High level visual guide to 1cycle policy

Hey! This is a quick visual guide to what the learning rates look like while training our model for 1 or more epochs with the fit_one_cycle method. This also shows what some of the function function parameters do : div, div_final, pct_start.

Some of the questions it addresses:

  1. From what value does the lr increase?
  2. When does it reverse it’s directions? (what is the stepsize?)
  3. Does the learning rate go up and down every epoch?

Hope this helps!