Sorting row values and displaying the column order

Question

I am trying to a data frame which is something like this

user_name   tag1   tag2   tag3   tag4
    user1    .65    .32    .91      0
    user2    .34    .44    .21    .56
    user3    .21      0      0    .19

I need to sort each row on the values of column and get the result columns with highest value and decreasing for each row, also need to remove columns having 0 value for particular user. . output should look like something this.

user_name       0      1      2     3
    user1    tag3   tag1   tag2  
    user2    tag4   tag2   tag1   tag3
    user3    tag1   tag4          

or transpose of this will also work. I need to do this in python2.7. Thank you.

Answer 1

If you replace the 0 values with NaN then you can apply a lambda to mask the index:

In [28]:
df.replace(0,np.NaN, inplace=True)
def func(x):
    val = x.sort_values(ascending=False).index.to_series()
    mask = pd.isnull(x)
    val[mask] = ''
    return val.values
df.ix[:, 'tag1':] = df.ix[:, 'tag1':].apply(lambda x: func(x), axis=1)
df

Out[28]:
  user_name  tag1  tag2  tag3  tag4
0     user1  tag3  tag1  tag2      
1     user2  tag4  tag2  tag1  tag3
2     user3  tag1  tag4            

I use NaN here rather than comparing against 0 because comparing against float scalar values are problematic and may not work:

In [32]:
def func(x):
    val = x.sort_values(ascending=False).index.to_series()
    mask = val == 0
    val[mask] = ''
    return val.values
df.ix[:, 'tag1':] = df.ix[:, 'tag1':].apply(lambda x: func(x), axis=1)
df

Out[32]:
  user_name  tag1  tag2  tag3  tag4
0     user1  tag3  tag1  tag2  tag4
1     user2  tag4  tag2  tag1  tag3
2     user3  tag1  tag4  tag3  tag2

I return a np array using .values because otherwise the Series that would have been returned would align against the original column names, so no ordering would occur

Additionally I have to call to_series on the Index object as you've can't mutate an Index object with this line: val[mask] = ''