How to pass a parameter to only one part of a pipeline object in scikit learn?

问题

I need to pass a parameter, sample_weight, to my RandomForestClassifier like so:

X = np.array([[2.0, 2.0, 1.0, 0.0, 1.0, 3.0, 3.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0,
        1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 5.0, 3.0,
        2.0, '0'],
       [15.0, 2.0, 5.0, 5.0, 0.466666666667, 4.0, 3.0, 2.0, 0.0, 0.0, 0.0,
        0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 1.0, 0.0, 0.0,
        7.0, 14.0, 2.0, '0'],
       [3.0, 4.0, 3.0, 1.0, 1.33333333333, 1.0, 1.0, 1.0, 0.0, 0.0, 0.0,
        0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0,
        9.0, 8.0, 2.0, '0'],
       [3.0, 2.0, 3.0, 0.0, 0.666666666667, 2.0, 2.0, 1.0, 0.0, 0.0, 0.0,
        0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0,
        5.0, 3.0, 1.0, '0']], dtype=object)

y = np.array([ 0.,  0.,  1.,  0.])

m = sklearn.ensemble.RandomForestClassifier(
        random_state=0, 
        oob_score=True, 
        n_estimators=100,
        min_samples_leaf=5, 
        max_depth=10)

m.fit(X, y, sample_weight=np.array([3,4,2,3]))

The above code works perfectly fine. Then, I try to do this in a pipeline object like so, using pipeline object instead of only random forest:

m = sklearn.pipeline.Pipeline([
    ('feature_selection', sklearn.feature_selection.SelectKBest(
        score_func=sklearn.feature_selection.f_regression,
        k=25)),
    ('model', sklearn.ensemble.RandomForestClassifier(
        random_state=0, 
        oob_score=True, 
        n_estimators=500,
        min_samples_leaf=5, 
        max_depth=10))])

m.fit(X, y, sample_weight=np.array([3,4,2,3]))

Now this breaks in the fit method with "ValueError: need more than 1 value to unpack".

ValueError                                Traceback (most recent call last)
<ipython-input-212-c4299f5b3008> in <module>()
     25         max_depth=10))])
     26 
---> 27 m.fit(X, y, sample_weights=np.array([3,4,2,3]))

/usr/local/lib/python2.7/dist-packages/sklearn/pipeline.pyc in fit(self, X, y, **fit_params)
    128         data, then fit the transformed data using the final estimator.
    129         """
--> 130         Xt, fit_params = self._pre_transform(X, y, **fit_params)
    131         self.steps[-1][-1].fit(Xt, y, **fit_params)
    132         return self

/usr/local/lib/python2.7/dist-packages/sklearn/pipeline.pyc in _pre_transform(self, X, y, **fit_params)
    113         fit_params_steps = dict((step, {}) for step, _ in self.steps)
    114         for pname, pval in six.iteritems(fit_params):
--> 115             step, param = pname.split('__', 1)
    116             fit_params_steps[step][param] = pval
    117         Xt = X

ValueError: need more than 1 value to unpack

I am using sklearn version 0.14.
I think that the problem is that the F selection step in the pipeline does not take in an argument for sample_weights. how do I pass this parameter to only one step in the pipeline with I run "fit"? Thanks.

回答1:

From the documentation:

The purpose of the pipeline is to assemble several steps that can be cross-validated together while setting different parameters. For this, it enables setting parameters of the various steps using their names and the parameter name separated by a ‘__’, as in the example below.

So you can simply insert model__ in front of whatever fit parameter kwargs you want to pass to your 'model' step:

m.fit(X, y, model__sample_weight=np.array([3,4,2,3]))

回答2:

You can also use the method set_params and prepend the name of the step.

m = sklearn.pipeline.Pipeline([
    ('feature_selection', sklearn.feature_selection.SelectKBest(
        score_func=sklearn.feature_selection.f_regression,
        k=25)),
    ('model', sklearn.ensemble.RandomForestClassifier(
        random_state=0, 
        oob_score=True, 
        n_estimators=500,
        min_samples_leaf=5, 
        max_depth=10))])

m.set_params(model__sample_weight=np.array([3,4,2,3]))

回答3:

Wish I could leave a comment on @rovyko post above instead of a separate answer but I don't have enough stackoverflow reputation yet to leave comments so here it is instead.

You cannot use:

Pipeline.set_params(model__sample_weight=np.array([3,4,2,3])

to set parameters for the RandomForestClassifier.fit() method. Pipeline.set_params() as indicated in the code (here) is only for initialization parameters for individual steps in the Pipeline. RandomForestClassifier has no initialization parameter called sample_weight (see its __init__() method here). sample_weight is actually an input parameter to RandomForestClassifier's fit() method and can therefore only be set by the method presented in the correctly marked answer be @ali_m, which is,

m.fit(X, y, model__sample_weight=np.array([3,4,2,3])).

来源：https://stackoverflow.com/questions/35632634/how-to-pass-a-parameter-to-only-one-part-of-a-pipeline-object-in-scikit-learn

标签

python

pandas

scikit-learn

pipeline