release v0.2.4.1

add a hook for on_tng_metrics so that users get access to the grad_no…

README.md

@@ -116,8 +116,8 @@ def validation_end(self, outputs):
     return tqdm_dic
 ```
    
-## TensorboardX    
-Lightning is fully integrated with tensorboardX.   
+## Tensorboard    
+Lightning is fully integrated with tensorboard.   
 
 <p align="center">
   <a href="https://williamfalcon.github.io/pytorch-lightning/">

pytorch_lightning/models/trainer.py

@@ -31,7 +31,8 @@ class Trainer(TrainerIO):
 
     def __init__(self,
                  experiment,
-                 checkpoint_callback, early_stop_callback,
+                 early_stop_callback=None,
+                 checkpoint_callback=None,
                  gradient_clip=0,
                  cluster=None,
                  process_position=0,
@@ -44,12 +45,14 @@ class Trainer(TrainerIO):
                  check_val_every_n_epoch=1,
                  fast_dev_run=False,
                  accumulate_grad_batches=1,
-                 enable_early_stop=True, max_nb_epochs=1000, min_nb_epochs=1,
-                 train_percent_check=1.0, val_percent_check=1.0, test_percent_check=1.0, val_check_interval=0.95,
+                 max_nb_epochs=1000, min_nb_epochs=1,
+                 train_percent_check=1.0, val_percent_check=1.0, test_percent_check=1.0,
+                 val_check_interval=0.95,
                  log_save_interval=100, add_log_row_interval=10,
                  lr_scheduler_milestones=None,
                  use_amp=False,
                  print_nan_grads=False,
+                 print_weights_summary=True,
                  amp_level='O2',
                  nb_sanity_val_steps=5):
 
@@ -57,7 +60,7 @@ class Trainer(TrainerIO):
         self.nb_gpu_nodes = nb_gpu_nodes
         self.gradient_clip = gradient_clip
         self.check_val_every_n_epoch = check_val_every_n_epoch
-        self.enable_early_stop = enable_early_stop
+        self.enable_early_stop = early_stop_callback is not None
         self.track_grad_norm = track_grad_norm
         self.fast_dev_run = fast_dev_run
         self.on_gpu = gpus is not None and torch.cuda.is_available()
@@ -67,8 +70,12 @@ class Trainer(TrainerIO):
         self.cluster = cluster
         self.process_position = process_position
         self.current_gpu_name = current_gpu_name
+        self.print_weights_summary = print_weights_summary
         self.checkpoint_callback = checkpoint_callback
-        self.checkpoint_callback.save_function = self.save_checkpoint
+
+        if self.checkpoint_callback is not None:
+            self.checkpoint_callback.save_function = self.save_checkpoint
+
         self.early_stop = early_stop_callback
         self.model = None
         self.max_nb_epochs = max_nb_epochs
@@ -212,9 +219,6 @@ class Trainer(TrainerIO):
         :param max_batches: Scalar
         :return:
         """
-        if self.proc_rank == 0:
-            print('validating...')
-
         # enable eval mode
         model.zero_grad()
         model.eval()
@@ -273,7 +277,7 @@ class Trainer(TrainerIO):
         self.test_dataloader = model.test_dataloader
         self.val_dataloader = model.val_dataloader
 
-        if self.data_parallel and type(self.tng_dataloader.sampler) is not DistributedSampler:
+        if self.data_parallel and not isinstance(self.tng_dataloader.sampler, DistributedSampler):
             msg = '''
             when using multiple gpus and multiple nodes you must pass a DistributedSampler to DataLoader(sampler).
             
@@ -300,7 +304,7 @@ class Trainer(TrainerIO):
             mp.spawn(self.dp_train, nprocs=len(self.data_parallel_device_ids), args=(model, ))
 
         # treat 1 gpu as a different case to avoid nccl bugs
-        elif len(self.data_parallel_device_ids) == 1:
+        elif self.data_parallel_device_ids is not None and len(self.data_parallel_device_ids) == 1:
             self.single_gpu_train(model)
 
         else:
@@ -443,7 +447,7 @@ class Trainer(TrainerIO):
                 self.lr_schedulers.append(scheduler)
 
         # print model summary
-        if self.proc_rank == 0:
+        if self.proc_rank == 0 and self.print_weights_summary:
             ref_model.summarize()
 
         # give model convenience properties
@@ -538,9 +542,11 @@ class Trainer(TrainerIO):
                     if self.track_grad_norm > 0:
                         model = self.__get_model()
                         grad_norm_dic = model.grad_norm(self.track_grad_norm)
-
                         metrics.update(grad_norm_dic)
 
+                    if self.__is_function_implemented('on_tng_metrics'):
+                        model.on_tng_metrics(metrics)
+
                     # log metrics
                     scalar_metrics = self.__metrics_to_scalars(metrics, blacklist=self.__log_vals_blacklist())
                     if self.proc_rank == 0:
@@ -562,9 +568,9 @@ class Trainer(TrainerIO):
                 model.on_epoch_end()
 
             # early stopping
-            if self.enable_early_stop:
+            met_min_epochs = epoch_nb > self.min_nb_epochs
+            if self.enable_early_stop and met_min_epochs:
                 should_stop = self.early_stop_callback.on_epoch_end(epoch=epoch_nb, logs=self.__tng_tqdm_dic)
-                met_min_epochs = epoch_nb > self.min_nb_epochs
 
                 # stop training
                 stop = should_stop and met_min_epochs
@@ -719,6 +725,6 @@ class Trainer(TrainerIO):
             self.prog_bar.set_postfix(**tqdm_metrics)
 
         # model checkpointing
-        if self.proc_rank == 0:
+        if self.proc_rank == 0 and self.checkpoint_callback:
             print('save callback...')
             self.checkpoint_callback.on_epoch_end(epoch=self.current_epoch, logs=self.__tng_tqdm_dic)

pytorch_lightning/root_module/hooks.py

@@ -19,3 +19,6 @@ class ModelHooks(torch.nn.Module):
     def on_post_performance_check(self):
         pass
 
+    def on_tng_metrics(self, metrics):
+        pass
+

pytorch_lightning/root_module/root_module.py

@@ -78,7 +78,7 @@ class LightningModule(GradInformation, ModelIO, OptimizerConfig, ModelHooks):
         :param logs:
         :return:
         """
-        raise NotImplementedError
+        return logs
 
     def loss(self, *args, **kwargs):
         """

setup.py

@@ -7,7 +7,7 @@ from setuptools import setup, find_packages
 # http://blog.ionelmc.ro/2014/05/25/python-packaging/
 setup(
     name="pytorch-lightning",
-    version='0.2.3',
+    version='0.2.4.1',
     description="The Keras for ML researchers using PyTorch",
     author="William Falcon",
     author_email="waf2107@columbia.edu",
@@ -19,8 +19,7 @@ setup(
     install_requires=[
         "torch>=1.1.0",
         "tqdm",
-        "test-tube>=0.6.6",
-        "tensorflow>=1.14.0"
+        "test-tube>=0.6.7.1",
     ],
     packages=find_packages(),
     long_description=open("README.md", encoding="utf-8").read(),