How to use GridSearchCV for comparing multiple models along with pipeline and hyper-parameter tuning in python

Question

I am using two estimators, Randomforest and SVM

random_forest_pipeline=Pipeline([   
    ('vectorizer',CountVectorizer(stop_words='english')),
    ('random_forest',RandomForestClassifier())
])
svm_pipeline=Pipeline([
    ('vectorizer',CountVectorizer(stop_words='english')),
    ('svm',LinearSVC())
])

I want to first vectorize the data and then use the estimator, I was going through this online tutorial . then I use the hyper parameter as follows

parameters=[
    {
        'vectorizer__max_features':[500,1000,1500],
        'random_forest__min_samples_split':[50,100,250,500]
    },
    {
        'vectorizer__max_features':[500,1000,1500],
        'svm__C':[1,3,5]
    }
]

and passed to the GridSearchCV

pipelines=[random_forest_pipeline,svm_pipeline]
grid_search=GridSearchCV(pipelines,param_grid=parameters,cv=3,n_jobs=-1)
grid_search.fit(x_train,y_train)

but, when I run the code I get an error

TypeError: estimator should be an estimator implementing 'fit' method

Don't know why am I getting this error

Answer 1

It is quite possible to do it in a single Pipeline/GridSearchCV, based on an example here.

You just have to explicitly mention the scoring method for the pipeline since we are not declaring the final estimator initially.

Example:

from sklearn.feature_extraction.text import CountVectorizer
from sklearn.pipeline import Pipeline
from sklearn.ensemble import RandomForestClassifier
from sklearn.svm import LinearSVC


my_pipeline = Pipeline([
    ('vectorizer', CountVectorizer(stop_words='english')),
    ('clf', 'passthrough')
])


parameters = [
    {
        'vectorizer__max_features': [500, 1000],
        'clf':[RandomForestClassifier()],
        'clf__min_samples_split':[50, 100,]
    },
    {
        'vectorizer__max_features': [500, 1000],
        'clf':[LinearSVC()],
        'clf__C':[1, 3]
    }
]

grid_search = GridSearchCV(my_pipeline, param_grid=parameters, cv=3, n_jobs=-1, scoring='accuracy')
grid_search.fit(X, y)

grid_search.best_params_

> # {'clf': RandomForestClassifier(bootstrap=True, ccp_alpha=0.0, class_weight=None,
#                         criterion='gini', max_depth=None, max_features='auto',
#                         max_leaf_nodes=None, max_samples=None,
#                         min_impurity_decrease=0.0, min_impurity_split=None,
#                         min_samples_leaf=1, min_samples_split=100,
#                         min_weight_fraction_leaf=0.0, n_estimators=100,
#                         n_jobs=None, oob_score=False, random_state=None,
#                         verbose=0, warm_start=False),
#  'clf__min_samples_split': 100,
#  'vectorizer__max_features': 1000}




pd.DataFrame(grid_search.cv_results_)[['param_vectorizer__max_features',
                                       'param_clf__min_samples_split',
                                       'param_clf__C','mean_test_score',
                                       'rank_test_score']]

Answer 2

The problem is the pipelines=[random_forest_pipeline,svm_pipeline] that is a list not having the fit method.

Even if you could make it work this way, at some point the 'random_forest__min_samples_split':[50,100,250,500] would be passed in the svm_pipeline and this would raise an error.

ValueError: Invalid parameter svm for estimator Pipeline

You cannot mix this way 2 pipelines because at some point you request the svm_pipeline to be evaluated using the values of random_forest__min_samples_split and this is INVALID.

---

Solution: Fit a GridSearch object for the Random forest model and another GridSearch object for the SVC model

pipelines=[random_forest_pipeline,svm_pipeline]

grid_search_1=GridSearchCV(pipelines[0],param_grid=parameters[0],cv=3,n_jobs=-1)
grid_search_1.fit(X,y)

grid_search_2=GridSearchCV(pipelines[1],param_grid=parameters[1],cv=3,n_jobs=-1)
grid_search_2.fit(X,y)

---

Full code:

random_forest_pipeline=Pipeline([   
    ('vectorizer',CountVectorizer(stop_words='english')),
    ('random_forest',RandomForestClassifier())
])
svm_pipeline=Pipeline([
    ('vectorizer',CountVectorizer(stop_words='english')),
    ('svm',LinearSVC())
])

parameters=[
    {
        'vectorizer__max_features':[500,1000,1500],
        'random_forest__min_samples_split':[50,100,250,500]
    },
    {
        'vectorizer__max_features':[500,1000,1500],
        'svm__C':[1,3,5]
    }
]

pipelines=[random_forest_pipeline,svm_pipeline]

# gridsearch only for the Random Forest model
grid_search_1 =GridSearchCV(pipelines[0],param_grid=parameters[0],cv=3,n_jobs=-1)
grid_search_1.fit(X,y)

# gridsearch only for the SVC model
grid_search_2 =GridSearchCV(pipelines[1],param_grid=parameters[1],cv=3,n_jobs=-1)
grid_search_2.fit(X,y)

EDIT

If you explicitly define the models into the param_grid list then it is possible based on the documentation.

Link: https://scikit-learn.org/stable/auto_examples/compose/plot_compare_reduction.html?highlight=pipeline%20gridsearch

Code from doc:

import numpy as np
import matplotlib.pyplot as plt
from sklearn.datasets import load_digits
from sklearn.model_selection import GridSearchCV
from sklearn.pipeline import Pipeline
from sklearn.svm import LinearSVC
from sklearn.decomposition import PCA, NMF
from sklearn.feature_selection import SelectKBest, chi2

print(__doc__)

pipe = Pipeline([
    # the reduce_dim stage is populated by the param_grid
    ('reduce_dim', 'passthrough'),
    ('classify', LinearSVC(dual=False, max_iter=10000))
])

N_FEATURES_OPTIONS = [2, 4, 8]
C_OPTIONS = [1, 10, 100, 1000]
param_grid = [
    {
        'reduce_dim': [PCA(iterated_power=7), NMF()],
        'reduce_dim__n_components': N_FEATURES_OPTIONS,
        'classify__C': C_OPTIONS
    },
    {
        'reduce_dim': [SelectKBest(chi2)],
        'reduce_dim__k': N_FEATURES_OPTIONS,
        'classify__C': C_OPTIONS
    },
]
reducer_labels = ['PCA', 'NMF', 'KBest(chi2)']

grid = GridSearchCV(pipe, n_jobs=1, param_grid=param_grid)
X, y = load_digits(return_X_y=True)
grid.fit(X, y)