In preprocessing training data for deep learning model, why do we need to shuffle the data?

Group of answer choices

To prepare for stateful Deep Neural Network.

To make it easier to create batches.

To generate more training set.

To help gradient descent works better.