finetune!

			finetune!(learner, nepochs[, base_lr = 0.002; kwargs...])

Behaves like the fastai implementation fastai.Learner.fine_tune .

Keyword arguments

• freezeepochs = 1: Number of epochs to train with the backbone completely frozen.

• grouper = FastAI.defaultgrouper(learner.model): ParamGrouper which assigns groups 1 (backbone) or 2 (head) for every parameter in learner.model. The default expects learner.model to be a Chain(backbone, head).

• backbone_factor = 0.1: Factor by which updates to backbone model are discounted during the second phase of training.

Any additional keyword arguments are passed to fitonecycle!.