ScikitLearn GridSearchCV and pipeline using different methods

Question

I am trying to evaluate these Machine Learning methods to the same data using GridSearchCV and pipeline, when I vary the parameters in the same method it works, but when I put multiple Methods it gives an error

pipe_steps = [
    ('scaler', StandardScaler()), 
    ('logistic', LogisticRegression()),
    ('SVM',SVC()),
    ('KNN',KNeighborsClassifier())]
check_params={
    'logistic__C':[1,1e5],
    'SVM__C':[1,1e5],
    'KNN__n_neighbors':[3,5],
    'KNN__metric':['euclidean','manhattan']
    }
pipeline = Pipeline(pipe_steps)
GridS = GridSearchCV(pipeline, param_grid=check_params)
GridS.fit(X, y)
print('Score %3.2f' %GridS.score(X, y))
print('Best Fit')
print(GridS.best_params_)

gives the error message on pipeline line below

TypeError                                 Traceback (most recent call last)
<ipython-input-139-75960299bc1c> in <module>
     13     }
     14 
---> 15 pipeline = Pipeline(pipe_steps)
     16 
     17 BCX_Grid = GridSearchCV(pipeline, param_grid=check_params)

C:\ProgramData\Anaconda3\lib\site-packages\sklearn\pipeline.py in __init__(self, steps, memory, verbose)
    133     def __init__(self, steps, memory=None, verbose=False):
    134         self.steps = steps
--> 135         self._validate_steps()
    136         self.memory = memory
    137         self.verbose = verbose

C:\ProgramData\Anaconda3\lib\site-packages\sklearn\pipeline.py in _validate_steps(self)
    183                                 "transformers and implement fit and transform "
    184                                 "or be the string 'passthrough' "
--> 185                                 "'%s' (type %s) doesn't" % (t, type(t)))
    186 
    187         # We allow last estimator to be None as an identity transformation

TypeError: All intermediate steps should be transformers and implement fit and transform or be the string 'passthrough' 'LogisticRegression(C=1.0, class_weight=None, dual=False, fit_intercept=True,
                   intercept_scaling=1, l1_ratio=None, max_iter=100,
                   multi_class='warn', n_jobs=None, penalty='l2',
                   random_state=None, solver='warn', tol=0.0001, verbose=0,
                   warm_start=False)' (type <class 'sklearn.linear_model.logistic.LogisticRegression'>) doesn't

Thanks

Do you wish to grid search using multiple classifiers? Like at the same time grid search with SVC, KNN and LogisticRegression? — Bernardo Duarte, May 16 '20 at 17:08

score 1 · Accepted Answer · answered May 16 '20 at 17:23

You need to split the pipeline into multiple pipelines, for that I have a solution that requires a list of grid params that determines each step of the pipeline.

pipeline = Pipeline([
    ('transformer', StandardScaler(),),
    ('model', 'passthrough',),
])

params = [
    {
        'model': (LogisticRegression(),),
        'model__C': (1, 1e5,),
    },
    {
        'model': (SVC(),),
        'model__C': (1, 1e5,),
    },
    {
        'model': (KNeighborsClassifier(),),
        'model__n_neighbors': (3, 5,),
        'model__metric': ('euclidean', 'manhattan',),
    }
]

grid_Search = GridSearchCV(pipeline, params)

With this strategy you can define the steps of the pipeline dynamically.

Thanks Bernardo!!! It worked, I had just to define as the program asked: scoring = 'accuracy' grid_Search = GridSearchCV(pipeline, params, scoring = 'accuracy') Very nice!!! — Loup70, May 16 '20 at 17:59

score 0 · Answer 2 · answered May 16 '20 at 16:31

0

Your problem is not the hyperparameters as they are defined correctly. The problem is that all intermediate step should be transformers, as the error indicates. in your pipeline SVM is not a transformer.

see this post

answered May 16 '20 at 16:31

Alireza

656
1
6
20

So, how may I implement it ? – Loup70 May 16 '20 at 16:53
1

technically speaking, whatever class that is, it should have a `.transform()` method *and trivial `.fit_transform()` as a chain of fit and transform methods. Therefore if you inherit a class from SVC and write a transform method for it, that works. then in that you return X that is required by the next step: your KNN. note that Y is given to the next step of the transformer as it is and without change – Alireza May 16 '20 at 17:51
1

Or you can start writing your own transformer (standard way) and inside that, call the SVC. take a loot at this: https://scikit-learn.org/stable/modules/generated/sklearn.base.TransformerMixin.html – Alireza May 16 '20 at 17:53

ScikitLearn GridSearchCV and pipeline using different methods

2 Answers2