kumoai 2.13.0.dev202512031731__cp312-cp312-macosx_11_0_arm64.whl → 2.14.0.dev202512181731__cp312-cp312-macosx_11_0_arm64.whl

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

@@ -0,0 +1,315 @@
+from typing import TYPE_CHECKING, Literal
+
+import numpy as np
+import pandas as pd
+from kumoapi.pquery import ValidatedPredictiveQuery
+
+from kumoai.experimental.rfm.backend.local import LocalGraphStore
+from kumoai.experimental.rfm.base import Sampler, SamplerOutput
+from kumoai.experimental.rfm.pquery import PQueryPandasExecutor
+from kumoai.utils import ProgressLogger
+
+if TYPE_CHECKING:
+    from kumoai.experimental.rfm import Graph
+
+
+class LocalSampler(Sampler):
+    def __init__(
+        self,
+        graph: 'Graph',
+        verbose: bool | ProgressLogger = True,
+    ) -> None:
+        super().__init__(graph=graph, verbose=verbose)
+
+        import kumoai.kumolib as kumolib
+
+        self._graph_store = LocalGraphStore(graph, verbose)
+        self._graph_sampler = kumolib.NeighborSampler(
+            list(self.table_stype_dict.keys()),
+            self.edge_types,
+            {
+                '__'.join(edge_type): colptr
+                for edge_type, colptr in self._graph_store.colptr_dict.items()
+            },
+            {
+                '__'.join(edge_type): row
+                for edge_type, row in self._graph_store.row_dict.items()
+            },
+            self._graph_store.time_dict,
+        )
+
+    def _get_min_max_time_dict(
+        self,
+        table_names: list[str],
+    ) -> dict[str, tuple[pd.Timestamp, pd.Timestamp]]:
+        return {
+            key: value
+            for key, value in self._graph_store.min_max_time_dict.items()
+            if key in table_names
+        }
+
+    def _sample_subgraph(
+        self,
+        entity_table_name: str,
+        entity_pkey: pd.Series,
+        anchor_time: pd.Series | Literal['entity'],
+        columns_dict: dict[str, set[str]],
+        num_neighbors: list[int],
+    ) -> SamplerOutput:
+
+        index = self._graph_store.get_node_id(entity_table_name, entity_pkey)
+
+        if isinstance(anchor_time, pd.Series):
+            time = anchor_time.astype(int).to_numpy() // 1000**3  # to seconds
+        else:
+            assert anchor_time == 'entity'
+            time = self._graph_store.time_dict[entity_table_name][index]
+
+        (
+            row_dict,
+            col_dict,
+            node_dict,
+            batch_dict,
+            num_sampled_nodes_dict,
+            num_sampled_edges_dict,
+        ) = self._graph_sampler.sample(
+            {
+                '__'.join(edge_type): num_neighbors
+                for edge_type in self.edge_types
+            },
+            {},
+            entity_table_name,
+            index,
+            time,
+        )
+
+        df_dict: dict[str, pd.DataFrame] = {}
+        inverse_dict: dict[str, np.ndarray] = {}
+        for table_name, node in node_dict.items():
+            df = self._graph_store.df_dict[table_name]
+            columns = columns_dict[table_name]
+            if self.end_time_column_dict.get(table_name, None) in columns:
+                df = df.iloc[node]
+            elif len(columns) == 0:
+                df = df.iloc[node]
+            else:
+                # Only store unique rows in `df` above a certain threshold:
+                unique_node, inverse = np.unique(node, return_inverse=True)
+                if len(node) > 1.05 * len(unique_node):
+                    df = df.iloc[unique_node]
+                    inverse_dict[table_name] = inverse
+                else:
+                    df = df.iloc[node]
+            df = df.reset_index(drop=True)
+            df = df[list(columns)]
+            df_dict[table_name] = df
+
+        num_sampled_nodes_dict = {
+            table_name: num_sampled_nodes.tolist()
+            for table_name, num_sampled_nodes in
+            num_sampled_nodes_dict.items()
+        }
+
+        row_dict = {
+            edge_type: row_dict['__'.join(edge_type)]
+            for edge_type in self.edge_types
+        }
+        col_dict = {
+            edge_type: col_dict['__'.join(edge_type)]
+            for edge_type in self.edge_types
+        }
+        num_sampled_edges_dict = {
+            edge_type: num_sampled_edges_dict['__'.join(edge_type)].tolist()
+            for edge_type in self.edge_types
+        }
+
+        return SamplerOutput(
+            anchor_time=time * 1000**3,  # to nanoseconds
+            df_dict=df_dict,
+            inverse_dict=inverse_dict,
+            batch_dict=batch_dict,
+            num_sampled_nodes_dict=num_sampled_nodes_dict,
+            row_dict=row_dict,
+            col_dict=col_dict,
+            num_sampled_edges_dict=num_sampled_edges_dict,
+        )
+
+    def _sample_entity_table(
+        self,
+        table_name: str,
+        columns: set[str],
+        num_rows: int,
+        random_seed: int | None = None,
+    ) -> pd.DataFrame:
+        pkey_map = self._graph_store.pkey_map_dict[table_name]
+        if len(pkey_map) > num_rows:
+            pkey_map = pkey_map.sample(
+                n=num_rows,
+                random_state=random_seed,
+                ignore_index=True,
+            )
+        df = self._graph_store.df_dict[table_name]
+        df = df.iloc[pkey_map['arange']][list(columns)]
+        return df
+
+    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,
+            pkey=entity_df[self.primary_key_dict[query.entity_table]],
+            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,
+            pkey=entity_df[self.primary_key_dict[query.entity_table]],
+            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
+
+    # Helper Methods ##########################################################
+
+    def _sample_target_set(
+        self,
+        query: ValidatedPredictiveQuery,
+        pkey: pd.Series,
+        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]:
+
+        num_hops = 1 if len(time_offset_dict) > 0 else 0
+        num_neighbors_dict: dict[str, list[int]] = {}
+        unix_time_offset_dict: dict[str, list[list[int | None]]] = {}
+        for edge_type, (start, end) in time_offset_dict.items():
+            unix_time_offset_dict['__'.join(edge_type)] = [[
+                date_offset_to_seconds(start) if start is not None else None,
+                date_offset_to_seconds(end),
+            ]]
+        for edge_type in set(self.edge_types) - set(time_offset_dict.keys()):
+            num_neighbors_dict['__'.join(edge_type)] = [0] * num_hops
+
+        if anchor_time.dtype != 'datetime64[ns]':
+            anchor_time = anchor_time.astype('datetime64')
+
+        count = 0
+        ys: list[pd.Series] = []
+        mask = np.full(len(index), False, dtype=bool)
+        for start in range(0, len(index), batch_size):
+            subset = pkey.iloc[index[start:start + batch_size]]
+            time = anchor_time.iloc[start:start + batch_size]
+
+            _, _, node_dict, batch_dict, _, _ = self._graph_sampler.sample(
+                num_neighbors_dict,
+                unix_time_offset_dict,
+                query.entity_table,
+                self._graph_store.get_node_id(query.entity_table, subset),
+                time.astype(int).to_numpy() // 1000**3,  # to seconds
+            )
+
+            feat_dict: dict[str, pd.DataFrame] = {}
+            time_dict: dict[str, pd.Series] = {}
+            for table_name, columns in columns_dict.items():
+                df = self._graph_store.df_dict[table_name]
+                df = df.iloc[node_dict[table_name]].reset_index(drop=True)
+                df = df[list(columns)]
+                feat_dict[table_name] = df
+
+                time_column = self.time_column_dict.get(table_name)
+                if time_column in columns:
+                    time_dict[table_name] = df[time_column]
+
+            y, _mask = PQueryPandasExecutor().execute(
+                query=query,
+                feat_dict=feat_dict,
+                time_dict=time_dict,
+                batch_dict=batch_dict,
+                anchor_time=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
+
+
+# Helper Functions ############################################################
+
+
+def date_offset_to_seconds(offset: pd.DateOffset) -> int:
+    r"""Convert a :class:`pandas.DateOffset` into a number of seconds.
+
+    .. note::
+        We are conservative and take months and years as their maximum value.
+        Additional values are then dropped in label computation where we know
+        the actual dates.
+    """
+    MAX_DAYS_IN_MONTH = 31
+    MAX_DAYS_IN_YEAR = 366
+
+    SECONDS_IN_MINUTE = 60
+    SECONDS_IN_HOUR = 60 * SECONDS_IN_MINUTE
+    SECONDS_IN_DAY = 24 * SECONDS_IN_HOUR
+
+    total_sec = 0
+    multiplier = getattr(offset, 'n', 1)  # The multiplier (if present).
+
+    for attr, value in offset.__dict__.items():
+        if value is None or value == 0:
+            continue
+        scaled_value = value * multiplier
+        if attr == 'years':
+            total_sec += scaled_value * MAX_DAYS_IN_YEAR * SECONDS_IN_DAY
+        elif attr == 'months':
+            total_sec += scaled_value * MAX_DAYS_IN_MONTH * SECONDS_IN_DAY
+        elif attr == 'days':
+            total_sec += scaled_value * SECONDS_IN_DAY
+        elif attr == 'hours':
+            total_sec += scaled_value * SECONDS_IN_HOUR
+        elif attr == 'minutes':
+            total_sec += scaled_value * SECONDS_IN_MINUTE
+        elif attr == 'seconds':
+            total_sec += scaled_value
+
+    return total_sec

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

@@ -1,8 +1,10 @@
-from typing import List, Optional
+import warnings
+from typing import cast
 
 import pandas as pd
+from kumoapi.model_plan import MissingType
 
-from kumoai.experimental.rfm.base import SourceColumn, SourceForeignKey, Table
+from kumoai.experimental.rfm.base import DataBackend, SourceColumn, Table
 from kumoai.experimental.rfm.infer import infer_dtype
 
 
@@ -51,9 +53,9 @@ class LocalTable(Table):
         self,
         df: pd.DataFrame,
         name: str,
-        primary_key: Optional[str] = None,
-        time_column: Optional[str] = None,
-        end_time_column: Optional[str] = None,
+        primary_key: MissingType | str | None = MissingType.VALUE,
+        time_column: str | None = None,
+        end_time_column: str | None = None,
     ) -> None:
 
         if df.empty:
@@ -75,21 +77,36 @@ class LocalTable(Table):
             end_time_column=end_time_column,
         )
 
-    def _get_source_columns(self) -> List[SourceColumn]:
-        return [
-            SourceColumn(
+    @property
+    def backend(self) -> DataBackend:
+        return cast(DataBackend, DataBackend.LOCAL)
+
+    def _get_source_columns(self) -> list[SourceColumn]:
+        source_columns: list[SourceColumn] = []
+        for column in self._data.columns:
+            ser = self._data[column]
+            try:
+                dtype = infer_dtype(ser)
+            except Exception:
+                warnings.warn(f"Data type inference for column '{column}' in "
+                              f"table '{self.name}' failed. Consider changing "
+                              f"the data type of the column to use it within "
+                              f"this table.")
+                continue
+
+            source_column = SourceColumn(
                 name=column,
-                dtype=infer_dtype(self._data[column]),
+                dtype=dtype,
                 is_primary_key=False,
                 is_unique_key=False,
-            ) for column in self._data.columns
-        ]
+                is_nullable=True,
+            )
+            source_columns.append(source_column)
 
-    def _get_source_foreign_keys(self) -> List[SourceForeignKey]:
-        return []
+        return source_columns
 
     def _get_sample_df(self) -> pd.DataFrame:
         return self._data
 
-    def _get_num_rows(self) -> Optional[int]:
+    def _get_num_rows(self) -> int | None:
         return len(self._data)

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

@@ -27,9 +27,11 @@ def connect(**kwargs: Any) -> Connection:
 
 
 from .table import SnowTable  # noqa: E402
+from .sampler import SnowSampler  # noqa: E402
 
 __all__ = [
     'connect',
     'Connection',
     'SnowTable',
+    'SnowSampler',
 ]

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

@@ -0,0 +1,252 @@
+import json
+from collections.abc import Iterator
+from contextlib import contextmanager
+
+import numpy as np
+import pandas as pd
+import pyarrow as pa
+from kumoapi.pquery import ValidatedPredictiveQuery
+
+from kumoai.experimental.rfm.backend.snow import Connection
+from kumoai.experimental.rfm.base import SQLSampler
+from kumoai.experimental.rfm.pquery import PQueryPandasExecutor
+from kumoai.utils import quote_ident
+
+
+@contextmanager
+def paramstyle(connection: Connection, style: str = 'qmark') -> Iterator[None]:
+    _style = connection._paramstyle
+    connection._paramstyle = style
+    yield
+    connection._paramstyle = _style
+
+
+class SnowSampler(SQLSampler):
+    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 paramstyle(self._connection), self._connection.cursor() as cursor:
+            cursor.execute(sql, table_names)
+            rows = cursor.fetchall()
+        for table_name, _min, _max in rows:
+            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 Snowflake does support `SEED` only as part of `SYSTEM` sampling.
+        num_rows = min(num_rows, 1_000_000)  # Snowflake's upper limit.
+
+        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")
+
+        sql = (f"SELECT {', '.join(quote_ident(col) for col in columns)}\n"
+               f"FROM {self.fqn_dict[table_name]}\n"
+               f"SAMPLE ROW ({num_rows} ROWS)")
+        if len(filters) > 0:
+            sql += f"\nWHERE{' AND'.join(filters)}"
+
+        with self._connection.cursor() as cursor:
+            # NOTE This may return duplicate primary keys. This is okay.
+            cursor.execute(sql)
+            table = cursor.fetch_arrow_all()
+
+        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]:
+
+        # NOTE For Snowflake, we execute everything at once to pay minimal
+        # query initialization costs.
+        index = np.concatenate([train_index, test_index])
+        time = pd.concat([train_time, test_time], axis=0, ignore_index=True)
+
+        entity_df = entity_df.iloc[index].reset_index(drop=True)
+
+        feat_dict: dict[str, pd.DataFrame] = {query.entity_table: entity_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] = entity_df[time_column]
+        batch_dict: dict[str, np.ndarray] = {
+            query.entity_table: np.arange(len(entity_df)),
+        }
+        for edge_type, (min_offset, max_offset) 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=entity_df[self.primary_key_dict[query.entity_table]],
+                anchor_time=time,
+                min_offset=min_offset,
+                max_offset=max_offset,
+                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=time,
+            num_forecasts=query.num_forecasts,
+        )
+
+        train_mask = mask[:len(train_index)]
+        test_mask = mask[len(train_index):]
+
+        boundary = int(train_mask.sum())
+        train_y = y.iloc[:boundary]
+        test_y = y.iloc[boundary:].reset_index(drop=True)
+
+        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]
+        source_table = self.source_table_dict[table_name]
+
+        payload = json.dumps(list(pkey))
+
+        sql = ("WITH TMP as (\n"
+               "  SELECT\n"
+               "    f.index as BATCH,\n")
+        if source_table[pkey_name].dtype.is_int():
+            sql += "    f.value::NUMBER as ID\n"
+        elif source_table[pkey_name].dtype.is_float():
+            sql += "    f.value::FLOAT as ID\n"
+        else:
+            sql += "    f.value::VARCHAR as ID\n"
+        sql += (f"  FROM TABLE(FLATTEN(INPUT => PARSE_JSON(?))) f\n"
+                f")\n"
+                f"SELECT TMP.BATCH as __BATCH__, "
+                f"{', '.join('ENT.' + quote_ident(col) for col in columns)}\n"
+                f"FROM TMP\n"
+                f"JOIN {self.fqn_dict[table_name]} ENT\n"
+                f"  ON ENT.{quote_ident(pkey_name)} = TMP.ID")
+
+        with paramstyle(self._connection), self._connection.cursor() as cursor:
+            cursor.execute(sql, (payload, ))
+            table = cursor.fetch_arrow_all()
+
+        # Remove any duplicated primary keys in post-processing:
+        tmp = table.append_column('__TMP__', pa.array(range(len(table))))
+        gb = tmp.group_by('__BATCH__').aggregate([('__TMP__', 'min')])
+        table = table.take(gb['__TMP___min'])
+
+        batch = table['__BATCH__'].cast(pa.int64()).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]:
+
+        end_time = anchor_time + max_offset
+        end_time = end_time.dt.strftime("%Y-%m-%d %H:%M:%S")
+        if min_offset is not None:
+            start_time = anchor_time + min_offset
+            start_time = start_time.dt.strftime("%Y-%m-%d %H:%M:%S")
+            payload = json.dumps(list(zip(pkey, end_time, start_time)))
+        else:
+            payload = json.dumps(list(zip(pkey, end_time)))
+
+        # Based on benchmarking, JSON payload is the fastest way to query by
+        # custom indices (compared to large `IN` clauses or temporary tables):
+        source_table = self.source_table_dict[table_name]
+        time_column = self.time_column_dict[table_name]
+        sql = ("WITH TMP as (\n"
+               "  SELECT\n"
+               "    f.index as BATCH,\n")
+        if source_table[fkey].dtype.is_int():
+            sql += "    f.value[0]::NUMBER as ID,\n"
+        elif source_table[fkey].dtype.is_float():
+            sql += "    f.value[0]::FLOAT as ID,\n"
+        else:
+            sql += "    f.value[0]::VARCHAR as ID,\n"
+        sql += "    f.value[1]::TIMESTAMP_NTZ as END_TIME"
+        if min_offset is not None:
+            sql += ",\n    f.value[2]::TIMESTAMP_NTZ as START_TIME"
+        sql += (f"\n"
+                f"  FROM TABLE(FLATTEN(INPUT => PARSE_JSON(?))) f\n"
+                f")\n"
+                f"SELECT TMP.BATCH as __BATCH__, "
+                f"{', '.join('FACT.' + quote_ident(col) for col in columns)}\n"
+                f"FROM 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_TIME")
+        if min_offset is not None:
+            sql += f"\n AND FACT.{quote_ident(time_column)} > TMP.START_TIME"
+
+        with paramstyle(self._connection), self._connection.cursor() as cursor:
+            cursor.execute(sql, (payload, ))
+            table = cursor.fetch_arrow_all()
+
+        batch = table['__BATCH__'].cast(pa.int64()).to_numpy()
+        table = table.remove_column(table.schema.get_field_index('__BATCH__'))
+
+        return self._sanitize(table_name, table), batch
+
+    def _sanitize(self, table_name: str, table: pa.table) -> pd.DataFrame:
+        return table.to_pandas(types_mapper=pd.ArrowDtype)