Looks pretty good. Any chance you could show that first graph overlayed with a scatter plot of the training samples you used?
