Gradient descent methods, from plain GD to stochastic gradient descent and adaptive learning rates