Interleave training data from different labels prior to training.

[john-philipp]

Avoid overfitting by interleaving data alternatingly prior to training on it.