You can see how this works for a random dataset opposite. We start with a bad setup and converge to a good one over time. Things won't work the same each time, and there are cases in which this sequence of steps does not converge well at all.

You can see how this works for a random dataset opposite. We start with a bad setup and converge to a good one over time. Things won't work the same each time, and there are cases in which this sequence of steps does not converge well at all.