tuning_parameters_MLP.txt

{
# --- Feature selection (SelectKBest) ---
# used only to modify selected SKB behavior
'feature_selection__k': [25, 50], # number of top features to select based on univariate scores
'feature_selection__score_func': [mutual_info_classif], # scoring function estimating mutual information for classification tasks

# --- Feature selection (SelectFromModel with Lasso Logistic Regression) ---
# used only to modify laSFM behavior
#'feature_selection__threshold': [-float('inf')], # keep all non-zero weighted features
#'feature_selection__max_features': [25, 50], # number of retained features
#'feature_selection__estimator__penalty': ['l1'], # L1 regularization (lasso)
#'feature_selection__estimator__solver': ['liblinear'], # required for L1 penalty
#'feature_selection__estimator__C': [0.1, 1.0], # inverse regularization strength
#'feature_selection__estimator__max_iter': [300], # iterations for convergence

# --- Feature selection (SelectFromModel with ElasticNet Logistic Regression) ---
# used only to modify enSFM behavior
#'feature_selection__threshold': [-float('inf')], # keep all non-zero weighted features
#'feature_selection__max_features': [25, 50], # number of retained features
#'feature_selection__estimator__penalty': ['elasticnet'], # elasticnet regularization
#'feature_selection__estimator__solver': ['saga'], # required for L1/L2 penalties
#'feature_selection__estimator__l1_ratio': [0.5], # mix of L1/L2 regularization
#'feature_selection__estimator__C': [1.0], # inverse regularization strength

# --- Feature selection (SelectFromModel with Ridge-regularized Logistic Regression) ---
# used only to modify riSFM behavior
#'feature_selection__threshold': [-float('inf')], # keep all features (no threshold cutoff)
#'feature_selection__max_features': [25, 50], # number of retained features
#'feature_selection__estimator__penalty': ['l2'], # L2 regularization (ridge)
#'feature_selection__estimator__solver': ['lbfgs'], # stable solver commonly used for L2-regularized logistic regression
#'feature_selection__estimator__C': [0.5, 1.0], # inverse regularization strength

# --- Feature selection (SelectFromModel with RandomForestClassifier) ---
# used only to modify rfSFM behavior
#'feature_selection__threshold': [-float('inf')], # keep all non-zero importance features
#'feature_selection__max_features': [25, 50], # limit the number of features kept based on feature importance
#'feature_selection__estimator__n_estimators': [200], # number of trees
#'feature_selection__estimator__max_depth': [5, 10], # tree depth control

# --- Model tuning (MLPClassifier) ---
# used only to modify MLP behavior
'model__hidden_layer_sizes': [(50,), (100,)], # hidden layers
'model__alpha': [0.001], # regularization
'model__max_iter': [500] # iterations
}