All of the labels seem to be less than 300, is this an intentional charaistic of the data?

[pennswood]

This seems to contrast your baseline algorithm, which makes predictions up to 1400+ (of course, it’s called a baseline algorithm for a reason, but still).