Multiple Transforms for Multiple Columns

[adiv5]

Hello,

I wanted to know is there any way to do multiple transforms on multiple columns , treating each one seperately.

I was able to implement it using Sklearn's ColumnTransformer as follows:

ct = ColumnTransformer(
    [(
        'numeric',
        Pipeline([
            ('handle_na',NAHandler(is_train=True,nan_cols=[])),
            ('standardize',StandardScaler()),
            ('PCA',PCA(n_components=4))
        ]),
            ['col1','col2','col3','col4']
    
    ),
  )],
  remainder='passthrough'
)

However SKlearn pandas' documentation doesnt point me to something like this.
I can see there are 2 sections --- one for single column , multiple transforms and other for multiple colums, single transform.

I couldnt see multiple cols multiple transforms

for now i am able to do what i intend by writing transforms for each and every column seperately . i.e

mapper = DataFrameMapper(
        [
                (
                        ['col1'],
                        [NAHandler(is_train=True,nan_cols=[]),StandardScaler(),PCA(n_components=4)]
                ),
                (
                        ['col2'],
                        [NAHandler(is_train=True,nan_cols=[]),StandardScaler(),PCA(n_components=4)]
                )

               ........
        ],
        input_df=True,
        df_out=True,
        default=None
        )

But what i was actually looking for is ColumnTransformer- like usage .

something like this :

mapper = DataFrameMapper(
        [
                (
                        ['col1','col2','col3','col4'],
                        [NAHandler(is_train=True,nan_cols=[]),StandardScaler(),PCA(n_components=4)]
                )
        ],
        input_df=True,
        df_out=True,
        default=None
        )

will such functionality be supported in upcoming builds? Can be very helpful!

[hu-minghao]

你好，已收到，谢谢。