Struggling to integrate sklearn and pandas in simple Kaggle task
-
16-10-2019 - |
문제
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.
해결책
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)
다른 팁
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')