Column name corresponding to largest value in pandas DataFrame

Question

for below dataframe data:

x  y  a b c 
2  6 12 1 2
1  2  4 6 8

I want result in new column(i.e d) that returns name of column with max value only among a,b,c.

cols
a
c

I'm trying to find maximum values from three columns and return column name.But instead of selecting all the rows of dataset,I want to select rows of only these three columns.I'm using the following code:

def returncolname(row, colnames):
    return colnames[np.argmax(row.values)]
data['colmax'] = data.apply(lambda x: returncolname(x, data.columns), axis=1)

Answer 1

The fastest solution I can think of is DataFrame.dot:

df.eq(df.max(1), axis=0).dot(df.columns)

Details
First, compute the maximum per row:

df.max(1)
0    12
1     8
dtype: int64

Next, find the positions these values come from:

df.eq(df.max(1), axis=0)     
       x      y      a      b      c
0  False  False   True  False  False
1  False  False  False  False   True

I use eq to make sure the comparison is broadcasted correctly across columns.

Next, compute the dot product with the column list:

df.eq(df.max(1), axis=0).dot(df.columns)
0    a
1    c
dtype: object

---

If the max is not unique, use

df.eq(df.max(1), axis=0).dot(df.columns + ',').str.rstrip(',')

To get a comma separated list of columns. For example,

Change a couple values:

df.at[0, 'c'] = 12
df.at[1, 'y'] = 8

Everything is the same, but notice I append a comma to every column:

df.columns + ','
Index(['x,', 'y,', 'a,', 'b,', 'c,'], dtype='object')

df.eq(df.max(1), axis=0).dot(df.columns + ',')
0    a,c,
1    y,c,
dtype: object

From this, strip any trailing commas:

df.eq(df.max(1), axis=0).dot(df.columns + ',').str.rstrip(',') 
0    a,c
1    y,c
dtype: object

Answer 2

If you don't mind returning the first occurence of the maximum (or if maximum will always be unique), use idxmax

df[['a', 'b', 'c']].idxmax(1)

0    a
1    c
dtype: object

If you need all occurences of a maximum, use coldspeed's answer