multiple columns alias name is not effect!

[feng-1985]

import pandas as pd
import numpy as np
import sklearn.preprocessing, sklearn.decomposition, \
    sklearn.linear_model, sklearn.pipeline, sklearn.metrics
from sklearn_pandas import DataFrameMapper
from sklearn_pandas import CategoricalImputer

data = pd.DataFrame({'pet':      ['cat', 'dog', 'dog', 'fish', None, 'dog', 'cat', 'fish'],
                      'children': [4., 6, 3, 3, 2, 3, 5, 4],
                      'salary':   [90, 24, 44, 27, 32, 59, 36, 27],
                     'age':[12,24,21,17,18,25,19,15]})

mapper = DataFrameMapper([
     ('pet', [CategoricalImputer(), sklearn.preprocessing.LabelBinarizer()]),
     (['children', 'salary'], sklearn.preprocessing.StandardScaler(), {'alias': 'children_scaled',
                                                                       'alias1':'salary_scaled'})

 ], df_out=True, default=None)
mapper.fit_transform(data.copy())
print(mapper.transformed_names_)

['pet_cat', 'pet_dog', 'pet_fish', 'children_scaled_0', 'children_scaled_1', 'age']

[feng-1985]

`import pandas as pd
import numpy as np
import sklearn.preprocessing, sklearn.decomposition,
sklearn.linear_model, sklearn.pipeline, sklearn.metrics
from sklearn_pandas import DataFrameMapper
from sklearn_pandas import CategoricalImputer

data = pd.DataFrame({'pet': ['cat', 'dog', 'dog', 'fish', None, 'dog', 'cat', 'fish'],
'children': [4., 6, 3, 3, 2, 3, 5, 4],
'salary': [90, 24, 44, 27, 32, 59, 36, 27],
'age':[12,24,21,17,18,25,19,15]})

mapper = DataFrameMapper([
(['children', 'salary'], sklearn.preprocessing.StandardScaler())
], df_out=True, default=None)
mapper.fit_transform(data.copy()) # !!!
print(mapper.transformed_names_)
`

['children_salary_0', 'children_salary_1', 'age_pet_0', 'age_pet_1']

[feng-1985]

I will try to solve this problem!

[dukebody]

Can you describe what do you expect vs. what is the current behavior? I don't quite understand where is the issue.

The input alias1 in othe of the options dict for one of the features in your code doesn't do anything. A parameter with such a name is not expected nor used.

[feng-1985]

@dukebody
The first result I expect it should be:

['pet_cat', 'pet_dog', 'pet_fish', 'children_scaled', 'salary_scaled', 'age']

Yes, the input alias1 is unuse, so I using the list:

mapper = DataFrameMapper([
     ('pet', [CategoricalImputer(), sklearn.preprocessing.LabelBinarizer()]),
     (['children', 'salary'], sklearn.preprocessing.StandardScaler(), {'alias': ['children_scaled','salary_scaled']})
 ], df_out=True, default=None)

But this code will raise error, so I update dataframe_mapper.py code, add some functions, hope it helps.

The second result I expect it should be:

['children', 'salary', 'age', 'pet']

#129 checks failed, i don't know why.

[devforfu]

@bifeng See the CircleCI report (click Details link). Basically, your pull request has a few flake8 errors/warnings.

[feng-1985]

@devforfu finally remove these errors.

[dukebody]

@bifeng I believe you can get the columns you want using the right syntax with the DataFrameMapper, without modifying any code. Try the following:

import pandas as pd
import numpy as np
import sklearn.preprocessing, sklearn.decomposition, \
    sklearn.linear_model, sklearn.pipeline, sklearn.metrics
from sklearn_pandas import DataFrameMapper
from sklearn_pandas import CategoricalImputer

data = pd.DataFrame({'pet':      ['cat', 'dog', 'dog', 'fish', None, 'dog', 'cat', 'fish'],
                      'children': [4., 6, 3, 3, 2, 3, 5, 4],
                      'salary':   [90, 24, 44, 27, 32, 59, 36, 27],
                     'age':[12,24,21,17,18,25,19,15]})

mapper = DataFrameMapper([
     ('pet', [CategoricalImputer(), sklearn.preprocessing.LabelBinarizer()]),
     (['children'],
      sklearn.preprocessing.StandardScaler(), {'alias': 'children_scaled'}),
     (['salary'],
      sklearn.preprocessing.StandardScaler(), {'alias': 'salary_scaled'}),

 ], df_out=True, default=None)
mapper.fit_transform(data.copy())
print(mapper.transformed_names_)

It outputs:

['pet_cat', 'pet_dog', 'pet_fish', 'children_scaled', 'salary_scaled', 'age']

what is what you expect. Right?

The problem in your original code is that if you use the feature definition:

['children', 'salary'], sklearn.preprocessing.StandardScaler()

then it applies the StandardScaler to a numpy array consisting on both columns, producing a single output.

Let me know if the solution I outlined works for you.

[datajanko]

I think he wants to avoid redundancy. Actually, it might be best to allow an aliases variable in gen_features

[dukebody]

I honestly don't think it's too much writing... However to avoid redundancy one can write something like:

features = [('pet', [CategoricalImputer(), sklearn.preprocessing.LabelBinarizer()])]
for column, alias in zip(['children', 'salary'], ['children_scaled', 'salary_scaled']):
    features.append = ([column], sklearn.preprocessing.StandardScaler(), {'alias': alias})

mapper = DataFrameMapper(features)

[datajanko]

In that case, I have the feeling that we should modify gen_features (because it seems like your are essentially providing functionality of that function.

[devforfu]

@dukebody Do you think we should processed with the PR #129, or close it for now? As it was noted, the more straightforward solution would be to modify gen_features so it can accept aliases list.

[dukebody]

@devforfu I don't think #129 is necessary. We can modify gen_features to accept an aliases list perhaps, yes.

[Rendiere]

@dukebody any update on adding list of aliases to gen_features ? +1 from me for this functionality ^_^