Changing multiple column values to binary values

问题

I've asked this question before but the answer I got didn't quite work out as I thought it had, so that here I am.

Previous question: Defining a function for changing column values and creating new datasets

I am trying to define a function where it will take a dataframe and change values in a column to create multiple new dataframes.

As an example, from df1 looking like:

  df1:

  class    colB    colC
0   1      1b      1c
1   2      2b      2c
2   3      3b      3c
3   1      4b      4c
4   2      5b      5c

I am trying to create multiple binary classes to implement one-vs-all classification. So the function would create...

df2:
  class    colB    colC
0   1      1b      1c
1   -1      2b      2c
2   -1      3b      3c
3   1      4b      4c
4   -1      5b      5c

df3:
  class    colB    colC
0   -1      1b      1c
1   1      2b      2c
2   -1      3b      3c
3   -1      4b      4c
4   1      5b      5c

df4:
  class    colB    colC
0   -1      1b      1c
1   -1      2b      2c
2    1      3b      3c
3   -1      4b      4c
4   -1      5b      5c

and so on. All the unique values are an incremental value ranging from 1 to 120.

The problem with the previous answer give (np.identity) was that it created dataframes taking every single value as either 1 or -1 instead of categorizing identical values as the same class accordingly.

Thanks

回答1:

A similar idea using np.where and unique (again renaming your class column so it doesn't override a builtin name):

dfs = [
    df1.assign(class_=np.where(df1['class_'].eq(i), 1, -1)) for i in df1['class_'].unique()
]

for d in dfs:
    print(d, end='\n\n')

   class_ colB colC
0       1   1b   1c
1      -1   2b   2c
2      -1   3b   3c
3       1   4b   4c
4      -1   5b   5c

   class_ colB colC
0      -1   1b   1c
1       1   2b   2c
2      -1   3b   3c
3      -1   4b   4c
4       1   5b   5c

   class_ colB colC
0      -1   1b   1c
1      -1   2b   2c
2       1   3b   3c
3      -1   4b   4c
4      -1   5b   5c

回答2:

In similar vein to @user3483203, but using mask and fillna:

[df.assign(**{'class' : df['class'].mask(df['class'].ne(cls)).fillna(-1)}) 
     for cls in df['class'].unique()
]

[   class colB colC
 0    1.0   1b   1c
 1   -1.0   2b   2c
 2   -1.0   3b   3c
 3    1.0   4b   4c
 4   -1.0   5b   5c,    class colB colC
 0   -1.0   1b   1c
 1    2.0   2b   2c
 2   -1.0   3b   3c
 3   -1.0   4b   4c
 4    2.0   5b   5c,    class colB colC
 0   -1.0   1b   1c
 1   -1.0   2b   2c
 2    3.0   3b   3c
 3   -1.0   4b   4c
 4   -1.0   5b   5c]

来源：https://stackoverflow.com/questions/51914247/changing-multiple-column-values-to-binary-values

标签

python

pandas

function

dataframe

classification