warn user when dropping unpicklable hparams (#2874)

* refactored clean_namespace

* Update try except to handle pickling error

* Consolidated clean_namespace. Added is_picklable

* PEP8

* Change warning to use rank_zero_warn. Added Test to ensure proper hparam filtering

* Updated imports

* Corrected Test Case

pytorch_lightning/utilities/__init__.py

@@ -6,7 +6,7 @@ import torch
 
 from pytorch_lightning.utilities.apply_func import move_data_to_device
 from pytorch_lightning.utilities.distributed import rank_zero_only, rank_zero_warn, rank_zero_info
-from pytorch_lightning.utilities.parsing import AttributeDict, flatten_dict
+from pytorch_lightning.utilities.parsing import AttributeDict, flatten_dict, is_picklable
 
 try:
     from apex import amp

pytorch_lightning/utilities/parsing.py

@@ -13,9 +13,12 @@
 # limitations under the License.
 
 import inspect
+import pickle
 from argparse import Namespace
 from typing import Dict
 
+from pytorch_lightning.utilities import rank_zero_warn
+
 
 def str_to_bool(val):
     """Convert a string representation of truth to true (1) or false (0).
@@ -39,26 +42,28 @@ def str_to_bool(val):
         raise ValueError(f'invalid truth value {val}')
 
 
+def is_picklable(obj: object) -> bool:
+    """Tests if an object can be pickled"""
+
+    try:
+        pickle.dumps(obj)
+        return True
+    except pickle.PicklingError:
+        return False
+
+
 def clean_namespace(hparams):
-    """Removes all functions from hparams so we can pickle."""
+    """Removes all unpicklable entries from hparams"""
 
+    hparams_dict = hparams
     if isinstance(hparams, Namespace):
-        del_attrs = []
-        for k in hparams.__dict__:
-            if callable(getattr(hparams, k)):
-                del_attrs.append(k)
+        hparams_dict = hparams.__dict__
+
+    del_attrs = [k for k, v in hparams_dict.items() if not is_picklable(v)]
 
     for k in del_attrs:
-            delattr(hparams, k)
-
-    elif isinstance(hparams, dict):
-        del_attrs = []
-        for k, v in hparams.items():
-            if callable(v):
-                del_attrs.append(k)
-
-        for k in del_attrs:
-            del hparams[k]
+        rank_zero_warn(f"attribute '{k}' removed from hparams because it cannot be pickled", UserWarning)
+        del hparams_dict[k]
 
 
 def get_init_args(frame) -> dict:

tests/models/test_hparams.py

@@ -11,7 +11,7 @@ from torch.utils.data import DataLoader
 
 from pytorch_lightning import Trainer, LightningModule
 from pytorch_lightning.core.saving import save_hparams_to_yaml, load_hparams_from_yaml
-from pytorch_lightning.utilities import AttributeDict
+from pytorch_lightning.utilities import AttributeDict, is_picklable
 from tests.base import EvalModelTemplate, TrialMNIST
 
 
@@ -282,7 +282,7 @@ def test_collect_init_arguments(tmpdir, cls):
     assert model.hparams.batch_size == 179
 
     if isinstance(model, AggSubClassEvalModel):
-        assert isinstance(model.hparams.my_loss, torch.nn.CrossEntropyLoss)
+        assert isinstance(model.hparams.my_loss, torch.nn.CosineEmbeddingLoss)
 
     if isinstance(model, DictConfSubClassEvalModel):
         assert isinstance(model.hparams.dict_conf, Container)
@@ -413,6 +413,23 @@ def test_hparams_pickle(tmpdir):
     assert ad == pickle.loads(pkl)
 
 
+class UnpickleableArgsEvalModel(EvalModelTemplate):
+    """ A model that has an attribute that cannot be pickled. """
+
+    def __init__(self, foo='bar', pickle_me=(lambda x: x + 1), **kwargs):
+        super().__init__(**kwargs)
+        assert not is_picklable(pickle_me)
+        self.save_hyperparameters()
+
+
+def test_hparams_pickle_warning(tmpdir):
+    model = UnpickleableArgsEvalModel()
+    trainer = Trainer(default_root_dir=tmpdir, max_steps=1)
+    with pytest.warns(UserWarning, match="attribute 'pickle_me' removed from hparams because it cannot be pickled"):
+        trainer.fit(model)
+    assert 'pickle_me' not in model.hparams
+
+
 def test_hparams_save_yaml(tmpdir):
     hparams = dict(batch_size=32, learning_rate=0.001, data_root='./any/path/here',
                    nasted=dict(any_num=123, anystr='abcd'))