Update imports and requirements (#50)

* update requirements

* simplify imports

* set macos version due to error in ci (try with 12.3)

* temporarily remove macos due to code-unreleated issues in ci

* Update NeptuneCallback docstring (#51)

* update NeptuneCallback docstring

- Update for neptune 1.0
- Streamline description and examples
- Rm extra links

* tweak arg description

* phrasing

* fix imports and tests

* simplify retrieve_fit_index

---------

Co-authored-by: Sabine <sabine.nyholm@neptune.ai>

.github/workflows/ci.yml

@@ -21,7 +21,7 @@ jobs:
     runs-on: ${{ matrix.os }}
     strategy:
       matrix:
-        os: [ubuntu-latest, macos-latest] #, windows-latest]
+        os: [ubuntu-latest] #, macos-12.3] #, windows-latest]
         python-version: [3.7, 3.8, 3.9]
     steps:
       - uses: actions/checkout@v2

CHANGELOG.md

@@ -1,3 +1,10 @@
+## neptune-fastai 1.1.0
+
+### Changes
+- Updated integration for compatibility with `neptune 1.X`
+- Removed `neptune` and `neptune-client` from base requirements - installation is checked at runtime
+- Functions outside the callback accept `Handler` as well
+
 ## neptune-fastai 1.0.0
 
   ### Changes

pyproject.toml

@@ -9,13 +9,12 @@ style = "semver"
 pattern = "default-unprefixed"
 
 [tool.poetry.dependencies]
-python = "^3.6"
+python = "^3.7"
 
 # Python lack of functionalities from future versions
 importlib-metadata = { version = "*", python = "<3.8" }
 
 # Base requirements
-neptune-client = ">=0.16.17"
 fastai = ">=2.4"
 
 # dev
@@ -24,6 +23,7 @@ pytest = { version = ">=5.0", optional = true }
 pytest-tap = { version = "3.2", optional = true }
 pytest-cov = { version = "2.10.1", optional = true }
 pytest-xdist = { version = "2.2.0", optional = true }
+neptune = { version = ">=1.0.0", optional = true }
 
 [tool.poetry.extras]
 dev = [
@@ -32,6 +32,7 @@ dev = [
     "pytest-cov",
     "pytest-tap",
     "pytest-xdist",
+    "neptune",
 ]
 
 [tool.poetry]

src/neptune_fastai/impl/__init__.py

@@ -39,9 +39,18 @@
     trainable_params,
 )
 
+from neptune_fastai.impl.version import __version__
+
 try:
-    # neptune-client=0.9.0+ package structure
-    import neptune.new as neptune
+    from neptune import Run
+    from neptune.handler import Handler
+    from neptune.integrations.utils import (
+        expect_not_an_experiment,
+        verify_type,
+    )
+    from neptune.types import File
+    from neptune.utils import stringify_unsupported
+except ImportError:
     from neptune.new import Run
     from neptune.new.handler import Handler
     from neptune.new.integrations.utils import (
@@ -50,16 +59,6 @@
     )
     from neptune.new.types import File
     from neptune.new.utils import stringify_unsupported
-except ImportError:
-    # neptune-client>=1.0.0 package structure
-    import neptune  # isort:skip
-    from neptune.integrations.utils import expect_not_an_experiment, verify_type
-    from neptune.handler import Handler
-    from neptune import Run
-    from neptune.types import File  # isort:skip
-    from neptune.utils import stringify_unsupported
-
-from neptune_fastai.impl.version import __version__
 
 INTEGRATION_VERSION_KEY = "source_code/integrations/neptune-fastai"
 
@@ -70,30 +69,14 @@ class NeptuneCallback(Callback):
     The callback logs paramaters, metrics, losses, model configuration,
     optimizer configuration, and info about the dataset: path, number of samples, and hash value.
 
-    Metrics and losses are logged separately for every `learner.fit()` call.
-    For example, when you call `learner.fit(n)` the first time, it will create
-    a folder named fit_0 under the folder metrics that contains optimizer hyperparameters,
-    batch, and loader-level metrics.
-
-        metrics
-            |--> fit_0
-                |--> batch
-                |--> loader
-                |--> optimizer hyperparameters
-            |--> ...
-            |--> fit_n
-
-    Note: To try the integration without registering, you can use the public API token
-        `neptune.ANONYMOUS_API_TOKEN` and set the `project` argument to "common/fastai-integration".
-
     Args:
-        run: Neptune `Run` or namespace `Handler` object. A run is a representation of all metadata
-            that you log to Neptune.
-        base_namespace: Root namespace inside which all metadata will be logged.
+        run: Neptune run object. You can also pass a namespace handler object;
+            for example, run["test"], in which case all metadata is logged under
+            the "test" namespace inside the run.
+        base_namespace: Root namespace where all metadata logged by the callback is stored.
             If omitted, the metadata is logged without a common root namespace.
-        upload_saved_models: Which model checkpoints to upload.
-            - `"all"` (default): uploads all model checkpoints created by `SaveModelCallback()`.
-            - `"last"`: uploads the last model checkpoint created by `SaveModelCallback()`.
+        upload_saved_models: Which model checkpoints created by `SaveModelCallback()`
+            to upload: 'all' or 'last'.
 
     Examples:
 
@@ -102,8 +85,8 @@ class NeptuneCallback(Callback):
             from fastai.callback.all import SaveModelCallback
             from fastai.vision.all import (untar_data, ImageDataLoaders, ...)
 
-            import neptune.new as neptune
-            from neptune.new.integrations.fastai import NeptuneCallback
+            import neptune
+            from neptune.integrations.fastai import NeptuneCallback
 
             run = neptune.init_run()
 
@@ -119,10 +102,10 @@ class NeptuneCallback(Callback):
 
             n = 2
             learn = vision_learner(
-                ...
+                ...,
                 cbs=[
                     SaveModelCallback(every_epoch=n),
-                    NeptuneCallback(run=run, base_namespace="experiment_2", upload_saved_models="all"),
+                    NeptuneCallback(run=run, base_namespace="experiment_2"),
                 ],
             )
 
@@ -131,9 +114,6 @@ class NeptuneCallback(Callback):
     For more, see the docs:
         Tutorial: https://docs.neptune.ai/integrations/fastai
         API reference: https://docs.neptune.ai/api/integrations/fastai
-
-    Example scripts:
-        https://github.com/neptune-ai/examples/tree/main/integrations-and-supported-tools/fastai/scripts
     """
 
     order = SaveModelCallback.order + 1
@@ -148,7 +128,7 @@ def __init__(
         super().__init__(**kwargs)
 
         expect_not_an_experiment(run)
-        verify_type("run", run, (neptune.Run, neptune.handler.Handler))
+        verify_type("run", run, (Run, Handler))
         verify_type("base_namespace", base_namespace, str)
         verify_type("upload_saved_models", upload_saved_models, (str, type(None)))
 
@@ -158,7 +138,7 @@ def __init__(
         self.fit_index = retrieve_fit_index(run, f"{base_namespace}/metrics/")
 
         root_obj = run
-        if isinstance(root_obj, neptune.handler.Handler):
+        if isinstance(root_obj, Handler):
             root_obj = run.get_root_object()
         root_obj[INTEGRATION_VERSION_KEY] = __version__
 
@@ -343,7 +323,7 @@ def after_fit(self):
         self.fit_index += 1
 
 
-def _log_model_architecture(run: neptune.Run, base_namespace: str, learn: Learner):
+def _log_model_architecture(run: Union[Run, Handler], base_namespace: str, learn: Learner):
     if hasattr(learn, "arch"):
         run[f"{base_namespace}/config/model/architecture_name"] = getattr(learn.arch, "__name__", "")
 
@@ -353,7 +333,7 @@ def _log_model_architecture(run: neptune.Run, base_namespace: str, learn: Learne
     run[f"{base_namespace}/io_files/artifacts/model_architecture"].upload(model_architecture)
 
 
-def _log_dataset_metadata(run: neptune.Run, base_namespace: str, learn: Learner):
+def _log_dataset_metadata(run: Union[Run, Handler], base_namespace: str, learn: Learner):
     sha = hashlib.sha1(str(learn.dls.path).encode())
 
     run[f"{base_namespace}/io_files/resources/dataset"] = stringify_unsupported(
@@ -365,17 +345,22 @@ def _log_dataset_metadata(run: neptune.Run, base_namespace: str, learn: Learner)
     )
 
 
-def _log_or_assign_metric(run: neptune.Run, number_of_epochs: int, metric: str, value):
+def _log_or_assign_metric(run: Union[Run, Handler], number_of_epochs: int, metric: str, value):
     if number_of_epochs > 1:
         run[metric].append(value)
     else:
         run[metric] = value
 
 
-def retrieve_fit_index(run: neptune.Run, path: str) -> int:
-    return len(run.get_attribute(path) or [])
+def retrieve_fit_index(run: Union[Run, Handler], path: str) -> int:
+    root = run
+
+    if isinstance(run, Handler):
+        root = run.get_root_object()
+
+    return len(root.get_attribute(path) or [])
 
 
-def _log_optimizer_hyperparams(run: neptune.Run, prefix: str, optimizer_hyperparams: dict, n_epoch: int):
+def _log_optimizer_hyperparams(run: Union[Run, Handler], prefix: str, optimizer_hyperparams: dict, n_epoch: int):
     for param, value in optimizer_hyperparams.items():
         _log_or_assign_metric(run, n_epoch, f"{prefix}/{param}", value)

src/neptune_fastai/impl/version.py

@@ -1,6 +1,7 @@
 __all__ = ["__version__"]
 
 import sys
+from importlib.util import find_spec
 
 if sys.version_info >= (3, 8):
     from importlib.metadata import (
@@ -13,6 +14,14 @@
         version,
     )
 
+if not (find_spec("neptune") or find_spec("neptune-client")):
+    msg = """
+            The Neptune client library was not found.
+            Install the neptune package with
+                `pip install neptune`
+            Need help? -> https://docs.neptune.ai/setup/installation/"""
+    raise PackageNotFoundError(msg)
+
 try:
     __version__ = version("neptune-fastai")
 except PackageNotFoundError:

tests/conftest.py

@@ -13,6 +13,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 #
+import neptune
 import pytest
 from fastai.basics import (
     URLs,
@@ -25,13 +26,6 @@
     TabularDataLoaders,
 )
 
-try:
-    # neptune-client=0.9.0 package structure
-    from neptune import new as neptune
-except ImportError:
-    # neptune-client=1.0.0 package structure
-    import neptune
-
 
 @pytest.fixture()
 def run():

tests/neptune_fastai/test_e2e.py

@@ -15,7 +15,6 @@
 #
 from itertools import islice
 
-import neptune.new as neptune
 from fastai.basics import (
     URLs,
     accuracy,
@@ -37,7 +36,13 @@
     get_image_files,
     squeezenet1_0,
 )
-from neptune.new.integrations.fastai import NeptuneCallback
+
+try:
+    from neptune import init_run
+    from neptune.integrations.fastai import NeptuneCallback
+except ImportError:
+    from neptune.new import init_run
+    from neptune.new.integrations.fastai import NeptuneCallback
 
 import neptune_fastai
 
@@ -47,9 +52,10 @@ def is_cat(x):
 
 
 class TestE2E:
-    def test_vision_classification(self):
+    def test_vision_classification_with_handler(self):
         # given (Subject)
-        run = neptune.init_run(name="Integration fastai (vision classification)")
+        run = init_run(name="Integration fastai (vision classification)")["test"]
+        root_obj = run.get_root_object()
 
         path = untar_data(URLs.PETS) / "images"
 
@@ -71,11 +77,11 @@ def test_vision_classification(self):
         )
 
         learn.fit(1)
-        run.sync()
+        root_obj.sync()
 
         # then
         # correct integration version is logged
-        logged_version = run["source_code/integrations/neptune-fastai"].fetch()
+        logged_version = root_obj["source_code/integrations/neptune-fastai"].fetch()
         assert logged_version == neptune_fastai.__version__
 
         # and
@@ -96,7 +102,7 @@ def test_vision_classification(self):
 
     def test_tabular_model(self):
         # given (Subject)
-        run = neptune.init_run(name="Integration fastai (tabular model)")
+        run = init_run(name="Integration fastai (tabular model)")
 
         path = untar_data(URLs.ADULT_SAMPLE)