The pros of Mini-Batch Gradient Descent(MBGD):
- It's good at a large dataset while BGD is not.
- It's good at online learning. *Online learning is the way which a model incrementally learns from a stream of dataset in real-time while BGD is not.
- It doesn't need the repreparation of a whole dataset if you want to update a model while BGD needs.
- It less gets stuck in local minima or saddle points than BGD.
But the cons is that MBGD outputs a less accurate value than BGD.