Most of the real world data is biased with only a few source endpoints (Around Vancouver/BC). Consider what this would mean for the training, validation, and testing. More generalization required? We do not want to overly overfit to the validation/test data, since we are aiming for more general source-destination pairs.
Most of the real world data is biased with only a few source endpoints (Around Vancouver/BC). Consider what this would mean for the training, validation, and testing. More generalization required? We do not want to overly overfit to the validation/test data, since we are aiming for more general source-destination pairs.