those values ending in KT. And I'm creating a second new column for those values starting in T.

To start, I attempted pulling out all of the data that satisfied my criterion in rows 5 and 6 like so:

df1=df[df[5].str.contains("KT")].iloc[:,[0,5]]
df2=df[df[6].str.contains("KT")].iloc[:,[0,6]]

the .iloc value was an attempt to merge the results together. There has to be a slicker way to get this formatted. Any thoughts?

If helpful, here's a more simple data set:

row1=['a','b','c1K','d','e','foo','foo','f1111T','g','$']
row2=['a','b','foo','c2K','d','e','f4321T','g','$','$']
row3=['a','b','c3K','d','e','f1234T','g','$']
df=ps.DataFrame(zip(row1,row2,row3)).T
df1=df[df[2].str.contains("K")].iloc[:,[0,2]]
df2=df[df[3].str.contains("K")].iloc[:,[0,3]]

trying ps.concat([df1,df2],axis=0,join='outer') doesn't give what I'd like, it gives

   0    2    3
0  a  c1K  NaN
2  a  c3K  NaN
1  a  NaN  c2K

something like this would be prettier:

      0   
1  a  c1K  
2  a  c3K 
3  a  c2K

Answer 1

The following can be done in one pass (using loc as iloc doesn't allow boolean masking):

df1 = df[df[5].str.contains("KT")].iloc[:,[0,5]]

df1 = df.loc[df[5].str.contains("KT"), [0, 5]]

To get the end result, you could either concat these as a Series (to avoid aligning the columns), or change the name of the columns to be more descriptive before concating:

df1.columns = ['letter', 'code']
df2.columns = ['letter', 'code']
pd.concat([df1, df2], axis=0, ignore_index=True)