snowflake-ml-python 1.1.0__py3-none-any.whl → 1.1.2__py3-none-any.whl

snowflake/ml/modeling/_internal/snowpark_trainer.py

@@ -0,0 +1,331 @@
+import importlib
+import inspect
+import os
+import posixpath
+from typing import Any, Callable, Dict, List, Optional, Tuple
+
+import cloudpickle as cp
+
+from snowflake.ml._internal import telemetry
+from snowflake.ml._internal.exceptions import (
+    error_codes,
+    exceptions,
+    modeling_error_messages,
+)
+from snowflake.ml._internal.utils import identifier, snowpark_dataframe_utils
+from snowflake.ml._internal.utils.query_result_checker import SqlResultValidator
+from snowflake.ml._internal.utils.temp_file_utils import (
+    cleanup_temp_files,
+    get_temp_file_path,
+)
+from snowflake.ml.modeling._internal.model_specifications import (
+    ModelSpecifications,
+    ModelSpecificationsBuilder,
+)
+from snowflake.snowpark import DataFrame, Session, exceptions as snowpark_exceptions
+from snowflake.snowpark._internal.utils import (
+    TempObjectType,
+    random_name_for_temp_object,
+)
+from snowflake.snowpark.functions import sproc
+from snowflake.snowpark.stored_procedure import StoredProcedure
+
+cp.register_pickle_by_value(inspect.getmodule(get_temp_file_path))
+cp.register_pickle_by_value(inspect.getmodule(identifier.get_inferred_name))
+
+_PROJECT = "ModelDevelopment"
+
+
+class SnowparkModelTrainer:
+    """
+    A class for training models on Snowflake data using the Sproc.
+
+    TODO (snandamuri): Introduce the concept of executor that would take the training function
+    and execute it on the target environments like, local, Snowflake warehouse, or SPCS, etc.
+    """
+
+    def __init__(
+        self,
+        estimator: object,
+        dataset: DataFrame,
+        session: Session,
+        input_cols: List[str],
+        label_cols: Optional[List[str]],
+        sample_weight_col: Optional[str],
+        autogenerated: bool = False,
+        subproject: str = "",
+    ) -> None:
+        """
+        Initializes the SnowparkModelTrainer with a model, a Snowpark DataFrame, feature, and label column names.
+
+        Args:
+            estimator: SKLearn compatible estimator or transformer object.
+            dataset: The dataset used for training the model.
+            session: Snowflake session object to be used for training.
+            input_cols: The name(s) of one or more columns in a DataFrame containing a feature to be used for training.
+            label_cols: The name(s) of one or more columns in a DataFrame representing the target variable(s) to learn.
+            sample_weight_col: The column name representing the weight of training examples.
+            autogenerated: A boolean denoting if the trainer is being used by autogenerated code or not.
+            subproject: subproject name to be used in telemetry.
+        """
+        self.estimator = estimator
+        self.dataset = dataset
+        self.session = session
+        self.input_cols = input_cols
+        self.label_cols = label_cols
+        self.sample_weight_col = sample_weight_col
+        self._autogenerated = autogenerated
+        self._subproject = subproject
+        self._class_name = estimator.__class__.__name__
+
+    def _create_temp_stage(self) -> str:
+        """
+        Creates temporary stage.
+
+        Returns:
+            Temp stage name.
+        """
+        # Create temp stage to upload pickled model file.
+        transform_stage_name = random_name_for_temp_object(TempObjectType.STAGE)
+        stage_creation_query = f"CREATE OR REPLACE TEMPORARY STAGE {transform_stage_name};"
+        SqlResultValidator(session=self.session, query=stage_creation_query).has_dimensions(
+            expected_rows=1, expected_cols=1
+        ).validate()
+        return transform_stage_name
+
+    def _upload_model_to_stage(self, stage_name: str) -> Tuple[str, str]:
+        """
+        Util method to pickle and upload the model to a temp Snowflake stage.
+
+        Args:
+            stage_name: Stage name to save model.
+
+        Returns:
+            a tuple containing stage file paths for pickled input model for training and location to store trained
+            models(response from training sproc).
+        """
+        # Create a temp file and dump the transform to that file.
+        local_transform_file_name = get_temp_file_path()
+        with open(local_transform_file_name, mode="w+b") as local_transform_file:
+            cp.dump(self.estimator, local_transform_file)
+
+        # Use posixpath to construct stage paths
+        stage_transform_file_name = posixpath.join(stage_name, os.path.basename(local_transform_file_name))
+        stage_result_file_name = posixpath.join(stage_name, os.path.basename(local_transform_file_name))
+
+        statement_params = telemetry.get_function_usage_statement_params(
+            project=_PROJECT,
+            subproject=self._subproject,
+            function_name=telemetry.get_statement_params_full_func_name(inspect.currentframe(), self._class_name),
+            api_calls=[sproc],
+            custom_tags=dict([("autogen", True)]) if self._autogenerated else None,
+        )
+        # Put locally serialized transform on stage.
+        self.session.file.put(
+            local_transform_file_name,
+            stage_transform_file_name,
+            auto_compress=False,
+            overwrite=True,
+            statement_params=statement_params,
+        )
+
+        cleanup_temp_files([local_transform_file_name])
+        return (stage_transform_file_name, stage_result_file_name)
+
+    def _fetch_model_from_stage(self, dir_path: str, file_name: str, statement_params: Dict[str, str]) -> object:
+        """
+        Downloads the serialized model from a stage location and unpickels it.
+
+        Args:
+            dir_path: Stage directory path where results are stored.
+            file_name: File name with in the directory where results are stored.
+            statement_params: Statement params to be attached to the SQL queries issue form this method.
+
+        Returns:
+            Deserialized model object.
+        """
+        local_result_file_name = get_temp_file_path()
+        self.session.file.get(
+            posixpath.join(dir_path, file_name),
+            local_result_file_name,
+            statement_params=statement_params,
+        )
+
+        with open(os.path.join(local_result_file_name, file_name), mode="r+b") as result_file_obj:
+            fit_estimator = cp.load(result_file_obj)
+
+        cleanup_temp_files([local_result_file_name])
+        return fit_estimator
+
+    def _build_fit_wrapper_sproc(
+        self,
+        model_spec: ModelSpecifications,
+    ) -> Callable[[Any, List[str], str, str, List[str], List[str], Optional[str], Dict[str, str]], str]:
+        """
+        Constructs and returns a python stored procedure function to be used for training model.
+
+        Args:
+            model_spec: ModelSpecifications object that contains model specific information
+                like required imports, package dependencies, etc.
+
+        Returns:
+            A callable that can be registered as a stored procedure.
+        """
+        imports = model_spec.imports  # In order for the sproc to not resolve this reference in snowflake.ml
+
+        def fit_wrapper_function(
+            session: Session,
+            sql_queries: List[str],
+            stage_transform_file_name: str,
+            stage_result_file_name: str,
+            input_cols: List[str],
+            label_cols: List[str],
+            sample_weight_col: Optional[str],
+            statement_params: Dict[str, str],
+        ) -> str:
+            import inspect
+            import os
+
+            import cloudpickle as cp
+            import pandas as pd
+
+            for import_name in imports:
+                importlib.import_module(import_name)
+
+            # Execute snowpark queries and obtain the results as pandas dataframe
+            # NB: this implies that the result data must fit into memory.
+            for query in sql_queries[:-1]:
+                _ = session.sql(query).collect(statement_params=statement_params)
+            sp_df = session.sql(sql_queries[-1])
+            df: pd.DataFrame = sp_df.to_pandas(statement_params=statement_params)
+            df.columns = sp_df.columns
+
+            local_transform_file_name = get_temp_file_path()
+
+            session.file.get(stage_transform_file_name, local_transform_file_name, statement_params=statement_params)
+
+            local_transform_file_path = os.path.join(
+                local_transform_file_name, os.listdir(local_transform_file_name)[0]
+            )
+            with open(local_transform_file_path, mode="r+b") as local_transform_file_obj:
+                estimator = cp.load(local_transform_file_obj)
+
+            argspec = inspect.getfullargspec(estimator.fit)
+            args = {"X": df[input_cols]}
+            if label_cols:
+                label_arg_name = "Y" if "Y" in argspec.args else "y"
+                args[label_arg_name] = df[label_cols].squeeze()
+
+            if sample_weight_col is not None and "sample_weight" in argspec.args:
+                args["sample_weight"] = df[sample_weight_col].squeeze()
+
+            estimator.fit(**args)
+
+            local_result_file_name = get_temp_file_path()
+
+            with open(local_result_file_name, mode="w+b") as local_result_file_obj:
+                cp.dump(estimator, local_result_file_obj)
+
+            session.file.put(
+                local_result_file_name,
+                stage_result_file_name,
+                auto_compress=False,
+                overwrite=True,
+                statement_params=statement_params,
+            )
+
+            # Note: you can add something like  + "|" + str(df) to the return string
+            # to pass debug information to the caller.
+            return str(os.path.basename(local_result_file_name))
+
+        return fit_wrapper_function
+
+    def _get_fit_wrapper_sproc(self, statement_params: Dict[str, str]) -> StoredProcedure:
+        # If the sproc already exists, don't register.
+        if not hasattr(self.session, "_FIT_WRAPPER_SPROCS"):
+            self.session._FIT_WRAPPER_SPROCS: Dict[str, StoredProcedure] = {}  # type: ignore[attr-defined, misc]
+
+        model_spec = ModelSpecificationsBuilder.build(model=self.estimator)
+        fit_sproc_key = model_spec.__class__.__name__
+        if fit_sproc_key in self.session._FIT_WRAPPER_SPROCS:  # type: ignore[attr-defined]
+            fit_sproc: StoredProcedure = self.session._FIT_WRAPPER_SPROCS[fit_sproc_key]  # type: ignore[attr-defined]
+            return fit_sproc
+
+        fit_sproc_name = random_name_for_temp_object(TempObjectType.PROCEDURE)
+
+        fit_wrapper_sproc = self.session.sproc.register(
+            func=self._build_fit_wrapper_sproc(model_spec=model_spec),
+            is_permanent=False,
+            name=fit_sproc_name,
+            packages=["snowflake-snowpark-python"] + model_spec.pkgDependencies,  # type: ignore[arg-type]
+            replace=True,
+            session=self.session,
+            statement_params=statement_params,
+        )
+
+        self.session._FIT_WRAPPER_SPROCS[fit_sproc_key] = fit_wrapper_sproc  # type: ignore[attr-defined]
+
+        return fit_wrapper_sproc
+
+    def train(self) -> object:
+        """
+        Trains the model by pushing down the compute into Snowflake using stored procedures.
+
+        Returns:
+            Trained model
+
+        Raises:
+            e: Raises an exception if any of Snowflake operations fail because of any reason.
+            SnowflakeMLException: Know exception are caught and rethrow with more detailed error message.
+        """
+        dataset = snowpark_dataframe_utils.cast_snowpark_dataframe_column_types(self.dataset)
+
+        # TODO(snandamuri) : Handle the already in a stored procedure case in the in builder.
+
+        # Extract query that generated the dataframe. We will need to pass it to the fit procedure.
+        queries = dataset.queries["queries"]
+
+        transform_stage_name = self._create_temp_stage()
+        (stage_transform_file_name, stage_result_file_name) = self._upload_model_to_stage(
+            stage_name=transform_stage_name
+        )
+
+        # Call fit sproc
+        statement_params = telemetry.get_function_usage_statement_params(
+            project=_PROJECT,
+            subproject=self._subproject,
+            function_name=telemetry.get_statement_params_full_func_name(inspect.currentframe(), self._class_name),
+            api_calls=[Session.call],
+            custom_tags=dict([("autogen", True)]) if self._autogenerated else None,
+        )
+
+        fit_wrapper_sproc = self._get_fit_wrapper_sproc(statement_params=statement_params)
+
+        try:
+            sproc_export_file_name: str = fit_wrapper_sproc(
+                self.session,
+                queries,
+                stage_transform_file_name,
+                stage_result_file_name,
+                self.input_cols,
+                self.label_cols,
+                self.sample_weight_col,
+                statement_params,
+            )
+        except snowpark_exceptions.SnowparkClientException as e:
+            if "fit() missing 1 required positional argument: 'y'" in str(e):
+                raise exceptions.SnowflakeMLException(
+                    error_code=error_codes.NOT_FOUND,
+                    original_exception=RuntimeError(modeling_error_messages.ATTRIBUTE_NOT_SET.format("label_cols")),
+                ) from e
+            raise e
+
+        if "|" in sproc_export_file_name:
+            fields = sproc_export_file_name.strip().split("|")
+            sproc_export_file_name = fields[0]
+
+        return self._fetch_model_from_stage(
+            dir_path=stage_result_file_name,
+            file_name=sproc_export_file_name,
+            statement_params=statement_params,
+        )