Split a column by regex and keep only matches in pandas

I need to have a column per my regex match.

Consider below dataframe:

Letter      Actions
r1          a95,a20
r2          a30,a12-rf,a15,a75
r3          0
r4          a15,a74
r5          a11

I expect:

Letter      Action1      Action2      Action3      Action4
r1          a95          a20          
r2          a30          a12          a15          a75
r3          
r4          a15          a74
r5          a11

I have below but it keeps every thing else as well while I just need the pattern matches:

import pandas as pd

df = pd.DataFrame(
    [['r1', 'a95,a20'],
     ['r2', 'a30,a12-rf,a15,a75'],
     ['r3', '0'],
     ['r4', 'a15,a74'],
     ['r5', 'a11']],
    columns=['Letter', 'Actions'])

Actions = df['Actions'].str.split(r'([a]\d{2})', expand=True)

Here is an option using str.extractall()

(df.set_index('Letter')['Actions']
.str.extractall(r'([a]\d{2})')[0]
.unstack()
.rename(lambda x: 'Action{}'.format(x+1),axis=1)
.reset_index()
.rename_axis(None,axis=1))

Another possibility is using str.split()

df.set_index('Letter')['Actions'].str.split(r'(?:\D+)?,',expand=True)

Output:

  Letter Action1 Action2 Action3 Action4
0     r1     a95     a20     NaN     NaN
1     r2     a30     a12     a15     a75
2     r4     a15     a74     NaN     NaN
3     r5     a11     NaN     NaN     NaN

You've used split wrong. https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.Series.str.split.html

Something like this is probably what you want to do, although you will still keep the -rf parts:

split(',', expand=True)

     0       1     2     3
0  a95     a20  None  None
1  a30  a12-rf   a15   a75
2    0    None  None  None
3  a15     a74  None  None
4  a11    None  None  None

Also, your regex is hitting some common beginner problems so it would be a good idea to read up on regex in python. https://docs.python.org/3/howto/regex.html. In particular your [a] is quite meaningless since it is the same as a plain a.

If you know the maximum number of elements you can split to, then you can use a regex and extract which also does what you want. https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.Series.str.extract.html

Something like

extract(r'(a\d\d)(?:-[^,]+)?(?:,(a\d\d)(?:-[^,]+)?)?(?:,(a\d\d)(?:-[^,]+)?)?(?:,(a\d\d)(?:-[^,]+)?)?')

     0    1    2    3
0  a95  a20  NaN  NaN
1  a30  a12  a15  a75
2  NaN  NaN  NaN  NaN
3  a15  a74  NaN  NaN
4  a11  NaN  NaN  NaN

In terms of matching something like a12-rf and only keeping the a12 part, you need a regex like (a\d\d)(?:-[^,]+)? to capture a12 with (a\d\d), and then match but not keep -rf you use the non-capture group ?: modifier, and then match -rf with -[^,]+ where [^,]+ matches 1-or-more non-, characters.

This is how I got (a\d\d)(?:-[^,]+)?. For the other parts elements in your string, you can wrap this pattern in its own non-capture group to handle the being optional and adding a comma to match. (?:,<prev_pattern>)?. And then you copy-paste that ,<pattern> match as many times as you need.

If you do not know how many of these elements you could hit, then you can use extractall and then pivot the MultiIndex it uses. This sounds complicated (and I have to admit that doing it in my head is a bit painful) but is quite simple in practice.

extractall(r'(a\d\d)(?:-[^,]+)?').unstack()[0]

match    0    1    2    3
0      a95  a20  NaN  NaN
1      a30  a12  a15  a75
3      a15  a74  NaN  NaN
4      a11  NaN  NaN  NaN