根据条件重命名 Pandas 数据框的多列

Question

I have a df in which I need to rename 40 column names to empty string. this can be achieved by using .rename() , but I need to provide all the column names in dict, which needs to be renamed. I was searching for some better way to rename columns by some pattern matching. wherever it finds NULL/UNNAMED in column name, replace that with empty string.

df1: original df (In actual df, i have around 20 columns as NULL1-NULL20 and 20 columns as UNNAMED1-UNNAMED20)

    NULL1   NULL2   C1  C2  UNNAMED1    UNNAMED2
0   1   11  21  31  41  51
1   2   22  22  32  42  52
2   3   33  23  33  43  53
3   4   44  24  34  44  54

desired output df:

            C1  C2      
0   1   11  21  31  41  51
1   2   22  22  32  42  52
2   3   33  23  33  43  53
3   4   44  24  34  44  54

This can be achieved by

df.rename(columns={'NULL1':'', 'NULL2':'', 'UNNAMED1':'', 'UNNAMED2':''}, inplace=True)

But I dont want to create the long dictionary of 40 elements

Answer 1

Is it possible, but be carefull - then if need select one empty column get all empty columns, because duplicated columns names:

print (df[''])

0  1  11  41  51
1  2  22  42  52
2  3  33  43  53
3  4  44  44  54

Use startswith for get all columns by tuples in list comprehension:

df.columns = ['' if c.startswith(('NULL','UNNAMED')) else c for c in df.columns]

Your solution should be changed:

d = dict.fromkeys(df.columns[df.columns.str.startswith(('NULL','UNNAMED'))], '')
print (d)
{'NULL1': '', 'NULL2': '', 'UNNAMED1': '', 'UNNAMED2': ''}
df = df.rename(columns=d)

Answer 2

If you want to stick with rename :

def renaming_fun(x):
    if "NULL" in x or "UNNAMED" in x:
        return "" # or None
    return x

df = df.rename(columns=renaming_fun)

It can be handy if the mapping function gets more complex. Otherwise, list comprehensions will do:

df.columns = [renaming_fun(col) for col in cols]

Another possibility:

df.columns = map(renaming_fun, df.columns)

But as it was already mentioned, renaming with empty strings is not something you would usually do.

Answer 3

If you have few columns to retain their name. Use list-comprehension as below:

df.columns = [col if col in ('C1','C2') else "" for col in df.columns]

Answer 4

df.columns = [col if “NULL” not in col else “” for col in df.columns]

这应该有效，因为您可以通过将列表分配给数据框列变量来更改列名称。

Answer 5

You can use dict comprehension inside df.rename():

idx_filter = np.asarray([i for i, col in enumerate(df.columns) if SOME_STRING_CONDITION in col])
df.rename(columns={col: '' for col in df.columns[idx_filter]}, inplace=True)

In your case, it sounds like SOME_STRING_CONDITION would be 'NULL' or 'UNNAMED'.

I figured this out while looking for help on a thread for a more generic column renaming issue ( Renaming columns in pandas ) for a problem of my own. I didn't have enough reputation to add my solution as an answer or comment (I'm new-ish on stackoverflow), so I am posting it here!

This solution is also helpful if you need to keep part of the string that you were filtering for. For example, if you wanted to replace the "C" columns with "col_":

idx_filter = np.asarray([i for i, col in enumerate(df.columns) if 'C' in col])
df.rename(columns={col: col.replace('C', 'col_') for col in df.columns[idx_filter]}, inplace=True)