Learning Rate scheduler

[boynukaline]

Scheduler (reduce lr at epochs 218, 245, i.e. batches 400k, 450k)

# lf = lambda x: 1 - x / epochs  # linear ramp to zero
# lf = lambda x: 10 ** (-2 * x / epochs)  # exp ramp to lr0 * 1e-2
lf = lambda x: 1 - 10 ** (hyp['lrf'] * (1 - x / epochs))  # inv exp ramp to lr0 * 1e-2
scheduler = optim.lr_scheduler.LambdaLR(optimizer, lr_lambda=lf, last_epoch=start_epoch - 1)

---

Which epochs does the lr reduce other than 218 and 245? Does it stop at 245? Are we able to tune this for a custom dataset? What to do when the loss stagnates?

[glenn-jocher]

See #238 for all LR scheduling concerns.

The original darknet LR scheduler steps at 400k and 450k when training COCO to 500200 batches.