Apply X_train_sfs = sfs1.transform(X_train) to a different estimator

[Servando1990]

Hi,

I'm trying to reduce a dataset with SFS and then use NestedCV for different algorithms.

My question is: Does it makes sense to apply previously reduced X_train with SFS on different algorithms?

X_train_selected just works well with KNN?

How can I reduce the dataset with SFS and then perform some algorithm selection technique?

sfs1 = SFS(knn, # KNN as the estimator
           k_features=3, 
           forward=True, 
           floating=False, 
           scoring='accuracy',
           cv=5)
sfs1 = sfs1.fit(X_train, y_train)
X_train_selected = sfs1.transform(X_train) 

# Nested CV
outer_cv = StratifiedKFold(n_splits=5, shuffle=True, random_state=1)

for name, gs_est in sorted(gridcvs.items()): # Here I stored different algorithms e.j LR, RF, KNN, DT 
    nested_score = cross_val_score(gs_est, 
                                   X=X_train_selected,  # Transformed data from SFS with KNN estimator
                                   y=y_train, 
                                   cv=outer_cv,
                                   n_jobs=1) 
    print(f'{name:<7} | outer ACC {100*nested_score.mean():.2f}% +/- {100*nested_score.std():.2f}')