[AIR] Improve to_air_checkpoint with path (ray-project#26532)

Signed-off-by: Xiaowei Jiang <[email protected]>

python/ray/train/lightgbm/utils.py

@@ -17,19 +17,33 @@
 
 @PublicAPI(stability="alpha")
 def to_air_checkpoint(
-    path: str,
     booster: lightgbm.Booster,
+    *,
+    path: os.PathLike,
     preprocessor: Optional["Preprocessor"] = None,
 ) -> Checkpoint:
     """Convert a pretrained model to AIR checkpoint for serve or inference.
 
+    Example:
+
+    .. code-block:: python
+
+        import lightgbm
+        import tempfile
+        from ray.train.lightgbm import to_air_checkpoint, LightGBMPredictor
+
+        bst = lightgbm.Booster()
+        with tempfile.TemporaryDirectory() as tmpdir:
+            checkpoint = to_air_checkpoint(booster=bst, path=tmpdir)
+            predictor = LightGBMPredictor.from_checkpoint(checkpoint)
+
     Args:
-        path: The directory path where model and preprocessor steps are stored to.
         booster: A pretrained lightgbm model.
+        path: The directory where the checkpoint will be stored to.
         preprocessor: A fitted preprocessor. The preprocessing logic will
-            be applied to serve/inference.
+            be applied to the inputs for serving/inference.
     Returns:
-        A Ray Air checkpoint.
+        A Ray AIR checkpoint.
     """
     booster.save_model(os.path.join(path, MODEL_KEY))
 

python/ray/train/sklearn/utils.py

@@ -18,17 +18,31 @@
 
 @PublicAPI(stability="alpha")
 def to_air_checkpoint(
-    path: str,
     estimator: BaseEstimator,
+    *,
+    path: os.PathLike,
     preprocessor: Optional["Preprocessor"] = None,
 ) -> Checkpoint:
     """Convert a pretrained model to AIR checkpoint for serve or inference.
 
+    Example:
+
+    .. code-block:: python
+
+        import tempfile
+        from sklearn.ensemble import RandomForestClassifier
+        from ray.train.sklearn import to_air_checkpoint, SklearnPredictor
+
+        est = RandomForestClassifier()
+        with tempfile.TemporaryDirectory() as tmpdir:
+            checkpoint = to_air_checkpoint(estimator=est, path=tmpdir)
+            predictor = SklearnPredictor.from_checkpoint(checkpoint)
+
     Args:
-        path: The directory path where model and preprocessor steps are stored to.
         estimator: A pretrained model.
+        path: The directory where the checkpoint will be stored to.
         preprocessor: A fitted preprocessor. The preprocessing logic will
-            be applied to serve/inference.
+            be applied to the inputs for serving/inference.
     Returns:
         A Ray Air checkpoint.
     """

python/ray/train/tensorflow/utils.py

@@ -14,14 +14,14 @@
 
 @PublicAPI(stability="alpha")
 def to_air_checkpoint(
-    model: keras.Model, preprocessor: Optional["Preprocessor"] = None
+    model: keras.Model, *, preprocessor: Optional["Preprocessor"] = None
 ) -> Checkpoint:
     """Convert a pretrained model to AIR checkpoint for serve or inference.
 
     Args:
         model: A pretrained model.
         preprocessor: A fitted preprocessor. The preprocessing logic will
-            be applied to serve/inference.
+            be applied to the inputs for serving/inference.
     Returns:
         A Ray Air checkpoint.
     """

python/ray/train/tests/test_lightgbm_predictor.py

@@ -95,7 +95,7 @@ def test_predict_feature_columns_pandas():
 
 def test_predict_no_preprocessor_no_training():
     with tempfile.TemporaryDirectory() as tmpdir:
-        checkpoint = to_air_checkpoint(tmpdir, booster=model)
+        checkpoint = to_air_checkpoint(booster=model, path=tmpdir)
         predictor = LightGBMPredictor.from_checkpoint(checkpoint)
 
     data_batch = np.array([[1, 2], [3, 4], [5, 6]])

python/ray/train/tests/test_sklearn_predictor.py

@@ -146,7 +146,7 @@ def test_batch_prediction_with_set_cpus(ray_start_4_cpus):
 
 def test_sklearn_predictor_no_training():
     with tempfile.TemporaryDirectory() as tmpdir:
-        checkpoint = to_air_checkpoint(path=tmpdir, estimator=model)
+        checkpoint = to_air_checkpoint(estimator=model, path=tmpdir)
         batch_predictor = BatchPredictor.from_checkpoint(checkpoint, SklearnPredictor)
         test_dataset = ray.data.from_pandas(
             pd.DataFrame(dummy_data, columns=["A", "B"])

python/ray/train/tests/test_xgboost_predictor.py

@@ -97,7 +97,7 @@ def test_predict_feature_columns_pandas():
 
 def test_predict_no_preprocessor_no_training():
     with tempfile.TemporaryDirectory() as tmpdir:
-        checkpoint = to_air_checkpoint(tmpdir, booster=model)
+        checkpoint = to_air_checkpoint(booster=model, path=tmpdir)
         predictor = XGBoostPredictor.from_checkpoint(checkpoint)
 
     data_batch = np.array([[1, 2], [3, 4], [5, 6]])

python/ray/train/torch/utils.py

@@ -14,14 +14,14 @@
 
 @PublicAPI(stability="alpha")
 def to_air_checkpoint(
-    model: torch.nn.Module, preprocessor: Optional["Preprocessor"] = None
+    model: torch.nn.Module, *, preprocessor: Optional["Preprocessor"] = None
 ) -> Checkpoint:
     """Convert a pretrained model to AIR checkpoint for serve or inference.
 
     Args:
         model: A pretrained model.
         preprocessor: A fitted preprocessor. The preprocessing logic will
-            be applied to serve/inference.
+            be applied to the inputs for serving/inference.
     Returns:
         A Ray Air checkpoint.
     """

python/ray/train/xgboost/utils.py

@@ -17,19 +17,34 @@
 
 @PublicAPI(stability="alpha")
 def to_air_checkpoint(
-    path: str,
     booster: xgboost.Booster,
+    *,
+    path: os.PathLike,
     preprocessor: Optional["Preprocessor"] = None,
 ) -> Checkpoint:
     """Convert a pretrained model to AIR checkpoint for serve or inference.
 
+    Example:
+
+    .. code-block:: python
+
+        import xgboost
+        import tempfile
+        from ray.train.xgboost import to_air_checkpoint, XGBoostPredictor
+
+        bst = xgboost.Booster()
+        with tempfile.TemporaryDirectory() as tmpdir:
+            checkpoint = to_air_checkpoint(booster=bst, path=tmpdir)
+            predictor = XGBoostPredictor.from_checkpoint(checkpoint)
+
     Args:
-        path: The directory path where model and preprocessor steps are stored to.
         booster: A pretrained xgboost model.
+        path: The directory where the checkpoint will be stored to.
         preprocessor: A fitted preprocessor. The preprocessing logic will
-            be applied to serve/inference.
+            be applied to the inputs for serving/inference.
+
     Returns:
-        A Ray Air checkpoint.
+        A Ray AIR checkpoint.
     """
     booster.save_model(os.path.join(path, MODEL_KEY))