Struggling to integrate sklearn and pandas in simple Kaggle task

Question

I'm trying to use the sklearn_pandas module to extend the work I do in pandas and dip a toe into machine learning but I'm struggling with an error I don't really understand how to fix.

I was working through the following dataset on Kaggle.

It's essentially an unheadered table (1000 rows, 40 features) with floating point values.

import pandas as pdfrom sklearn import neighbors
from sklearn_pandas import DataFrameMapper, cross_val_score
path_train ="../kaggle/scikitlearn/train.csv"
path_labels ="../kaggle/scikitlearn/trainLabels.csv"
path_test = "../kaggle/scikitlearn/test.csv"

train = pd.read_csv(path_train, header=None)
labels = pd.read_csv(path_labels, header=None)
test = pd.read_csv(path_test, header=None)
mapper_train = DataFrameMapper([(list(train.columns),neighbors.KNeighborsClassifier(n_neighbors=3))])
mapper_train

Output:

DataFrameMapper(features=[([0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39], KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',
       n_neighbors=3, p=2, weights='uniform'))])

So far so good. But then I try the fit

mapper_train.fit_transform(train, labels)

Output:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-6-e3897d6db1b5> in <module>()
----> 1 mapper_train.fit_transform(train, labels)

//anaconda/lib/python2.7/site-packages/sklearn/base.pyc in fit_transform(self, X, y,     **fit_params)
    409         else:
    410             # fit method of arity 2 (supervised transformation)
--> 411             return self.fit(X, y, **fit_params).transform(X)
    412 
    413 

//anaconda/lib/python2.7/site-packages/sklearn_pandas/__init__.pyc in fit(self, X, y)
    116         for columns, transformer in self.features:
    117             if transformer is not None:
--> 118                 transformer.fit(self._get_col_subset(X, columns))
    119         return self
    120 

TypeError: fit() takes exactly 3 arguments (2 given)`

What am I doing wrong? While the data in this case is all the same, I'm planning to work up a workflow for mixtures categorical, nominal and floating point features and sklearn_pandas seemed to be a logical fit.

Answer 1

Here is an example of how to get pandas and sklearn to play nice

say you have 2 columns that are both strings and you wish to vectorize - but you have no idea which vectorization params will result in the best downstream performance.

create the vectorizer

to_vect = Pipeline([('vect',CountVectorizer(min_df =1,max_df=.9,ngram_range=(1,2),max_features=1000)),
                    ('tfidf', TfidfTransformer())])

create the DataFrameMapper obj.

full_mapper = DataFrameMapper([
        ('col_name1', to_vect),
        ('col_name2',to_vect)
    ])

this is the full pipeline

full_pipeline  = Pipeline([('mapper',full_mapper),('clf', SGDClassifier(n_iter=15, warm_start=True))])

define the params you want the scan to consider

full_params = {'clf__alpha': [1e-2,1e-3,1e-4],
                   'clf__loss':['modified_huber','hinge'],
                   'clf__penalty':['l2','l1'],
                   'mapper__features':[[('col_name1',deepcopy(to_vect)),
                                        ('col_name2',deepcopy(to_vect))],
                                       [('col_name1',deepcopy(to_vect).set_params(vect__analyzer= 'char_wb')),
                                        ('col_name2',deepcopy(to_vect))]]}

Thats it! - note however that mapper_features are a single item in this dictionary - so use a for loop or itertools.product to generate a FLAT list of all to_vect options you wish to consider - but that is a separate task outside the scope of the question.

Go on to create the optimal classifier or whatever else your pipeline ends with

gs_clf = GridSearchCV(full_pipe, full_params, n_jobs=-1)

Answer 2

I have never used sklearn_pandas, but from reading their source code, it looks like this is a bug on their side. If you look for the function that is throwing the exception, you can notice that they are discarding the y argument (it does not even survive until the docstring), and the inner fit function expects one argument more, which is probably y:

def fit(self, X, y=None):
    '''
    Fit a transformation from the pipeline

    X       the data to fit
    '''
    for columns, transformer in self.features:
        if transformer is not None:
            transformer.fit(self._get_col_subset(X, columns))
    return self

I would recommend that you open an issue in their bug tracker.

UPDATE:

You can test this if you run your code from IPython. To summarize, if you use the %pdb on magic right before you run the problematic call, the exception is captured by the Python debugger, so you can play around a bit and see that calling the fit function with the label values y[0] does work -- see the last line with the pdb> prompt. (The CSV files are downloaded from Kaggle, except for the largest one which is just a part of the real file).

In [1]: import pandas as pd

In [2]: from sklearn import neighbors

In [3]: from sklearn_pandas import DataFrameMapper, cross_val_score

In [4]: path_train ="train.csv"

In [5]: path_labels ="trainLabels.csv"

In [6]: path_test = "test.csv"

In [7]: train = pd.read_csv(path_train, header=None)

In [8]: labels = pd.read_csv(path_labels, header=None)

In [9]: test = pd.read_csv(path_test, header=None)

In [10]: mapper_train = DataFrameMapper([(list(train.columns),neighbors.KNeighborsClassifier(n_neighbors=3))])

In [13]: %pdb on

In [14]: mapper_train.fit_transform(train, labels)
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-14-e3897d6db1b5> in <module>()
----> 1 mapper_train.fit_transform(train, labels)

/opt/local/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/sklearn/base.pyc in fit_transform(self, X, y, **fit_params)
    409         else:
    410             # fit method of arity 2 (supervised transformation)
--> 411             return self.fit(X, y, **fit_params).transform(X)
    412
    413

/opt/local/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/sklearn_pandas/__init__.pyc in fit(self, X, y)
    116         for columns, transformer in self.features:
    117             if transformer is not None:
--> 118                 transformer.fit(self._get_col_subset(X, columns))
    119         return self
    120

TypeError: fit() takes exactly 3 arguments (2 given)
> /opt/local/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/sklearn_pandas/__init__.py(118)fit()
    117             if transformer is not None:
--> 118                 transformer.fit(self._get_col_subset(X, columns))
    119         return self

ipdb> l
    113
    114         X       the data to fit
    115         '''
    116         for columns, transformer in self.features:
    117             if transformer is not None:
--> 118                 transformer.fit(self._get_col_subset(X, columns))
    119         return self
    120
    121
    122     def transform(self, X):
    123         '''
ipdb> transformer.fit(self._get_col_subset(X, columns), y[0])
KNeighborsClassifier(algorithm='auto', leaf_size=30, metric='minkowski',
           n_neighbors=3, p=2, weights='uniform')