Learning Rate Finder Clarification

Part 1 (2020)
1

Hey everyone, I’d like to ask a clarification about the learning rate finder, particularly lr_steep from the following bit of code:

learn = cnn_learner(dls, resnet34, metrics=error_rate)
lr_min,lr_steep = learn.lr_find()

Is lr_steep supposed to be the learning rate at which the loss is decreasing the most?

1 Like
2

Yes that is correct. The docs include more info: