Support torch.optim.lr_scheduler.ReduceLROnPlateau (#320)

* feat: add reducelronplateau callback

* feat: use reducelronplateau callback in trainer

* feat: only on unsupported lr schedulers

* feat: last but not the least merge of master

* feat: merge master

* feat: support only on scheduler in reduceLrOnPlateauScheduler

* refactor: code style

* Update pt_callbacks.py

* Update trainer.py

* Update train_loop_mixin.py

* Update trainer.py

* Update train_loop_mixin.py

pytorch_lightning/trainer/train_loop_mixin.py

@@ -150,7 +150,8 @@ When this flag is enabled each batch is split into sequences of size truncated_b
 """
 
 import numpy as np
-import tqdm
+
+from pytorch_lightning.utilities.debugging import MisconfigurationException
 
 try:
     from apex import amp
@@ -213,7 +214,15 @@ class TrainerTrainLoopMixin(object):
             # update LR schedulers
             if self.lr_schedulers is not None:
                 for lr_scheduler in self.lr_schedulers:
-                    lr_scheduler.step(self.current_epoch)
+                    lr_scheduler.step(epoch=self.current_epoch)
+            if self.reduce_lr_on_plateau_scheduler is not None:
+                val_loss = self.callback_metrics.get('val_loss')
+                if val_loss is None:
+                    avail_metrics = ','.join(list(self.callback_metrics.keys()))
+                    m = f'ReduceLROnPlateau conditioned on metric val_loss ' \
+                        f'which is not available. Available metrics are: {avail_metrics}'
+                    raise MisconfigurationException(m)
+                self.reduce_lr_on_plateau_scheduler.step(val_loss, epoch=self.current_epoch)
 
             # early stopping
             met_min_epochs = epoch_nb > self.min_nb_epochs

pytorch_lightning/trainer/trainer.py

@@ -188,6 +188,8 @@ class Trainer(TrainerIOMixin,
         self.early_stop_callback = None
         self.configure_early_stopping(early_stop_callback, logger)
 
+        self.reduce_lr_on_plateau_scheduler = None
+
         # configure checkpoint callback
         self.checkpoint_callback = checkpoint_callback
         self.weights_save_path = weights_save_path
@@ -378,12 +380,20 @@ class Trainer(TrainerIOMixin,
         # two lists
         elif len(optimizers) == 2 and isinstance(optimizers[0], list):
             optimizers, lr_schedulers = optimizers
+            lr_schedulers, self.reduce_lr_on_plateau_scheduler = self.configure_schedulers(lr_schedulers)
             return optimizers, lr_schedulers
 
         # single list or tuple
         elif isinstance(optimizers, list) or isinstance(optimizers, tuple):
             return optimizers, []
 
+    def configure_schedulers(self, schedulers):
+        for i, scheduler in enumerate(schedulers):
+            if isinstance(scheduler, torch.optim.lr_scheduler.ReduceLROnPlateau):
+                reduce_lr_on_plateau_scheduler = schedulers.pop(i)
+                return schedulers, reduce_lr_on_plateau_scheduler
+        return schedulers, None
+
     def run_pretrain_routine(self, model):
         """
         Sanity check a few things before starting actual training