One cycle policy

Hi,

I am using OneCycleLR scheduler with a pct_start of 1.0 (i.e. go from a small learning rate up to a maximum learning rate and then stop), but I get an error at the last training iteration:
`computed_lr = self.anneal_func(group['max_lr'], group['min_lr'], down_step_num / self.step_size_down)`
`ZeroDivisionError: float division by zero`

On the other hand, if I use a pct_start of 0.0 (i.e. go from a large learning rate down to a minimum learning rate), the scheduler works as expected. Is this a missed edge case, or am I using the scheduler improperly?

Thank you!

How did you define `final_div_factor`, which should be used for the minimal learning rate?
Could this value be too high and you might thus run into a zero division?