Generating the shuffled datasets