How to specify the levels to iterate in a grid search with an ensemble classifier?

I have the following setup but I can’t find a way to pass levels to explore in the Grid search for svm* and mlp*:

steps = [('preprocessing', StandardScaler()),
         ('feature_selection', SelectKBest(mutual_info_classif, k=15)),
         ('clf', VotingClassifier(estimators=[("mlp1", mlp1),
                                              ("mlp2", mlp2),
                                              ("mlp3", mlp3),
                                              ("svm1", svm1),
                                              ("svm2", svm2)
                                             ], voting='soft'))
        ]

model = Pipeline(steps=steps)
params = [{
    'preprocessing': [StandardScaler(), MinMaxScaler(), MaxAbsScaler()],
    'feature_selection__score_func': [f_classif, mutual_info_classif]
}]

grid_search = GridSearchCV(model, params, cv=10, scoring='balanced_accuracy', verbose=1, n_jobs=20, refit=True)

>Solution :

Within VotingClassifier, nested parameters can be specified using the following structure:

clf__<estimator_name>__<hyperparameter_name>

clf refers to the VotingClassifier step in the pipeline.
<estimator_name> refers to the name of the estimator in the VotingClassifier (e.g. mlp1, svm1).
<hyperparameter_name> refers to the parameter of the individual estimator (e.g. hidden_layer_sizes for MLP or C for SVM).

Now you can include hyperparameters mlp1, mlp2, mlp3 (for MLP classifiers) and svm1, svm2 (for SVM classifiers) in the params dictionary. e.g.

params = [{
    # Preprocessing variations
    'preprocessing': [StandardScaler(), MinMaxScaler(), MaxAbsScaler()],
    
    # Feature selection variations
    'feature_selection__score_func': [f_classif, mutual_info_classif],
    
    # Hyperparameters for MLP classifiers
    'clf__mlp1__hidden_layer_sizes': [(50,), (100,), (50, 50)],
    'clf__mlp1__activation': ['relu', 'tanh'],
    'clf__mlp2__hidden_layer_sizes': [(50,), (100,)],
    'clf__mlp2__activation': ['relu', 'tanh'],
    'clf__mlp3__hidden_layer_sizes': [(50, 50), (100, 50)],
    'clf__mlp3__activation': ['relu', 'logistic'],
    
    # Hyperparameters for SVM classifiers
    'clf__svm1__C': [0.01, 0.1, 1, 10],
    'clf__svm1__kernel': ['linear', 'rbf'],
    'clf__svm2__C': [0.1, 1, 10],
    'clf__svm2__kernel': ['rbf', 'sigmoid']
}]