How do you evaluate model performance when the 'ground truth' is an expensive physical experiment at BigHat Biosciences?
