snowflake-ml-python 1.7.4__py3-none-any.whl → 1.8.0__py3-none-any.whl

snowflake/ml/model/_signatures/core.py

@@ -199,9 +199,16 @@ class DataType(Enum):
 class BaseFeatureSpec(ABC):
     """Abstract Class for specification of a feature."""
 
-    def __init__(self, name: str) -> None:
+    def __init__(self, name: str, shape: Optional[Tuple[int, ...]]) -> None:
         self._name = name
 
+        if shape and not isinstance(shape, tuple):
+            raise snowml_exceptions.SnowflakeMLException(
+                error_code=error_codes.INVALID_TYPE,
+                original_exception=TypeError("Shape should be a tuple if presented."),
+            )
+        self._shape = shape
+
     @final
     @property
     def name(self) -> str:
@@ -213,6 +220,11 @@ class BaseFeatureSpec(ABC):
         """Convert to corresponding Snowpark Type."""
         pass
 
+    @abstractmethod
+    def as_dtype(self, force_numpy_dtype: bool = False) -> Union[npt.DTypeLike, str, PandasExtensionTypes]:
+        """Convert to corresponding local Type."""
+        pass
+
     @abstractmethod
     def to_dict(self) -> Dict[str, Any]:
         """Serialization"""
@@ -256,7 +268,7 @@ class FeatureSpec(BaseFeatureSpec):
             SnowflakeMLException: TypeError: When the dtype input type is incorrect.
             SnowflakeMLException: TypeError: When the shape input type is incorrect.
         """
-        super().__init__(name=name)
+        super().__init__(name=name, shape=shape)
 
         if not isinstance(dtype, DataType):
             raise snowml_exceptions.SnowflakeMLException(
@@ -265,13 +277,6 @@ class FeatureSpec(BaseFeatureSpec):
             )
         self._dtype = dtype
 
-        if shape and not isinstance(shape, tuple):
-            raise snowml_exceptions.SnowflakeMLException(
-                error_code=error_codes.INVALID_TYPE,
-                original_exception=TypeError("Shape should be a tuple if presented."),
-            )
-        self._shape = shape
-
         self._nullable = nullable
 
     def as_snowpark_type(self) -> spt.DataType:
@@ -282,7 +287,7 @@ class FeatureSpec(BaseFeatureSpec):
             result_type = spt.ArrayType(result_type)
         return result_type
 
-    def as_dtype(self) -> Union[npt.DTypeLike, str, PandasExtensionTypes]:
+    def as_dtype(self, force_numpy_dtype: bool = False) -> Union[npt.DTypeLike, str, PandasExtensionTypes]:
         """Convert to corresponding local Type."""
 
         if not self._shape:
@@ -291,7 +296,7 @@ class FeatureSpec(BaseFeatureSpec):
                 return self._dtype._value
 
             np_type = self._dtype._numpy_type
-            if self._nullable:
+            if self._nullable and not force_numpy_dtype:
                 np_to_pd_dtype_mapping = {
                     np.int8: pd.Int8Dtype(),
                     np.int16: pd.Int16Dtype(),
@@ -386,15 +391,23 @@ class FeatureSpec(BaseFeatureSpec):
 class FeatureGroupSpec(BaseFeatureSpec):
     """Specification of a group of features in Snowflake native model packaging."""
 
-    def __init__(self, name: str, specs: List[FeatureSpec]) -> None:
+    def __init__(self, name: str, specs: List[BaseFeatureSpec], shape: Optional[Tuple[int, ...]] = None) -> None:
         """Initialize a feature group.
 
         Args:
             name: Name of the feature group.
             specs: A list of feature specifications that composes the group. All children feature specs have to have
                 name. And all of them should have the same type.
+            shape: Used to represent scalar feature, 1-d feature list,
+                or n-d tensor. Use -1 to represent variable length. Defaults to None.
+
+                Examples:
+                    - None: scalar
+                    - (2,): 1d list with a fixed length of 2.
+                    - (-1,): 1d list with variable length, used for ragged tensor representation.
+                    - (d1, d2, d3): 3d tensor.
         """
-        super().__init__(name=name)
+        super().__init__(name=name, shape=shape)
         self._specs = specs
         self._validate()
 
@@ -409,47 +422,52 @@ class FeatureGroupSpec(BaseFeatureSpec):
                 error_code=error_codes.INVALID_ARGUMENT,
                 original_exception=ValueError("All children feature specs have to have name."),
             )
-        if not (all(s._shape is None for s in self._specs) or all(s._shape is not None for s in self._specs)):
-            raise snowml_exceptions.SnowflakeMLException(
-                error_code=error_codes.INVALID_ARGUMENT,
-                original_exception=ValueError("All children feature specs have to have same shape."),
-            )
-        first_type = self._specs[0]._dtype
-        if not all(s._dtype == first_type for s in self._specs):
-            raise snowml_exceptions.SnowflakeMLException(
-                error_code=error_codes.INVALID_ARGUMENT,
-                original_exception=ValueError("All children feature specs have to have same type."),
-            )
 
     def as_snowpark_type(self) -> spt.DataType:
-        first_type = self._specs[0].as_snowpark_type()
-        return spt.MapType(spt.StringType(), first_type)
+        spt_type = spt.StructType(
+            fields=[
+                spt.StructField(
+                    s._name, datatype=s.as_snowpark_type(), nullable=s._nullable if isinstance(s, FeatureSpec) else True
+                )
+                for s in self._specs
+            ]
+        )
+        if not self._shape:
+            return spt_type
+        return spt.ArrayType(spt_type)
 
     def __eq__(self, other: object) -> bool:
         if isinstance(other, FeatureGroupSpec):
-            return self._specs == other._specs
+            return self._name == other._name and self._specs == other._specs and self._shape == other._shape
         else:
             return False
 
     def __repr__(self) -> str:
         spec_strs = ",\n\t\t".join(repr(spec) for spec in self._specs)
+        shape_str = f", shape={repr(self._shape)}" if self._shape else ""
         return textwrap.dedent(
             f"""FeatureGroupSpec(
                 name={repr(self._name)},
                 specs=[
                     {spec_strs}
-                ]
+                ]{shape_str}
             )
             """
         )
 
+    def as_dtype(self, force_numpy_dtype: bool = False) -> Union[npt.DTypeLike, str, PandasExtensionTypes]:
+        return np.object_
+
     def to_dict(self) -> Dict[str, Any]:
         """Serialize the feature group into a dict.
 
         Returns:
             A dict that serializes the feature group.
         """
-        return {"feature_group": {"name": self._name, "specs": [s.to_dict() for s in self._specs]}}
+        base_dict: Dict[str, Any] = {"name": self._name, "specs": [s.to_dict() for s in self._specs]}
+        if self._shape is not None:
+            base_dict["shape"] = self._shape
+        return base_dict
 
     @classmethod
     def from_dict(cls, input_dict: Dict[str, Any]) -> "FeatureGroupSpec":
@@ -462,10 +480,13 @@ class FeatureGroupSpec(BaseFeatureSpec):
             A feature group instance deserialized and created from the dict.
         """
         specs = []
-        for e in input_dict["feature_group"]["specs"]:
-            spec = FeatureSpec.from_dict(e)
+        for e in input_dict["specs"]:
+            spec = FeatureGroupSpec.from_dict(e) if "specs" in e else FeatureSpec.from_dict(e)
             specs.append(spec)
-        return FeatureGroupSpec(name=input_dict["feature_group"]["name"], specs=specs)
+        shape = input_dict.get("shape", None)
+        if shape:
+            shape = tuple(shape)
+        return FeatureGroupSpec(name=input_dict["name"], specs=specs, shape=shape)
 
 
 class ModelSignature:
@@ -525,7 +546,7 @@ class ModelSignature:
         sig_inputs = loaded["inputs"]
 
         deserialize_spec: Callable[[Dict[str, Any]], BaseFeatureSpec] = lambda sig_spec: (
-            FeatureGroupSpec.from_dict(sig_spec) if "feature_group" in sig_spec else FeatureSpec.from_dict(sig_spec)
+            FeatureGroupSpec.from_dict(sig_spec) if "specs" in sig_spec else FeatureSpec.from_dict(sig_spec)
         )
 
         return ModelSignature(

snowflake/ml/model/_signatures/dmatrix_handler.py

@@ -0,0 +1,98 @@
+from typing import TYPE_CHECKING, List, Literal, Optional, Sequence
+
+import numpy as np
+import pandas as pd
+from typing_extensions import TypeGuard
+
+from snowflake.ml._internal import type_utils
+from snowflake.ml._internal.exceptions import (
+    error_codes,
+    exceptions as snowml_exceptions,
+)
+from snowflake.ml.model import type_hints as model_types
+from snowflake.ml.model._signatures import base_handler, core
+
+if TYPE_CHECKING:
+    import xgboost
+
+
+class XGBoostDMatrixHandler(base_handler.BaseDataHandler["xgboost.DMatrix"]):
+    @staticmethod
+    def can_handle(data: model_types.SupportedDataType) -> TypeGuard["xgboost.DMatrix"]:
+        return type_utils.LazyType("xgboost.DMatrix").isinstance(data)
+
+    @staticmethod
+    def count(data: "xgboost.DMatrix") -> int:
+        return data.num_row()
+
+    @staticmethod
+    def truncate(data: "xgboost.DMatrix", length: int) -> "xgboost.DMatrix":
+
+        num_rows = min(
+            XGBoostDMatrixHandler.count(data),
+            length,
+        )
+        return data.slice(list(range(num_rows)))
+
+    @staticmethod
+    def validate(data: "xgboost.DMatrix") -> None:
+        if data.num_row() == 0:
+            raise snowml_exceptions.SnowflakeMLException(
+                error_code=error_codes.INVALID_DATA,
+                original_exception=ValueError("Data Validation Error: Empty data is found."),
+            )
+
+    @staticmethod
+    def infer_signature(data: "xgboost.DMatrix", role: Literal["input", "output"]) -> Sequence[core.BaseFeatureSpec]:
+        feature_prefix = f"{XGBoostDMatrixHandler.FEATURE_PREFIX}_"
+        features: List[core.BaseFeatureSpec] = []
+        role_prefix = (
+            XGBoostDMatrixHandler.INPUT_PREFIX if role == "input" else XGBoostDMatrixHandler.OUTPUT_PREFIX
+        ) + "_"
+
+        feature_names = data.feature_names or []
+        feature_types = data.feature_types or []
+
+        for i, (feature_name, dtype) in enumerate(zip(feature_names, feature_types)):
+            if not feature_name:
+                ft_name = f"{role_prefix}{feature_prefix}{i}"
+            else:
+                ft_name = feature_name
+
+            features.append(core.FeatureSpec(dtype=core.DataType.from_numpy_type(np.dtype(dtype)), name=ft_name))
+        return features
+
+    @staticmethod
+    def convert_to_df(data: "xgboost.DMatrix", ensure_serializable: bool = True) -> pd.DataFrame:
+        df = pd.DataFrame(data.get_data().toarray(), columns=data.feature_names)
+
+        feature_types = data.feature_types or []
+
+        if feature_types:
+            for idx, col in enumerate(df.columns):
+                dtype = feature_types[idx]
+                df[col] = df[col].astype(dtype)
+
+        return df
+
+    @staticmethod
+    def convert_from_df(
+        df: pd.DataFrame, features: Optional[Sequence[core.BaseFeatureSpec]] = None
+    ) -> "xgboost.DMatrix":
+        import xgboost as xgb
+
+        if not features:
+            return xgb.DMatrix(df)
+        else:
+            feature_names = []
+            feature_types = []
+            for feature in features:
+                if isinstance(feature, core.FeatureGroupSpec):
+                    raise snowml_exceptions.SnowflakeMLException(
+                        error_code=error_codes.NOT_IMPLEMENTED,
+                        original_exception=NotImplementedError("FeatureGroupSpec is not supported."),
+                    )
+                assert isinstance(feature, core.FeatureSpec), "Invalid feature kind."
+                feature_names.append(feature.name)
+                feature_types.append(feature._dtype._numpy_type)
+            return xgb.DMatrix(df, feature_names=feature_names, feature_types=feature_types)

snowflake/ml/model/_signatures/numpy_handler.py

@@ -1,5 +1,5 @@
 from collections import abc
-from typing import List, Literal, Sequence
+from typing import Literal, Sequence
 
 import numpy as np
 import pandas as pd
@@ -10,7 +10,7 @@ from snowflake.ml._internal.exceptions import (
     exceptions as snowml_exceptions,
 )
 from snowflake.ml.model import type_hints as model_types
-from snowflake.ml.model._signatures import base_handler, core
+from snowflake.ml.model._signatures import base_handler, core, pandas_handler
 
 
 class NumpyArrayHandler(base_handler.BaseDataHandler[model_types._SupportedNumpyArray]):
@@ -46,11 +46,15 @@ class NumpyArrayHandler(base_handler.BaseDataHandler[model_types._SupportedNumpy
     def infer_signature(
         data: model_types._SupportedNumpyArray, role: Literal["input", "output"]
     ) -> Sequence[core.BaseFeatureSpec]:
+        if data.dtype == np.object_:
+            return pandas_handler.PandasDataFrameHandler.infer_signature(
+                NumpyArrayHandler.convert_to_df(data), role=role
+            )
         feature_prefix = f"{NumpyArrayHandler.FEATURE_PREFIX}_"
         dtype = core.DataType.from_numpy_type(data.dtype)
         role_prefix = (NumpyArrayHandler.INPUT_PREFIX if role == "input" else NumpyArrayHandler.OUTPUT_PREFIX) + "_"
         if len(data.shape) == 1:
-            return [core.FeatureSpec(dtype=dtype, name=f"{role_prefix}{feature_prefix}0")]
+            return [core.FeatureSpec(dtype=dtype, name=f"{role_prefix}{feature_prefix}0", nullable=False)]
         else:
             # For high-dimension array, 0-axis is for batch, 1-axis is for column, further more is details of columns.
             features = []
@@ -59,9 +63,9 @@ class NumpyArrayHandler(base_handler.BaseDataHandler[model_types._SupportedNumpy
             for col_data, ft_name in zip(data[0], ft_names):
                 if isinstance(col_data, np.ndarray):
                     ft_shape = np.shape(col_data)
-                    features.append(core.FeatureSpec(dtype=dtype, name=ft_name, shape=ft_shape))
+                    features.append(core.FeatureSpec(dtype=dtype, name=ft_name, shape=ft_shape, nullable=False))
                 else:
-                    features.append(core.FeatureSpec(dtype=dtype, name=ft_name))
+                    features.append(core.FeatureSpec(dtype=dtype, name=ft_name, nullable=False))
             return features
 
     @staticmethod
@@ -108,21 +112,9 @@ class SeqOfNumpyArrayHandler(base_handler.BaseDataHandler[Sequence[model_types._
     def infer_signature(
         data: Sequence[model_types._SupportedNumpyArray], role: Literal["input", "output"]
     ) -> Sequence[core.BaseFeatureSpec]:
-        feature_prefix = f"{SeqOfNumpyArrayHandler.FEATURE_PREFIX}_"
-        features: List[core.BaseFeatureSpec] = []
-        role_prefix = (
-            SeqOfNumpyArrayHandler.INPUT_PREFIX if role == "input" else SeqOfNumpyArrayHandler.OUTPUT_PREFIX
-        ) + "_"
-
-        for i, data_col in enumerate(data):
-            dtype = core.DataType.from_numpy_type(data_col.dtype)
-            ft_name = f"{role_prefix}{feature_prefix}{i}"
-            if len(data_col.shape) == 1:
-                features.append(core.FeatureSpec(dtype=dtype, name=ft_name))
-            else:
-                ft_shape = tuple(data_col.shape[1:])
-                features.append(core.FeatureSpec(dtype=dtype, name=ft_name, shape=ft_shape))
-        return features
+        return pandas_handler.PandasDataFrameHandler.infer_signature(
+            SeqOfNumpyArrayHandler.convert_to_df(data, ensure_serializable=False), role=role
+        )
 
     @staticmethod
     def convert_to_df(

snowflake/ml/model/_signatures/pandas_handler.py

@@ -72,13 +72,6 @@ class PandasDataFrameHandler(base_handler.BaseDataHandler[pd.DataFrame]):
         df_col_dtypes = [data[col].dtype for col in data.columns]
         for df_col, df_col_dtype in zip(df_cols, df_col_dtypes):
             df_col_data = data[df_col]
-            if df_col_data.isnull().all():
-                raise snowml_exceptions.SnowflakeMLException(
-                    error_code=error_codes.INVALID_DATA,
-                    original_exception=ValueError(
-                        f"Data Validation Error: There is no non-null data in column {df_col}."
-                    ),
-                )
             if df_col_data.isnull().any():
                 warnings.warn(
                     (
@@ -104,25 +97,7 @@ class PandasDataFrameHandler(base_handler.BaseDataHandler[pd.DataFrame]):
                         ),
                     )
 
-                if isinstance(df_col_data.iloc[0], list):
-                    arr = utils.convert_list_to_ndarray(df_col_data.iloc[0])
-                    arr_dtype = core.DataType.from_numpy_type(arr.dtype)
-
-                    converted_data_list = [utils.convert_list_to_ndarray(data_row) for data_row in df_col_data]
-
-                    if not all(
-                        core.DataType.from_numpy_type(converted_data.dtype) == arr_dtype
-                        for converted_data in converted_data_list
-                    ):
-                        raise snowml_exceptions.SnowflakeMLException(
-                            error_code=error_codes.INVALID_DATA,
-                            original_exception=ValueError(
-                                "Data Validation Error: "
-                                + f"Inconsistent type of element in object found in column data {df_col_data}."
-                            ),
-                        )
-
-                elif isinstance(df_col_data.iloc[0], np.ndarray):
+                if isinstance(df_col_data.iloc[0], np.ndarray):
                     arr_dtype = core.DataType.from_numpy_type(df_col_data.iloc[0].dtype)
 
                     if not all(core.DataType.from_numpy_type(data_row.dtype) == arr_dtype for data_row in df_col_data):
@@ -133,7 +108,7 @@ class PandasDataFrameHandler(base_handler.BaseDataHandler[pd.DataFrame]):
                                 + f"Inconsistent type of element in object found in column data {df_col_data}."
                             ),
                         )
-                elif not isinstance(df_col_data.iloc[0], (str, bytes)):
+                elif not isinstance(df_col_data.iloc[0], (str, bytes, dict, list)):
                     raise snowml_exceptions.SnowflakeMLException(
                         error_code=error_codes.INVALID_DATA,
                         original_exception=ValueError(
@@ -163,22 +138,38 @@ class PandasDataFrameHandler(base_handler.BaseDataHandler[pd.DataFrame]):
         specs = []
         for df_col, df_col_dtype, ft_name in zip(df_cols, df_col_dtypes, ft_names):
             df_col_data = data[df_col]
+
+            if df_col_data.isnull().all():
+                raise snowml_exceptions.SnowflakeMLException(
+                    error_code=error_codes.INVALID_DATA,
+                    original_exception=ValueError(
+                        "Data Validation Error: "
+                        f"There is no non-null data in column {df_col} so the signature cannot be inferred."
+                    ),
+                )
             if df_col_data.isnull().any():
                 df_col_data = utils.series_dropna(df_col_data)
             df_col_dtype = df_col_data.dtype
 
             if df_col_dtype == np.dtype("O"):
                 if isinstance(df_col_data.iloc[0], list):
-                    arr = utils.convert_list_to_ndarray(df_col_data.iloc[0])
-                    arr_dtype = core.DataType.from_numpy_type(arr.dtype)
-                    ft_shape = np.shape(df_col_data.iloc[0])
-
-                    converted_data_list = [utils.convert_list_to_ndarray(data_row) for data_row in df_col_data]
-
-                    if not all(np.shape(converted_data) == ft_shape for converted_data in converted_data_list):
-                        ft_shape = (-1,)
-
-                    specs.append(core.FeatureSpec(dtype=arr_dtype, name=ft_name, shape=ft_shape))
+                    spec_0 = utils.infer_list(ft_name, df_col_data.iloc[0])
+                    for i in range(1, len(df_col_data)):
+                        spec = utils.infer_list(ft_name, df_col_data.iloc[i])
+                        if spec._shape != spec_0._shape:
+                            spec_0._shape = (-1,)
+                            spec._shape = (-1,)
+                        if spec != spec_0:
+                            raise snowml_exceptions.SnowflakeMLException(
+                                error_code=error_codes.INVALID_DATA,
+                                original_exception=ValueError(
+                                    "Unable to construct signature: "
+                                    f"Ragged nested or Unsupported list-like data {df_col_data} confronted."
+                                ),
+                            )
+                    specs.append(spec_0)
+                elif isinstance(df_col_data.iloc[0], dict):
+                    specs.append(utils.infer_dict(ft_name, df_col_data.iloc[0]))
                 elif isinstance(df_col_data.iloc[0], np.ndarray):
                     arr_dtype = core.DataType.from_numpy_type(df_col_data.iloc[0].dtype)
                     ft_shape = np.shape(df_col_data.iloc[0])

snowflake/ml/model/_signatures/pytorch_handler.py

@@ -1,5 +1,5 @@
 from collections import abc
-from typing import TYPE_CHECKING, List, Literal, Optional, Sequence
+from typing import TYPE_CHECKING, Literal, Optional, Sequence
 
 import numpy as np
 import pandas as pd
@@ -11,12 +11,54 @@ from snowflake.ml._internal.exceptions import (
     exceptions as snowml_exceptions,
 )
 from snowflake.ml.model import type_hints as model_types
-from snowflake.ml.model._signatures import base_handler, core
+from snowflake.ml.model._signatures import base_handler, core, numpy_handler
 
 if TYPE_CHECKING:
     import torch
 
 
+class PyTorchTensorHandler(base_handler.BaseDataHandler["torch.Tensor"]):
+    @staticmethod
+    def can_handle(data: model_types.SupportedDataType) -> TypeGuard["torch.Tensor"]:
+        return type_utils.LazyType("torch.Tensor").isinstance(data)
+
+    @staticmethod
+    def count(data: "torch.Tensor") -> int:
+        return data.shape[0]
+
+    @staticmethod
+    def truncate(data: "torch.Tensor", length: int) -> "torch.Tensor":
+        return data[: min(PyTorchTensorHandler.count(data), length)]
+
+    @staticmethod
+    def validate(data: "torch.Tensor") -> None:
+        return numpy_handler.NumpyArrayHandler.validate(data.detach().cpu().numpy())
+
+    @staticmethod
+    def infer_signature(data: "torch.Tensor", role: Literal["input", "output"]) -> Sequence[core.BaseFeatureSpec]:
+        return numpy_handler.NumpyArrayHandler.infer_signature(data.detach().cpu().numpy(), role=role)
+
+    @staticmethod
+    def convert_to_df(data: "torch.Tensor", ensure_serializable: bool = True) -> pd.DataFrame:
+        return numpy_handler.NumpyArrayHandler.convert_to_df(
+            data.detach().cpu().numpy(), ensure_serializable=ensure_serializable
+        )
+
+    @staticmethod
+    def convert_from_df(df: pd.DataFrame, features: Optional[Sequence[core.BaseFeatureSpec]] = None) -> "torch.Tensor":
+        import torch
+
+        if features is None:
+            if any(dtype == np.dtype("O") for dtype in df.dtypes):
+                return torch.from_numpy(np.array(df.to_numpy().tolist()))
+            return torch.from_numpy(df.to_numpy())
+
+        assert isinstance(features[0], core.FeatureSpec)
+        return torch.from_numpy(
+            np.array(df.to_numpy().tolist(), dtype=features[0]._dtype._numpy_type),
+        )
+
+
 class SeqOfPyTorchTensorHandler(base_handler.BaseDataHandler[Sequence["torch.Tensor"]]):
     @staticmethod
     def can_handle(data: model_types.SupportedDataType) -> TypeGuard[Sequence["torch.Tensor"]]:
@@ -24,56 +66,28 @@ class SeqOfPyTorchTensorHandler(base_handler.BaseDataHandler[Sequence["torch.Ten
             return False
         if len(data) == 0:
             return False
-        if type_utils.LazyType("torch.Tensor").isinstance(data[0]):
-            return all(type_utils.LazyType("torch.Tensor").isinstance(data_col) for data_col in data)
-        return False
+        return all(PyTorchTensorHandler.can_handle(data_col) for data_col in data)
 
     @staticmethod
     def count(data: Sequence["torch.Tensor"]) -> int:
-        return min(data_col.shape[0] for data_col in data)  # type: ignore[no-any-return]
+        return min(PyTorchTensorHandler.count(data_col) for data_col in data)
 
     @staticmethod
     def truncate(data: Sequence["torch.Tensor"], length: int) -> Sequence["torch.Tensor"]:
-        return [data_col[: min(SeqOfPyTorchTensorHandler.count(data), 10)] for data_col in data]
+        return [data_col[: min(SeqOfPyTorchTensorHandler.count(data), length)] for data_col in data]
 
     @staticmethod
     def validate(data: Sequence["torch.Tensor"]) -> None:
-        import torch
-
         for data_col in data:
-            if data_col.shape == torch.Size([0]):
-                # Empty array
-                raise snowml_exceptions.SnowflakeMLException(
-                    error_code=error_codes.INVALID_DATA,
-                    original_exception=ValueError("Data Validation Error: Empty data is found."),
-                )
-
-            if data_col.shape == torch.Size([1]):
-                # scalar
-                raise snowml_exceptions.SnowflakeMLException(
-                    error_code=error_codes.INVALID_DATA,
-                    original_exception=ValueError("Data Validation Error: Scalar data is found."),
-                )
+            PyTorchTensorHandler.validate(data_col)
 
     @staticmethod
     def infer_signature(
         data: Sequence["torch.Tensor"], role: Literal["input", "output"]
     ) -> Sequence[core.BaseFeatureSpec]:
-        feature_prefix = f"{SeqOfPyTorchTensorHandler.FEATURE_PREFIX}_"
-        features: List[core.BaseFeatureSpec] = []
-        role_prefix = (
-            SeqOfPyTorchTensorHandler.INPUT_PREFIX if role == "input" else SeqOfPyTorchTensorHandler.OUTPUT_PREFIX
-        ) + "_"
-
-        for i, data_col in enumerate(data):
-            dtype = core.DataType.from_torch_type(data_col.dtype)
-            ft_name = f"{role_prefix}{feature_prefix}{i}"
-            if len(data_col.shape) == 1:
-                features.append(core.FeatureSpec(dtype=dtype, name=ft_name, nullable=False))
-            else:
-                ft_shape = tuple(data_col.shape[1:])
-                features.append(core.FeatureSpec(dtype=dtype, name=ft_name, shape=ft_shape, nullable=False))
-        return features
+        return numpy_handler.SeqOfNumpyArrayHandler.infer_signature(
+            [data_col.detach().cpu().numpy() for data_col in data], role=role
+        )
 
     @staticmethod
     def convert_to_df(data: Sequence["torch.Tensor"], ensure_serializable: bool = True) -> pd.DataFrame:
@@ -81,8 +95,8 @@ class SeqOfPyTorchTensorHandler(base_handler.BaseDataHandler[Sequence["torch.Ten
         # the content is still numpy array so that the type could be preserved.
         # But that would not serializable and cannot use as UDF input and output.
         if ensure_serializable:
-            return pd.DataFrame({i: data_col.detach().to("cpu").numpy().tolist() for i, data_col in enumerate(data)})
-        return pd.DataFrame({i: list(data_col.detach().to("cpu").numpy()) for i, data_col in enumerate(data)})
+            return pd.DataFrame({i: data_col.detach().cpu().numpy().tolist() for i, data_col in enumerate(data)})
+        return pd.DataFrame({i: list(data_col.detach().cpu().numpy()) for i, data_col in enumerate(data)})
 
     @staticmethod
     def convert_from_df(
@@ -95,8 +109,10 @@ class SeqOfPyTorchTensorHandler(base_handler.BaseDataHandler[Sequence["torch.Ten
             for feature in features:
                 if isinstance(feature, core.FeatureGroupSpec):
                     raise snowml_exceptions.SnowflakeMLException(
-                        error_code=error_codes.NOT_IMPLEMENTED,
-                        original_exception=NotImplementedError("FeatureGroupSpec is not supported."),
+                        error_code=error_codes.INVALID_DATA_TYPE,
+                        original_exception=NotImplementedError(
+                            "FeatureGroupSpec is not supported when converting to Tensorflow tensor."
+                        ),
                     )
                 assert isinstance(feature, core.FeatureSpec), "Invalid feature kind."
                 res.append(torch.from_numpy(np.stack(df[feature.name].to_numpy()).astype(feature._dtype._numpy_type)))

snowflake/ml/model/_signatures/snowpark_handler.py

@@ -65,12 +65,6 @@ class SnowparkDataFrameHandler(base_handler.BaseDataHandler[snowflake.snowpark.D
         dtype_map = {}
         if features:
             for feature in features:
-                if isinstance(feature, core.FeatureGroupSpec):
-                    raise snowml_exceptions.SnowflakeMLException(
-                        error_code=error_codes.NOT_IMPLEMENTED,
-                        original_exception=NotImplementedError("FeatureGroupSpec is not supported."),
-                    )
-                assert isinstance(feature, core.FeatureSpec), "Invalid feature kind."
                 dtype_map[feature.name] = feature.as_dtype()
         df_local = data.to_pandas()
 
@@ -122,12 +116,6 @@ class SnowparkDataFrameHandler(base_handler.BaseDataHandler[snowflake.snowpark.D
         column_names = []
         columns = []
         for feature in features:
-            if isinstance(feature, core.FeatureGroupSpec):
-                raise snowml_exceptions.SnowflakeMLException(
-                    error_code=error_codes.NOT_IMPLEMENTED,
-                    original_exception=NotImplementedError("FeatureGroupSpec is not supported."),
-                )
-            assert isinstance(feature, core.FeatureSpec), "Invalid feature kind."
             column_names.append(identifier.get_inferred_name(feature.name))
             columns.append(F.col(identifier.get_inferred_name(feature.name)).cast(feature.as_snowpark_type()))