[FIXED] pandas melt data with many id_vars into on column

Issue

I'd like to melt a DataFrame without using any loops. Suppose I have a DataFrame looking something like this:

df = pd.DataFrame({'var1': [1,2,3,4,5]*100,
                   'var2': [1,2,3,4,5]*100,
                   'col1': ['a','b']*250,
                   'col2': ['c','d']*250,})


     var1  var2 col1 col2
0       1     1    a    c
1       2     2    b    d
2       3     3    a    c
3       4     4    b    d
4       5     5    a    c
..    ...   ...  ...  ...
495     1     1    b    d
496     2     2    a    c
497     3     3    b    d
498     4     4    a    c
499     5     5    b    d

And now I want to melt the data:

df.melt(value_vars=['var1', 'var2'], var_name='var', id_vars=['col1', 'col2'])

    col1 col2   var  value
0      a    c  var1      1
1      b    d  var1      2
2      a    c  var1      3
3      b    d  var1      4
4      a    c  var1      5
..   ...  ...   ...    ...
995    b    d  var2      1
996    a    c  var2      2
997    b    d  var2      3
998    a    c  var2      4
999    b    d  var2      5

Is it possible to melt the data without using any loops to one column with id_vars? So it would end up looking something like this:

    col var value
0   a   var1    1
1   b   var1    2
2   a   var1    3
3   b   var1    4
4   a   var1    5
5   c   var2    1
6   d   var2    2
7   c   var2    3
8   d   var2    4
9   c   var2    5
..  ..  ....   ..

Solution

Use wide_to_long, it create new column by number after var, col columns, so if necessary added var strings:

df1 = (pd.wide_to_long(df.reset_index(), stubnames=['var','col'], i='index', j='new')
         .reset_index(level=1)
         .assign(new = lambda x: 'var' + x['new'].astype(str))
         .reset_index(drop=True)
         )
print (df1) 
      new  var col
0    var1    1   a
1    var1    2   b
2    var1    3   a
3    var1    4   b
4    var1    5   a
..    ...  ...  ..
995  var2    1   d
996  var2    2   c
997  var2    3   d
998  var2    4   c
999  var2    5   d

[1000 rows x 3 columns]

Possible a bit hack solution with melt:

df1 = df.melt(value_vars=['var1', 'var2'], var_name='var', id_vars=['col1', 'col2'])
df2 = df.melt(value_vars=['col1', 'col2'], var_name='var', id_vars=['var1', 'var2'])

df = pd.concat([df1[['var','value']], df2['value'].rename('col')], axis=1)
print (df)
      var  value col
0    var1      1   a
1    var1      2   b
2    var1      3   a
3    var1      4   b
4    var1      5   a
..    ...    ...  ..
995  var2      1   d
996  var2      2   c
997  var2      3   d
998  var2      4   c
999  var2      5   d

[1000 rows x 3 columns]

Answered By - jezrael