Hi everyone, I saw the tutorial in website about Automatic Learning Rate Finder. That is a real cool feature! I tried to use it in my code but got some trouble.
The Error is
pytorch_lightning.utilities.exceptions.MisconfigurationException: No
train_dataloader()method defined. LightningTrainerexpects as minimum atraining_step(),train_dataloader()andconfigure_optimizers()to be defined.
I know what happend, due to I worte the dataloader out of PL-class , I can not use this cool feature.
I want to know if there is a solution that I keep my code-structure(PL-Class without dataloader method) and still could use auto_lr_find?
There is my code in minimal :
model = unet_train(hparams=vars(args)) #unet_train is inherite from pl.LightningModule
train_loader, val_loader = get_loaders( )
trainer=pl.Trainer.from_argparse_args(args,auto_lr_find=True)
lr_finder=trainer.tuner.lr_find(model)
fig=lr_finder.plot(suggest=True)
fig.show()
model.hparams.learning_rate=lr_finder.suggestion()
trainer.fit(model,train_loader,val_loader)