It would be very useful if the TunedThresholdClassifierCV() function allowed threshold optimization through the use of an independent validation dataset (separate and different from both the training dataset and the test dataset). I make this suggestion because using the training data initially to optimize hyperparameters (even using stratified cross-validation) and, subsequently, reusing the same training data to optimize the discrimination threshold (also using stratified cross-validation) can generate overfitting and poor results on test data.