How to use Bootstrap when refitting final stage?

Hi all,

So basically, roughly I have:

cate = DMLIV(model_Y_X(), model_T_X(), model_T_XZ(),
             dmliv_model_effect(), dmliv_featurizer(),
             n_splits= 20)

##### defining some variable in here
cate.fit(Y, T[:,0], XW, Z, store_final=True)

##### choosing the preferred subset of XW to refit the final in here
cate.refit_final(ph_dmliv_model_effect(), dmliv_featurizer())

So first of all. Should I use BootstrapEstimator in both or just when refitting the final stage as I am guessing (since the final estimates are on the refitted final stage). Secondly, when doing:

BootstrapEstimator(cate, n_bootstrap_samples=100).refit_final(ph_dmliv_model_effect(), dmliv_featurizer())

the following error appears: "unsupported operand type(s) for +: 'DMLIV' and 'DMLIV'" after having finished the replicates.

What's going wrong here? Seems like I have to call an attribute of the class DMLIV like fit or refit, but I am not really sure...
Moreover, is the refitting to be done with the training set? That's because I am specifying fold0, fold1, fold00, folld11 of issue #94 in final model, but I don't know how Bootstrap acts. Moreover, by doing:

boot_est=BootstrapEstimator(est,n_bootstrap_samples=100)
te_pred_interval = boot_est.effect_interval(X_test, lower=1, upper=99)
te_pred_interval

the intervals turn out to be equal to the estimates and equal among each other.

Thank you again!

Start by reviewing the DMLIV and BootstrapEstimator usage shown in the issue, including fit, refit_final, effect_interval, and the reference to issue #94. Clarify whether bootstrapping applies during fitting or final-stage refitting, how the reported error arises, how folds and training data are handled, and why the example intervals match the estimates; document the expected usage and results.