yeah all the vars are in global scope though, so I'm 99% sure that the invocation of

train(..., warmstart_arg_here)

in the training step is just passing in that warm_start var whose value is

False

. It didn't get overwritten in between the var definition and the train function call I don't think