How to reorganize two pairs of columns and choose the order as well?

My input is a dataframe :

import pandas as pd
import numpy as np

df = pd.DataFrame({'node1': ['_abc-1', 'xyz-1', 'abc-1', '-xyz-2', 'xyz-2', 'abc-2'],
 'p1': [1, 10, 3, 1, 2, 6],
 'p2': [9, 2, 11, 4, 5, 3],
 'node2': ['xyz-1', 'abc-1', '-xyz-1', 'def-2', 'def-2', '-xyz-1']})

# print(df)

#     node1  p1  p2   node2
# 0  _abc-1   1   9   xyz-1
# 1   xyz-1  10   2   abc-1
# 2   abc-1   3  11  -xyz-1
# 3  -xyz-2   1   4   def-2
# 4   xyz-2   2   5   def-2
# 5   abc-2   6   3  -xyz-1

I want to keep only the rows that contains abc and xyz in the nodes and then reorganize them to have the same category of nodes in the same column. I want also to be able to say for example : Lets put the xyz nodes in the left part (it means, node1 and p1).
Thanks to you guys in an older post, I was able to build this code but I’m not getting my expected output anymore and I don’t know how to impose an order.

mask1 = (
    (
        df["node1"].str.contains("abc", case=False) &
        df["node2"].str.contains("xyz", case=False)
    )
    |
    (
        df["node2"].str.contains("abc", case=False) &
        df["node1"].str.contains("xyz", case=False)
    )
)

df = df.loc[mask1]

final = df[['node1', 'node2']].apply(np.sort, axis=1).apply(pd.Series)
final.columns = ['node1', 'node2']

mask2 = final.eq(df[['node1', 'node2']]).all(axis=1)

final[['p1', 'p2']] = (
    df[['p1', 'p2']].where(mask2, other=df[['p2', 'p1']].values)
)

final = final[df.columns]

print(final)
#     node1  p1  p2  node2
# 0  _abc-1   1   9  xyz-1
# 1   abc-1   2  10  xyz-1
# 2  -xyz-1  11   3  abc-1
# 5  -xyz-1   3   6  abc-2

I feel like my code can be shorten much more and fixed at the same time.

Do you guys have any ideas to share ? I’m open to any suggestion.

My expected output is this (if xyz is chosen to be at left) :

#     node1  p1  p2  node2
# 0   xyz-1   9   1  _abc-1
# 1   xyz-1  10   2  abc-1
# 2  -xyz-1  11   3  abc-1
# 5  -xyz-1   3   6  abc-2

>Solution :

Following your original logic, with boolean masks and simply inverting the 1/2 in the column names:

m1 = (df["node1"].str.contains("abc", case=False)
     &df["node2"].str.contains("xyz", case=False))

m2 = (df["node2"].str.contains("abc", case=False)
     &df["node1"].str.contains("xyz", case=False))

inverted_cols = ['node2', 'p2', 'p1', 'node1']
# or, programmatically
# d = {'2': '1', '1': '2'}
# inverted_cols = df.columns.str.replace(r'\d+$',
#                                        lambda m: d.get(m.group(0), m.group(0)),
#                                        regex=True)

out = pd.concat([df[m2],
                 df[m1].set_axis(inverted_cols, axis=1)]
               ).sort_index()

Output:

    node1  p1  p2   node2
0   xyz-1   9   1  _abc-1
1   xyz-1  10   2   abc-1
2  -xyz-1  11   3   abc-1
5  -xyz-1   3   6   abc-2