Comment by joshstrange

I wonder how expensive it would be to train a model to parse through all the training data and remove anything you didn't want then re-train the model. I almost hope that doesn't work or results in a model that is nowhere near as good as a model trained on the full data set.