utilities¶

AddDanStagesCallback¶

class AddDanStagesCallback(epoch_freq)[source]¶

Bases: sphinx.ext.autodoc.importer._MockObject

Callback to frequently activate new stages (if available)

Parameters:	epoch_freq (int) – number of epochs to wait, before activating the next stage

at_epoch_end(trainer, epoch, **kwargs)[source]¶

Function which activates the next stage if the current epoch is epoch_freq epochs after activating the last one

Parameters:	trainer (`delira.training.PyTorchNetworkTrainer`) – the trainer holding the model epoch (int) – the current epoch
Returns:	a dictionary with all updated values
Return type:	dict

create_optimizers_dan_whole_network(model: dan.model.DeepAlignmentNetwork, optim_cls, max_stages, **optim_params)[source]¶

Creates one optimizer containing all stages’ parameters

Parameters:	model (DeepAlignmentNetwork) – the model, whose parameters should be optimized optim_cls – the actual optimizer class max_stages (int) – the number of maximum stages to optimize
Returns:	dictionary containing the optimizer optimizers
Return type:	dict

create_optimizers_dan_per_stage(model: dan.model.DeepAlignmentNetwork, optim_cls, max_stages, **optim_params)[source]¶

Creates optimizers for differnt DAN stages

Parameters:	model (DeepAlignmentNetwork) – the model, whose parameters should be optimized optim_cls – the actual optimizer class max_stages (int) – the number of maximum stages to optimize
Returns:	dictionary containing all optimizers
Return type:	dict