Add resuming from specific checkpoint (#516)

* Add resume_from_checkpoint

* Fix variable name

* #515 Remove did_restore

* #515 Simplify code

* #515 Update doc for resume_from_checkpoint

* #515 Add on_gpu

pytorch_lightning/trainer/trainer.py

@@ -83,7 +83,8 @@ class Trainer(TrainerIOMixin,
                  weights_save_path=None,
                  amp_level='O1',
                  nb_sanity_val_steps=5,
-                 truncated_bptt_steps=None):
+                 truncated_bptt_steps=None,
+                 resume_from_checkpoint=None):
         """
 
         :param logger: Logger for experiment tracking
@@ -140,6 +141,7 @@ class Trainer(TrainerIOMixin,
         self.nb_sanity_val_steps = nb_sanity_val_steps
         self.print_nan_grads = print_nan_grads
         self.truncated_bptt_steps = truncated_bptt_steps
+        self.resume_from_checkpoint = resume_from_checkpoint
         self.shown_warnings = set()
 
         self.fast_dev_run = fast_dev_run

pytorch_lightning/trainer/trainer_io.py

@@ -138,8 +138,11 @@ class TrainerIOMixin(object):
             torch.cuda.empty_cache()
 
         if not did_restore_hpc_weights:
-            # restore weights if same exp version
-            self.restore_state_if_checkpoint_exists(model)
+            if self.resume_from_checkpoint is not None:
+                self.restore(self.resume_from_checkpoint, on_gpu=self.on_gpu)
+            else:
+                # restore weights if same exp version
+                self.restore_state_if_checkpoint_exists(model)
 
         # wait for all models to restore weights
         if self.use_ddp or self.use_ddp2: