[FIXED] IsADirectoryError when loading my pytorch model with load_from_checkpoint

Issue

Could someone please explain to me why this function:

def train_graph_classifier(model_name, **model_kwargs):
  pl.seed_everything(42)

  # Create a PyTorch Lightning trainer with the generation callback
  root_dir = os.path.join('/home/predictor2', "GraphLevel" + model_name)
  os.makedirs(root_dir, exist_ok=True)
  trainer = pl.Trainer(default_root_dir=root_dir,
                     callbacks=[ModelCheckpoint(save_weights_only=True, mode="max", monitor="val_acc")],
                     gpus=1 if str(device).startswith("cuda") else 0,
                     max_epochs=500,
                     progress_bar_refresh_rate=0)
  trainer.logger._default_hp_metric = None # Optional logging argument that we don't need

  # Check whether pretrained model exists. If yes, load it and skip training
  pretrained_filename = os.path.join('/home/predictor2', f"GraphLevel{model_name}.ckpt")
  if os.path.isfile(pretrained_filename):
    print("Found pretrained model, loading...")
    model = GraphLevelGNN.load_from_checkpoint(pretrained_filename)
  else:
    pl.seed_everything(42)
    model = GraphLevelGNN(c_in=dataset.num_node_features, 
                          c_out=1 if dataset.num_classes==2 else dataset.num_classes,  #change
                          **model_kwargs)
    trainer.fit(model, graph_train_loader, graph_val_loader)
    model = GraphLevelGNN.load_from_checkpoint(trainer.checkpoint_callback.best_model_path)

  # Test best model on validation and test set
  train_result = trainer.test(model, graph_train_loader, verbose=False)
  test_result = trainer.test(model, graph_test_loader, verbose=False)
  result = {"test": test_result[0]['test_acc'], "train": train_result[0]['test_acc']} 
  return model, result

Returns the error:

Traceback (most recent call last):
  File "stability_v3_alternative_net.py", line 604, in <module>
    dp_rate=0.2)
  File "stability_v3_alternative_net.py", line 591, in train_graph_classifier
    model = GraphLevelGNN.load_from_checkpoint(trainer.checkpoint_callback.best_model_path)
  File "/root/miniconda3/lib/python3.7/site-packages/pytorch_lightning/core/saving.py", line 139, in load_from_checkpoint
    checkpoint = pl_load(checkpoint_path, map_location=lambda storage, loc: storage)
  File "/root/miniconda3/lib/python3.7/site-packages/pytorch_lightning/utilities/cloud_io.py", line 46, in load
    with fs.open(path_or_url, "rb") as f:
  File "/root/miniconda3/lib/python3.7/site-packages/fsspec/spec.py", line 1043, in open
    **kwargs,
  File "/root/miniconda3/lib/python3.7/site-packages/fsspec/implementations/local.py", line 159, in _open
    return LocalFileOpener(path, mode, fs=self, **kwargs)
  File "/root/miniconda3/lib/python3.7/site-packages/fsspec/implementations/local.py", line 254, in __init__
    self._open()
  File "/root/miniconda3/lib/python3.7/site-packages/fsspec/implementations/local.py", line 259, in _open
    self.f = open(self.path, mode=self.mode)
IsADirectoryError: [Errno 21] Is a directory: '/home/predictor'

where /home/predictor is the current directory i'm working in? (I made predictor2 directory because I get the same error when I replace predictor2 with predictor in the above code).

I understand that it's telling me that it's trying to write a file or something but it's finding that the location in a directory, I can get that from seeing other people's answers. But I can't see specifically here what the issue is becaues I don't name my working directory anywhere? The code was taken from this example.

Solution

This:

model=GraphLevelGNN.load_from_checkpoint(trainer.checkpoint_callback.best_model_path)

is failing because you are trying to open a directory and not a file. Check that trainer.checkpoint_callback.best_model_path actually is the path to your .ckpt. As it looks like it is not, then you need to figure out why your callback isn't storing the right path. You can of course hardcode it yourself for an ugly solution.

Answered By - Mikel B