问题
I'd like to find the highest values in each row and return the column header for the value in python. For example, I'd like to find the top two in each row:
df =
A B C D
5 9 8 2
4 1 2 3
I'd like my for my output to look like this:
df =
B C
A D
回答1:
You can use a dictionary comprehension to generate the largest_n
values in each row of the dataframe. I transposed the dataframe and then applied nlargest
to each of the columns. I used .index.tolist()
to extract the desired top_n
columns. Finally, I transposed this result to get the dataframe back into the desired shape.
top_n = 2
>>> pd.DataFrame({n: df.T[col].nlargest(top_n).index.tolist()
for n, col in enumerate(df.T)}).T
0 1
0 B C
1 A D
回答2:
I decided to go with an alternative way: Apply the pd.Series.nlargest()
function to each row.
Path to Solution
>>> df.apply(pd.Series.nlargest, axis=1, n=2)
A B C D
0 NaN 9.0 8.0 NaN
1 4.0 NaN NaN 3.0
This gives us the highest values for each row, but keeps the original columns, resulting in ugly NaN values where a column is not everywhere part of the top n values. Actually, we want to receive the index of the nlargest()
result.
>>> df.apply(lambda s, n: s.nlargest(n).index, axis=1, n=2)
0 Index(['B', 'C'], dtype='object')
1 Index(['A', 'D'], dtype='object')
dtype: object
Almost there. Only thing left is to convert the Index objects into Series.
Solution
df.apply(lambda s, n: pd.Series(s.nlargest(n).index), axis=1, n=2)
0 1
0 B C
1 A D
Note that I'm not using the Index.to_series()
function since I do not want to preserve the original index.
来源:https://stackoverflow.com/questions/34518634/finding-highest-values-in-each-row-in-a-data-frame-for-python