Pytorch Lightning module
Hippynn incldues support for distributed training using pytorch-lightning.
This can be accessed using the hippynn.experiment.HippynnLightningModule
class.
The class has two class-methods for creating the lightning module using the same
types of arguments that would be used for an ordinary hippynn experiment.
These are from_experiment_setup()
and from_train_setup()
.
Alternatively, you may construct and supply the arguments for the module yourself.
Finally, in additional to the usual pytorch lightning arguments,
the hippynn lightning module saves an additional file, experiment_structure.pt,
which needs to be provided as an argument to the
load_from_checkpoint()
constructor.