Merge pull request #54 from neptune-ai/fix/optim_name_not_found

Don't fail if optim name not found.

CHANGELOG.md

@@ -1,3 +1,8 @@
+## neptune-fastai 1.1.1
+
+### Fixes
+- Don't error if `optim.__name__` is not present. (https://github.com/neptune-ai/neptune-fastai/pull/54)
+
 ## neptune-fastai 1.1.0
 
 ### Changes

src/neptune_fastai/impl/__init__.py

@@ -154,7 +154,17 @@ def _batch_size(self) -> int:
 
     @property
     def _optimizer_name(self) -> Optional[str]:
-        return self.opt_func.__name__
+        NA = "N/A"
+        optim_name = getattr(self.opt_func, "__name__", NA)
+        if optim_name == NA:
+            warning_msg = (
+                "NeptuneCallback: Couldn't retrieve the optimizer name, "
+                "so it will be logged as 'N/A'. You can set the optimizer "
+                "name by assigning it to the __name__ attribute. "
+                "Eg. >>> optimizer.__name__ = 'NAME'"
+            )
+            warnings.warn(warning_msg)
+        return optim_name
 
     @property
     def _device(self) -> str:

tests/neptune_fastai/test_e2e.py

@@ -13,6 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 #
+from functools import partial
 from itertools import islice
 from pathlib import Path
 
@@ -24,6 +25,7 @@
     untar_data,
 )
 from fastai.callback.all import SaveModelCallback
+from fastai.optimizer import Adam
 from fastai.tabular.all import (
     Categorify,
     FillMissing,
@@ -71,12 +73,15 @@ def test_vision_classification_with_handler(self):
             device=torch.device("cpu"),
         )
 
+        opt_func = partial(Adam, lr=3e-3, wd=0.01)
+
         learn = cnn_learner(
             dls,
             squeezenet1_0,
             metrics=error_rate,
             cbs=[NeptuneCallback(run, "experiment")],
             pretrained=False,
+            opt_func=opt_func,
         )
 
         learn.fit(1)
@@ -91,6 +96,7 @@ def test_vision_classification_with_handler(self):
         exp_config = run["experiment/config"].fetch()
         assert exp_config["batch_size"] == 64
         assert exp_config["criterion"] == "CrossEntropyLoss()"
+        assert exp_config["optimizer"]["name"] == "N/A"
         assert exp_config["input_shape"] == {"x": "[3, 224, 224]", "y": 1}
 
         # and