I'm looking at using sklearn's Gradient Boosting Classifier (GBC) to predict the sign of stock returns. My question is regarding the parameter "validation_fraction" used for early stopping. Since I'm working with time series financial data, any bootstrapping/resampling etc where later observations are used as a training set to predict past observations would be problematic. For example when I use the RandomForestClassifier model I have to set bootstrap to False.
My question is, with this parameter, does the GBC use any resample and/or bootstrapping of any kind ? If so I would have to find a way to modify the source code to ensure the integrity of the training/validation set split.