kumoai 2.13.0.dev202512040649__cp313-cp313-win_amd64.whl → 2.14.0.dev202512211732__cp313-cp313-win_amd64.whl

kumoai/experimental/rfm/backend/snow/table.py

@@ -1,20 +1,35 @@
 import re
-from typing import List, Optional, Sequence
+from collections.abc import Sequence
+from typing import cast
 
 import pandas as pd
+from kumoapi.model_plan import MissingType
 from kumoapi.typing import Dtype
 
-from kumoai.experimental.rfm.backend.sqlite import Connection
-from kumoai.experimental.rfm.base import SourceColumn, SourceForeignKey, Table
+from kumoai.experimental.rfm.backend.snow import Connection
+from kumoai.experimental.rfm.base import (
+    ColumnExpressionSpec,
+    ColumnExpressionType,
+    DataBackend,
+    SourceColumn,
+    SourceForeignKey,
+    SQLTable,
+)
+from kumoai.utils import quote_ident
 
 
-class SnowTable(Table):
+class SnowTable(SQLTable):
     r"""A table backed by a :class:`sqlite` database.
 
     Args:
         connection: The connection to a :class:`snowflake` database.
-        name: The name of this table.
-        columns: The selected columns of this table.
+        name: The logical name of this table.
+        source_name: The physical name of this table in the database. If set to
+            ``None``, ``name`` is being used.
+        database: The database.
+        schema: The schema.
+        columns: The selected physical columns of this table.
+        column_expressions: The logical columns of this table.
         primary_key: The name of the primary key of this table, if it exists.
         time_column: The name of the time column of this table, if it exists.
         end_time_column: The name of the end time column of this table, if it
@@ -24,17 +39,20 @@ class SnowTable(Table):
         self,
         connection: Connection,
         name: str,
+        source_name: str | None = None,
         database: str | None = None,
         schema: str | None = None,
-        columns: Optional[Sequence[str]] = None,
-        primary_key: Optional[str] = None,
-        time_column: Optional[str] = None,
-        end_time_column: Optional[str] = None,
+        columns: Sequence[str] | None = None,
+        column_expressions: Sequence[ColumnExpressionType] | None = None,
+        primary_key: MissingType | str | None = MissingType.VALUE,
+        time_column: str | None = None,
+        end_time_column: str | None = None,
     ) -> None:
 
         if database is not None and schema is None:
-            raise ValueError(f"Missing 'schema' for table '{name}' in "
-                             f"database '{database}'")
+            raise ValueError(f"Unspecified 'schema' for table "
+                             f"'{source_name or name}' in database "
+                             f"'{database}'")
 
         self._connection = connection
         self._database = database
@@ -42,47 +60,67 @@ class SnowTable(Table):
 
         super().__init__(
             name=name,
+            source_name=source_name,
             columns=columns,
+            column_expressions=column_expressions,
             primary_key=primary_key,
             time_column=time_column,
             end_time_column=end_time_column,
         )
 
+    @staticmethod
+    def to_dtype(snowflake_dtype: str | None) -> Dtype | None:
+        if snowflake_dtype is None:
+            return None
+        snowflake_dtype = snowflake_dtype.strip().upper()
+        # TODO 'NUMBER(...)' is not always an integer!
+        if snowflake_dtype.startswith('NUMBER'):
+            return Dtype.int
+        elif snowflake_dtype.startswith('VARCHAR'):
+            return Dtype.string
+        elif snowflake_dtype == 'FLOAT':
+            return Dtype.float
+        elif snowflake_dtype == 'BOOLEAN':
+            return Dtype.bool
+        elif re.search('DATE|TIMESTAMP', snowflake_dtype):
+            return Dtype.date
+        return None
+
+    @property
+    def backend(self) -> DataBackend:
+        return cast(DataBackend, DataBackend.SNOWFLAKE)
+
     @property
-    def fqn_name(self) -> str:
-        names: List[str] = []
+    def fqn(self) -> str:
+        r"""The fully-qualified quoted table name."""
+        names: list[str] = []
         if self._database is not None:
-            assert self._schema is not None
-            names.extend([self._database, self._schema])
-        elif self._schema is not None:
-            names.append(self._schema)
-        names.append(self._name)
-        return '.'.join(names)
-
-    def _get_source_columns(self) -> List[SourceColumn]:
-        source_columns: List[SourceColumn] = []
+            names.append(quote_ident(self._database))
+        if self._schema is not None:
+            names.append(quote_ident(self._schema))
+        return '.'.join(names + [quote_ident(self._source_name)])
+
+    def _get_source_columns(self) -> list[SourceColumn]:
+        source_columns: list[SourceColumn] = []
         with self._connection.cursor() as cursor:
             try:
-                cursor.execute(f"DESCRIBE TABLE {self.fqn_name}")
+                sql = f"DESCRIBE TABLE {self.fqn}"
+                cursor.execute(sql)
             except Exception as e:
-                raise ValueError(
-                    f"Table '{self.fqn_name}' does not exist") from e
+                names: list[str] = []
+                if self._database is not None:
+                    names.append(self._database)
+                if self._schema is not None:
+                    names.append(self._schema)
+                source_name = '.'.join(names + [self._source_name])
+                raise ValueError(f"Table '{source_name}' does not exist in "
+                                 f"the remote data backend") from e
 
             for row in cursor.fetchall():
-                column, type, _, _, _, is_pkey, is_unique = row[:7]
-
-                type = type.strip().upper()
-                if type.startswith('NUMBER'):
-                    dtype = Dtype.int
-                elif type.startswith('VARCHAR'):
-                    dtype = Dtype.string
-                elif type == 'FLOAT':
-                    dtype = Dtype.float
-                elif type == 'BOOLEAN':
-                    dtype = Dtype.bool
-                elif re.search('DATE|TIMESTAMP', type):
-                    dtype = Dtype.date
-                else:
+                column, type, _, null, _, is_pkey, is_unique, *_ = row
+
+                dtype = self.to_dtype(type)
+                if dtype is None:
                     continue
 
                 source_column = SourceColumn(
@@ -90,26 +128,42 @@ class SnowTable(Table):
                     dtype=dtype,
                     is_primary_key=is_pkey.strip().upper() == 'Y',
                     is_unique_key=is_unique.strip().upper() == 'Y',
+                    is_nullable=null.strip().upper() == 'Y',
                 )
                 source_columns.append(source_column)
 
         return source_columns
 
-    def _get_source_foreign_keys(self) -> List[SourceForeignKey]:
-        source_fkeys: List[SourceForeignKey] = []
+    def _get_source_foreign_keys(self) -> list[SourceForeignKey]:
+        source_fkeys: list[SourceForeignKey] = []
         with self._connection.cursor() as cursor:
-            cursor.execute(f"SHOW IMPORTED KEYS IN TABLE {self.fqn_name}")
+            sql = f"SHOW IMPORTED KEYS IN TABLE {self.fqn}"
+            cursor.execute(sql)
             for row in cursor.fetchall():
-                _, _, _, dst_table, pkey, _, _, _, fkey = row[:9]
+                _, _, _, dst_table, pkey, _, _, _, fkey, *_ = row
                 source_fkeys.append(SourceForeignKey(fkey, dst_table, pkey))
         return source_fkeys
 
-    def _get_sample_df(self) -> pd.DataFrame:
+    def _get_source_sample_df(self) -> pd.DataFrame:
         with self._connection.cursor() as cursor:
-            columns = ', '.join(self._source_column_dict.keys())
-            cursor.execute(f"SELECT {columns} FROM {self.fqn_name} LIMIT 1000")
+            columns = [quote_ident(col) for col in self._source_column_dict]
+            sql = f"SELECT {', '.join(columns)} FROM {self.fqn} LIMIT 1000"
+            cursor.execute(sql)
             table = cursor.fetch_arrow_all()
             return table.to_pandas(types_mapper=pd.ArrowDtype)
 
-    def _get_num_rows(self) -> Optional[int]:
+    def _get_num_rows(self) -> int | None:
         return None
+
+    def _get_expression_sample_df(
+        self,
+        specs: Sequence[ColumnExpressionSpec],
+    ) -> pd.DataFrame:
+        with self._connection.cursor() as cursor:
+            columns = [
+                f"{spec.expr} AS {quote_ident(spec.name)}" for spec in specs
+            ]
+            sql = f"SELECT {', '.join(columns)} FROM {self.fqn} LIMIT 1000"
+            cursor.execute(sql)
+            table = cursor.fetch_arrow_all()
+            return table.to_pandas(types_mapper=pd.ArrowDtype)

kumoai/experimental/rfm/backend/sqlite/__init__.py

@@ -1,5 +1,5 @@
 from pathlib import Path
-from typing import Any, TypeAlias, Union
+from typing import Any, TypeAlias
 
 try:
     import adbc_driver_sqlite.dbapi as adbc
@@ -11,7 +11,7 @@ except ImportError:
 Connection: TypeAlias = adbc.AdbcSqliteConnection
 
 
-def connect(uri: Union[str, Path, None] = None, **kwargs: Any) -> Connection:
+def connect(uri: str | Path | None = None, **kwargs: Any) -> Connection:
     r"""Opens a connection to a :class:`sqlite` database.
 
     uri: The path to the database file to be opened.
@@ -22,9 +22,11 @@ def connect(uri: Union[str, Path, None] = None, **kwargs: Any) -> Connection:
 
 
 from .table import SQLiteTable  # noqa: E402
+from .sampler import SQLiteSampler  # noqa: E402
 
 __all__ = [
     'connect',
     'Connection',
     'SQLiteTable',
+    'SQLiteSampler',
 ]

kumoai/experimental/rfm/backend/sqlite/sampler.py

@@ -0,0 +1,349 @@
+import warnings
+from collections import defaultdict
+from typing import TYPE_CHECKING
+
+import numpy as np
+import pandas as pd
+import pyarrow as pa
+from kumoapi.pquery import ValidatedPredictiveQuery
+from kumoapi.typing import Stype
+
+from kumoai.experimental.rfm.base import SQLSampler
+from kumoai.experimental.rfm.pquery import PQueryPandasExecutor
+from kumoai.utils import ProgressLogger, quote_ident
+
+if TYPE_CHECKING:
+    from kumoai.experimental.rfm import Graph
+
+
+class SQLiteSampler(SQLSampler):
+    def __init__(
+        self,
+        graph: 'Graph',
+        verbose: bool | ProgressLogger = True,
+        optimize: bool = False,
+    ) -> None:
+        super().__init__(graph=graph, verbose=verbose)
+
+        if optimize:
+            with self._connection.cursor() as cursor:
+                cursor.execute("PRAGMA temp_store = MEMORY")
+                cursor.execute("PRAGMA cache_size = -2000000")  # 2 GB
+
+        # Collect database indices to speed-up sampling:
+        index_dict: dict[str, set[tuple[str, ...]]] = defaultdict(set)
+        for table_name, primary_key in self.primary_key_dict.items():
+            source_table = self.source_table_dict[table_name]
+            if not source_table[primary_key].is_unique_key:
+                index_dict[table_name].add((primary_key, ))
+        for src_table_name, foreign_key, _ in graph.edges:
+            source_table = self.source_table_dict[src_table_name]
+            if source_table[foreign_key].is_unique_key:
+                pass
+            elif time_column := self.time_column_dict.get(src_table_name):
+                index_dict[src_table_name].add((foreign_key, time_column))
+            else:
+                index_dict[src_table_name].add((foreign_key, ))
+
+        # Only maintain missing indices:
+        with self._connection.cursor() as cursor:
+            for table_name in list(index_dict.keys()):
+                indices = index_dict[table_name]
+                sql = f"PRAGMA index_list({self.fqn_dict[table_name]})"
+                cursor.execute(sql)
+                for _, index_name, *_ in cursor.fetchall():
+                    sql = f"PRAGMA index_info({quote_ident(index_name)})"
+                    cursor.execute(sql)
+                    index = tuple(info[2] for info in sorted(
+                        cursor.fetchall(), key=lambda x: x[0]))
+                    indices.discard(index)
+                if len(indices) == 0:
+                    del index_dict[table_name]
+
+        num = sum(len(indices) for indices in index_dict.values())
+        index_repr = '1 index' if num == 1 else f'{num} indices'
+        num = len(index_dict)
+        table_repr = '1 table' if num == 1 else f'{num} tables'
+
+        if optimize and len(index_dict) > 0:
+            if not isinstance(verbose, ProgressLogger):
+                verbose = ProgressLogger.default(
+                    msg="Optimizing SQLite database",
+                    verbose=verbose,
+                )
+
+            with verbose as logger, self._connection.cursor() as cursor:
+                for table_name, indices in index_dict.items():
+                    for index in indices:
+                        name = f"kumo_index_{table_name}_{'_'.join(index)}"
+                        name = quote_ident(name)
+                        columns = ', '.join(quote_ident(v) for v in index)
+                        columns += ' DESC' if len(index) > 1 else ''
+                        sql = (f"CREATE INDEX IF NOT EXISTS {name}\n"
+                               f"ON {self.fqn_dict[table_name]}({columns})")
+                        cursor.execute(sql)
+                self._connection.commit()
+                logger.log(f"Created {index_repr} in {table_repr}")
+
+        elif len(index_dict) > 0:
+            warnings.warn(f"Missing {index_repr} in {table_repr} for optimal "
+                          f"database querying. For improving runtime, we "
+                          f"strongly suggest to create these indices by "
+                          f"instantiating KumoRFM via "
+                          f"`KumoRFM(graph, optimize=True)`.")
+
+    def _get_min_max_time_dict(
+        self,
+        table_names: list[str],
+    ) -> dict[str, tuple[pd.Timestamp, pd.Timestamp]]:
+        selects: list[str] = []
+        for table_name in table_names:
+            time_column = self.time_column_dict[table_name]
+            select = (f"SELECT\n"
+                      f"  ? as table_name,\n"
+                      f"  MIN({quote_ident(time_column)}) as min_date,\n"
+                      f"  MAX({quote_ident(time_column)}) as max_date\n"
+                      f"FROM {self.fqn_dict[table_name]}")
+            selects.append(select)
+        sql = "\nUNION ALL\n".join(selects)
+
+        out_dict: dict[str, tuple[pd.Timestamp, pd.Timestamp]] = {}
+        with self._connection.cursor() as cursor:
+            cursor.execute(sql, table_names)
+            for table_name, _min, _max in cursor.fetchall():
+                out_dict[table_name] = (
+                    pd.Timestamp.max if _min is None else pd.Timestamp(_min),
+                    pd.Timestamp.min if _max is None else pd.Timestamp(_max),
+                )
+        return out_dict
+
+    def _sample_entity_table(
+        self,
+        table_name: str,
+        columns: set[str],
+        num_rows: int,
+        random_seed: int | None = None,
+    ) -> pd.DataFrame:
+        # NOTE SQLite does not natively support passing a `random_seed`.
+
+        filters: list[str] = []
+        primary_key = self.primary_key_dict[table_name]
+        if self.source_table_dict[table_name][primary_key].is_nullable:
+            filters.append(f" {quote_ident(primary_key)} IS NOT NULL")
+        time_column = self.time_column_dict.get(table_name)
+        if (time_column is not None and
+                self.source_table_dict[table_name][time_column].is_nullable):
+            filters.append(f" {quote_ident(time_column)} IS NOT NULL")
+
+        # TODO Make this query more efficient - it does full table scan.
+        sql = (f"SELECT {', '.join(quote_ident(col) for col in columns)}\n"
+               f"FROM {self.fqn_dict[table_name]}")
+        if len(filters) > 0:
+            sql += f"\nWHERE{' AND'.join(filters)}"
+        sql += f"\nORDER BY RANDOM() LIMIT {num_rows}"
+
+        with self._connection.cursor() as cursor:
+            # NOTE This may return duplicate primary keys. This is okay.
+            cursor.execute(sql)
+            table = cursor.fetch_arrow_table()
+
+        return self._sanitize(table_name, table)
+
+    def _sample_target(
+        self,
+        query: ValidatedPredictiveQuery,
+        entity_df: pd.DataFrame,
+        train_index: np.ndarray,
+        train_time: pd.Series,
+        num_train_examples: int,
+        test_index: np.ndarray,
+        test_time: pd.Series,
+        num_test_examples: int,
+        columns_dict: dict[str, set[str]],
+        time_offset_dict: dict[
+            tuple[str, str, str],
+            tuple[pd.DateOffset | None, pd.DateOffset],
+        ],
+    ) -> tuple[pd.Series, np.ndarray, pd.Series, np.ndarray]:
+        train_y, train_mask = self._sample_target_set(
+            query=query,
+            entity_df=entity_df,
+            index=train_index,
+            anchor_time=train_time,
+            num_examples=num_train_examples,
+            columns_dict=columns_dict,
+            time_offset_dict=time_offset_dict,
+        )
+
+        test_y, test_mask = self._sample_target_set(
+            query=query,
+            entity_df=entity_df,
+            index=test_index,
+            anchor_time=test_time,
+            num_examples=num_test_examples,
+            columns_dict=columns_dict,
+            time_offset_dict=time_offset_dict,
+        )
+
+        return train_y, train_mask, test_y, test_mask
+
+    def _by_pkey(
+        self,
+        table_name: str,
+        pkey: pd.Series,
+        columns: set[str],
+    ) -> tuple[pd.DataFrame, np.ndarray]:
+        pkey_name = self.primary_key_dict[table_name]
+
+        tmp = pa.table([pa.array(pkey)], names=['id'])
+        tmp_name = f'tmp_{table_name}_{pkey_name}_{id(tmp)}'
+
+        if self.source_table_dict[table_name][pkey_name].is_unique_key:
+            sql = (f"SELECT tmp.rowid - 1 as __batch__, "
+                   f"{', '.join('ent.' + quote_ident(c) for c in columns)}\n"
+                   f"FROM {quote_ident(tmp_name)} tmp\n"
+                   f"JOIN {self.fqn_dict[table_name]} ent\n"
+                   f"  ON ent.{quote_ident(pkey_name)} = tmp.id")
+        else:
+            sql = (f"SELECT tmp.rowid - 1 as __batch__, "
+                   f"{', '.join('ent.' + quote_ident(c) for c in columns)}\n"
+                   f"FROM {quote_ident(tmp_name)} tmp\n"
+                   f"JOIN {self.fqn_dict[table_name]} ent\n"
+                   f"  ON ent.rowid = (\n"
+                   f"    SELECT rowid FROM {self.fqn_dict[table_name]}\n"
+                   f"    WHERE {quote_ident(pkey_name)} == tmp.id\n"
+                   f"    LIMIT 1\n"
+                   f")")
+
+        with self._connection.cursor() as cursor:
+            cursor.adbc_ingest(tmp_name, tmp, mode='replace')
+            cursor.execute(sql)
+            table = cursor.fetch_arrow_table()
+
+        batch = table['__batch__'].to_numpy()
+        table = table.remove_column(table.schema.get_field_index('__batch__'))
+
+        return table.to_pandas(), batch  # TODO Use `self._sanitize`.
+
+    # Helper Methods ##########################################################
+
+    def _by_time(
+        self,
+        table_name: str,
+        fkey: str,
+        pkey: pd.Series,
+        anchor_time: pd.Series,
+        min_offset: pd.DateOffset | None,
+        max_offset: pd.DateOffset,
+        columns: set[str],
+    ) -> tuple[pd.DataFrame, np.ndarray]:
+        # NOTE SQLite does not have a native datetime format. Currently, we
+        # assume timestamps are given as `TEXT` in `ISO-8601 UTC`:
+        tmp = pa.table([pa.array(pkey)], names=['id'])
+        end_time = anchor_time + max_offset
+        end_time = end_time.dt.strftime("%Y-%m-%d %H:%M:%S")
+        tmp = tmp.append_column('end', pa.array(end_time))
+        if min_offset is not None:
+            start_time = anchor_time + min_offset
+            start_time = start_time.dt.strftime("%Y-%m-%d %H:%M:%S")
+            tmp = tmp.append_column('start', pa.array(start_time))
+        tmp_name = f'tmp_{table_name}_{fkey}_{id(tmp)}'
+
+        time_column = self.time_column_dict[table_name]
+        sql = (f"SELECT tmp.rowid - 1 as __batch__, "
+               f"{', '.join('fact.' + quote_ident(col) for col in columns)}\n"
+               f"FROM {quote_ident(tmp_name)} tmp\n"
+               f"JOIN {self.fqn_dict[table_name]} fact\n"
+               f"  ON fact.{quote_ident(fkey)} = tmp.id\n"
+               f" AND fact.{quote_ident(time_column)} <= tmp.end")
+        if min_offset is not None:
+            sql += f"\n AND fact.{quote_ident(time_column)} > tmp.start"
+
+        with self._connection.cursor() as cursor:
+            cursor.adbc_ingest(tmp_name, tmp, mode='replace')
+            cursor.execute(sql)
+            table = cursor.fetch_arrow_table()
+
+        batch = table['__batch__'].to_numpy()
+        table = table.remove_column(table.schema.get_field_index('__batch__'))
+
+        return self._sanitize(table_name, table), batch
+
+    def _sample_target_set(
+        self,
+        query: ValidatedPredictiveQuery,
+        entity_df: pd.DataFrame,
+        index: np.ndarray,
+        anchor_time: pd.Series,
+        num_examples: int,
+        columns_dict: dict[str, set[str]],
+        time_offset_dict: dict[
+            tuple[str, str, str],
+            tuple[pd.DateOffset | None, pd.DateOffset],
+        ],
+        batch_size: int = 10_000,
+    ) -> tuple[pd.Series, np.ndarray]:
+
+        count = 0
+        ys: list[pd.Series] = []
+        mask = np.full(len(index), False, dtype=bool)
+        for start in range(0, len(index), batch_size):
+            df = entity_df.iloc[index[start:start + batch_size]]
+            time = anchor_time.iloc[start:start + batch_size]
+
+            feat_dict: dict[str, pd.DataFrame] = {query.entity_table: df}
+            time_dict: dict[str, pd.Series] = {}
+            time_column = self.time_column_dict.get(query.entity_table)
+            if time_column in columns_dict[query.entity_table]:
+                time_dict[query.entity_table] = df[time_column]
+            batch_dict: dict[str, np.ndarray] = {
+                query.entity_table: np.arange(len(df)),
+            }
+            for edge_type, (_min, _max) in time_offset_dict.items():
+                table_name, fkey, _ = edge_type
+                feat_dict[table_name], batch_dict[table_name] = self._by_time(
+                    table_name=table_name,
+                    fkey=fkey,
+                    pkey=df[self.primary_key_dict[query.entity_table]],
+                    anchor_time=time,
+                    min_offset=_min,
+                    max_offset=_max,
+                    columns=columns_dict[table_name],
+                )
+                time_column = self.time_column_dict.get(table_name)
+                if time_column in columns_dict[table_name]:
+                    time_dict[table_name] = feat_dict[table_name][time_column]
+
+            y, _mask = PQueryPandasExecutor().execute(
+                query=query,
+                feat_dict=feat_dict,
+                time_dict=time_dict,
+                batch_dict=batch_dict,
+                anchor_time=anchor_time,
+                num_forecasts=query.num_forecasts,
+            )
+            ys.append(y)
+            mask[start:start + batch_size] = _mask
+
+            count += len(y)
+            if count >= num_examples:
+                break
+
+        if len(ys) == 0:
+            y = pd.Series([], dtype=float)
+        elif len(ys) == 1:
+            y = ys[0]
+        else:
+            y = pd.concat(ys, axis=0, ignore_index=True)
+
+        return y, mask
+
+    def _sanitize(self, table_name: str, table: pa.table) -> pd.DataFrame:
+        df = table.to_pandas(types_mapper=pd.ArrowDtype)
+
+        stype_dict = self.table_stype_dict[table_name]
+        for column_name in df.columns:
+            if stype_dict.get(column_name) == Stype.timestamp:
+                df[column_name] = pd.to_datetime(df[column_name])
+
+        return df