Learning rate

Applies to: python, general

A learning rate controls how large each training update is. Too small is slow; too large can diverge.

w -= 0.01 * grad_w

See also: gradient-descent