kumoai 2.12.0.dev202511031731__cp313-cp313-macosx_11_0_arm64.whl → 2.13.0.dev202512061731__cp313-cp313-macosx_11_0_arm64.whl

kumoai/experimental/rfm/base/sampler.py

@@ -0,0 +1,287 @@
+import copy
+import re
+from abc import ABC, abstractmethod
+from dataclasses import dataclass
+from typing import TYPE_CHECKING
+
+import numpy as np
+import pandas as pd
+from kumoapi.rfm.context import EdgeLayout, Link, Subgraph, Table
+from kumoapi.typing import Stype
+
+if TYPE_CHECKING:
+    from kumoai.experimental.rfm import Graph
+
+
+@dataclass
+class EdgeSpec:
+    num_neighbors: int | None = None
+    time_offsets: tuple[
+        pd.DateOffset | None,
+        pd.DateOffset,
+    ] | None = None
+
+    def __post_init__(self) -> None:
+        if (self.num_neighbors is None) == (self.time_offsets is None):
+            raise ValueError("Only one of 'num_neighbors' and 'time_offsets' "
+                             "must be provided")
+
+
+@dataclass
+class SamplerOutput:
+    df_dict: dict[str, pd.DataFrame]
+    inverse_dict: dict[str, np.ndarray]
+    batch_dict: dict[str, np.ndarray]
+    num_sampled_nodes_dict: dict[str, list[int]]
+    row_dict: dict[tuple[str, str, str], np.ndarray] | None = None
+    col_dict: dict[tuple[str, str, str], np.ndarray] | None = None
+    num_sampled_edges_dict: dict[tuple[str, str, str], list[int]] | None = None
+
+
+class Sampler(ABC):
+    def __init__(self, graph: 'Graph') -> None:
+        self._edge_types: list[tuple[str, str, str]] = []
+        for edge in graph.edges:
+            edge_type = (edge.src_table, edge.fkey, edge.dst_table)
+            self._edge_types.append(edge_type)
+            self._edge_types.append(Subgraph.rev_edge_type(edge_type))
+
+        self._primary_key_dict: dict[str, str] = {
+            table.name: table._primary_key
+            for table in graph.tables.values()
+            if table._primary_key is not None
+        }
+
+        self._time_column_dict: dict[str, str] = {
+            table.name: table._time_column
+            for table in graph.tables.values()
+            if table._time_column is not None
+        }
+
+        self._end_time_column_dict: dict[str, str] = {
+            table.name: table._end_time_column
+            for table in graph.tables.values()
+            if table._end_time_column is not None
+        }
+
+        foreign_keys = {(edge.src_table, edge.fkey) for edge in graph.edges}
+        self._table_stype_dict: dict[str, dict[str, Stype]] = {}
+        for table in graph.tables.values():
+            self._table_stype_dict[table.name] = {}
+            for column in table.columns:
+                if column == table.primary_key:
+                    continue
+                if (table.name, column.name) in foreign_keys:
+                    continue
+                self._table_stype_dict[table.name][column.name] = column.stype
+
+    @property
+    def edge_types(self) -> list[tuple[str, str, str]]:
+        return self._edge_types
+
+    @property
+    def primary_key_dict(self) -> dict[str, str]:
+        return self._primary_key_dict
+
+    @property
+    def time_column_dict(self) -> dict[str, str]:
+        return self._time_column_dict
+
+    @property
+    def end_time_column_dict(self) -> dict[str, str]:
+        return self._end_time_column_dict
+
+    @property
+    def table_stype_dict(self) -> dict[str, dict[str, Stype]]:
+        return self._table_stype_dict
+
+    def sample_subgraph(
+        self,
+        entity_table_names: tuple[str, ...],
+        entity_pkey: pd.Series,
+        anchor_time: pd.Series,
+        num_neighbors: list[int],
+        exclude_cols_dict: dict[str, list[str]] | None = None,
+    ) -> Subgraph:
+
+        edge_spec_dict: dict[tuple[str, str, str], list[EdgeSpec]] = {
+            edge_type: [EdgeSpec(value) for value in num_neighbors]
+            for edge_type in self.edge_types
+        }
+
+        # Exclude all columns that leak target information:
+        table_stype_dict: dict[str, dict[str, Stype]] = self._table_stype_dict
+        if exclude_cols_dict is not None:
+            table_stype_dict = copy.deepcopy(table_stype_dict)
+            for table_name, exclude_cols in exclude_cols_dict.items():
+                for column_name in exclude_cols:
+                    del table_stype_dict[table_name][column_name]
+
+        # Collect all columns being used as features:
+        column_spec_dict: dict[str, list[str]] = {
+            table_name: list(stype_dict.keys())
+            for table_name, stype_dict in table_stype_dict.items()
+        }
+        # Make sure to store primary key information for entity tables:
+        for table_name in entity_table_names:
+            column_spec_dict[table_name] = (
+                [self.primary_key_dict[table_name]] +
+                column_spec_dict[table_name])
+
+        if anchor_time.dtype != 'datetime64[ns]':
+            anchor_time = anchor_time.astype('datetime64[ns]')
+        out = self.sample(
+            entity_table_name=entity_table_names[0],
+            entity_pkey=entity_pkey,
+            anchor_time=anchor_time,
+            column_spec_dict=column_spec_dict,
+            edge_spec_dict=edge_spec_dict,
+            drop_duplicates=True,
+            return_edges=True,
+        )
+
+        subgraph = Subgraph(
+            anchor_time=anchor_time.astype(int).to_numpy(),
+            table_dict={},
+            link_dict={},
+        )
+
+        for table_name, batch in out.batch_dict.items():
+            if len(batch) == 0:
+                continue
+
+            primary_key = None
+            if table_name in entity_table_names:
+                primary_key = self.primary_key_dict.get(table_name, None)
+
+            df = out.df_dict[table_name].reset_index(drop=True)
+            if table_name in self.end_time_column_dict:
+                # Set end time to NaT for all values greater than anchor time:
+                end_time_column = self.end_time_column_dict[table_name]
+                ser = df[end_time_column]
+                if ser.dtype != 'datetime64[ns]':
+                    ser = ser.astype('datetime64[ns]')
+                mask = ser > anchor_time.iloc[batch]
+                ser.iloc[mask] = pd.NaT
+                df[end_time_column] = ser
+
+            stype_dict = table_stype_dict[table_name]
+            for column_name, stype in stype_dict.items():
+                if stype == Stype.text:
+                    df[column_name] = _normalize_text(df[column_name])
+
+            subgraph.table_dict[table_name] = Table(
+                df=df,
+                row=out.inverse_dict.get(table_name),
+                batch=batch,
+                num_sampled_nodes=out.num_sampled_nodes_dict[table_name],
+                stype_dict=stype_dict,
+                primary_key=primary_key,
+            )
+
+        assert out.row_dict is not None
+        assert out.col_dict is not None
+        assert out.num_sampled_edges_dict is not None
+        for edge_type in out.row_dict.keys():
+            row: np.ndarray | None = out.row_dict[edge_type]
+            col: np.ndarray | None = out.col_dict[edge_type]
+
+            if row is None or col is None or len(row) == 0:
+                continue
+
+            # Do not store reverse edge type if it is an exact replica:
+            rev_edge_type = Subgraph.rev_edge_type(edge_type)
+            if (rev_edge_type in subgraph.link_dict
+                    and np.array_equal(row, out.col_dict[rev_edge_type])
+                    and np.array_equal(col, out.row_dict[rev_edge_type])):
+                subgraph.link_dict[edge_type] = Link(
+                    layout=EdgeLayout.REV,
+                    row=None,
+                    col=None,
+                    num_sampled_edges=out.num_sampled_edges_dict[edge_type],
+                )
+                continue
+
+            # Do not store non-informative edges:
+            layout = EdgeLayout.COO
+            if np.array_equal(row, np.arange(len(row))):
+                row = None
+            if np.array_equal(col, np.arange(len(col))):
+                col = None
+
+            # Store in compressed representation if more efficient:
+            num_cols = subgraph.table_dict[edge_type[2]].num_rows
+            if col is not None and len(col) > num_cols + 1:
+                layout = EdgeLayout.CSC
+                colcount = np.bincount(col, minlength=num_cols)
+                col = np.empty(num_cols + 1, dtype=col.dtype)
+                col[0] = 0
+                np.cumsum(colcount, out=col[1:])
+
+            subgraph.link_dict[edge_type] = Link(
+                layout=layout,
+                row=row,
+                col=col,
+                num_sampled_edges=out.num_sampled_edges_dict[edge_type],
+            )
+
+        return subgraph
+
+    # Abstract Methods ########################################################
+
+    @abstractmethod
+    def sample(
+        self,
+        entity_table_name: str,
+        entity_pkey: pd.Series,
+        anchor_time: pd.Series,
+        column_spec_dict: dict[str, list[str]],
+        edge_spec_dict: dict[tuple[str, str, str], list[EdgeSpec]],
+        drop_duplicates: bool = False,
+        return_edges: bool = False,
+    ) -> SamplerOutput:
+        pass
+
+
+# Helper Functions ############################################################
+
+PUNCTUATION = re.compile(r"[\'\"\.,\(\)\!\?\;\:]")
+MULTISPACE = re.compile(r"\s+")
+
+
+def _normalize_text(
+    ser: pd.Series,
+    max_words: int | None = 50,
+) -> pd.Series:
+    r"""Normalizes text into a list of lower-case words.
+
+    Args:
+        ser: The :class:`pandas.Series` to normalize.
+        max_words: The maximum number of words to return.
+            This will auto-shrink any large text column to avoid blowing up
+            context size.
+    """
+    if len(ser) == 0 or pd.api.types.is_list_like(ser.iloc[0]):
+        return ser
+
+    def normalize_fn(line: str) -> list[str]:
+        line = PUNCTUATION.sub(" ", line)
+        line = re.sub(r"<br\s*/?>", " ", line)  # Handle <br /> or <br>
+        line = MULTISPACE.sub(" ", line)
+        words = line.split()
+        if max_words is not None:
+            words = words[:max_words]
+        return words
+
+    ser = ser.fillna('').astype(str)
+
+    if max_words is not None:
+        # We estimate the number of words as 5 characters + 1 space in an
+        # English text on average. We need this pre-filter here, as word
+        # splitting on a giant text can be very expensive:
+        ser = ser.str[:6 * max_words]
+
+    ser = ser.str.lower()
+    ser = ser.map(normalize_fn)
+
+    return ser

kumoai/experimental/rfm/base/source.py

@@ -0,0 +1,18 @@
+from dataclasses import dataclass
+
+from kumoapi.typing import Dtype
+
+
+@dataclass
+class SourceColumn:
+    name: str
+    dtype: Dtype
+    is_primary_key: bool
+    is_unique_key: bool
+
+
+@dataclass
+class SourceForeignKey:
+    name: str
+    dst_table: str
+    primary_key: str

kumoai/experimental/rfm/{local_table.py → base/table.py}

@@ -1,115 +1,35 @@
-from dataclasses import dataclass
-from typing import Any, Dict, List, Optional
+from abc import ABC, abstractmethod
+from collections import defaultdict
+from functools import cached_property
+from typing import Dict, List, Optional, Sequence, Set
 
 import pandas as pd
 from kumoapi.source_table import UnavailableSourceTable
 from kumoapi.table import Column as ColumnDefinition
 from kumoapi.table import TableDefinition
-from kumoapi.typing import Dtype, Stype
+from kumoapi.typing import Stype
 from typing_extensions import Self
 
-from kumoai import in_notebook
-from kumoai.experimental.rfm import utils
+from kumoai import in_notebook, in_snowflake_notebook
+from kumoai.experimental.rfm.base import Column, SourceColumn, SourceForeignKey
+from kumoai.experimental.rfm.infer import (
+    contains_categorical,
+    contains_id,
+    contains_multicategorical,
+    contains_timestamp,
+    infer_primary_key,
+    infer_time_column,
+)
 
 
-@dataclass(init=False, repr=False, eq=False)
-class Column:
-    stype: Stype
-
-    def __init__(
-        self,
-        name: str,
-        dtype: Dtype,
-        stype: Stype,
-        is_primary_key: bool = False,
-        is_time_column: bool = False,
-        is_end_time_column: bool = False,
-    ) -> None:
-        self._name = name
-        self._dtype = Dtype(dtype)
-        self._is_primary_key = is_primary_key
-        self._is_time_column = is_time_column
-        self._is_end_time_column = is_end_time_column
-        self.stype = Stype(stype)
-
-    @property
-    def name(self) -> str:
-        return self._name
-
-    @property
-    def dtype(self) -> Dtype:
-        return self._dtype
-
-    def __setattr__(self, key: str, val: Any) -> None:
-        if key == 'stype':
-            if isinstance(val, str):
-                val = Stype(val)
-            assert isinstance(val, Stype)
-            if not val.supports_dtype(self.dtype):
-                raise ValueError(f"Column '{self.name}' received an "
-                                 f"incompatible semantic type (got "
-                                 f"dtype='{self.dtype}' and stype='{val}')")
-            if self._is_primary_key and val != Stype.ID:
-                raise ValueError(f"Primary key '{self.name}' must have 'ID' "
-                                 f"semantic type (got '{val}')")
-            if self._is_time_column and val != Stype.timestamp:
-                raise ValueError(f"Time column '{self.name}' must have "
-                                 f"'timestamp' semantic type (got '{val}')")
-            if self._is_end_time_column and val != Stype.timestamp:
-                raise ValueError(f"End time column '{self.name}' must have "
-                                 f"'timestamp' semantic type (got '{val}')")
-
-        super().__setattr__(key, val)
-
-    def __hash__(self) -> int:
-        return hash((self.name, self.stype, self.dtype))
-
-    def __eq__(self, other: Any) -> bool:
-        if not isinstance(other, Column):
-            return False
-        return hash(self) == hash(other)
-
-    def __repr__(self) -> str:
-        return (f'{self.__class__.__name__}(name={self.name}, '
-                f'stype={self.stype}, dtype={self.dtype})')
-
-
-class LocalTable:
-    r"""A table backed by a :class:`pandas.DataFrame`.
-
-    A :class:`LocalTable` fully specifies the relevant metadata, *i.e.*
-    selected columns, column semantic types, primary keys and time columns.
-    :class:`LocalTable` is used to create a :class:`LocalGraph`.
-
-    .. code-block:: python
-
-        import pandas as pd
-        import kumoai.experimental.rfm as rfm
-
-        # Load data from a CSV file:
-        df = pd.read_csv("data.csv")
-
-        # Create a table from a `pandas.DataFrame` and infer its metadata ...
-        table = rfm.LocalTable(df, name="my_table").infer_metadata()
-
-        # ... or create a table explicitly:
-        table = rfm.LocalTable(
-            df=df,
-            name="my_table",
-            primary_key="id",
-            time_column="time",
-            end_time_column=None,
-        )
-
-        # Verify metadata:
-        table.print_metadata()
-
-        # Change the semantic type of a column:
-        table[column].stype = "text"
+class Table(ABC):
+    r"""A :class:`Table` fully specifies the relevant metadata of a single
+    table, *i.e.* its selected columns, data types, semantic types, primary
+    keys and time columns.
 
     Args:
-        df: The data frame to create the table from.
-        name: The name of the table.
+        name: The name of this table.
+        columns: The selected 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
@@ -117,46 +37,62 @@ class LocalTable:
     """
     def __init__(
         self,
-        df: pd.DataFrame,
         name: str,
+        columns: Optional[Sequence[str]] = None,
         primary_key: Optional[str] = None,
         time_column: Optional[str] = None,
         end_time_column: Optional[str] = None,
     ) -> None:
 
-        if df.empty:
-            raise ValueError("Data frame must have at least one row")
-        if isinstance(df.columns, pd.MultiIndex):
-            raise ValueError("Data frame must not have a multi-index")
-        if not df.columns.is_unique:
-            raise ValueError("Data frame must have unique column names")
-        if any(col == '' for col in df.columns):
-            raise ValueError("Data frame must have non-empty column names")
-
-        df = df.copy(deep=False)
-
-        self._data = df
         self._name = name
         self._primary_key: Optional[str] = None
         self._time_column: Optional[str] = None
         self._end_time_column: Optional[str] = None
 
+        if len(self._source_column_dict) == 0:
+            raise ValueError(f"Table '{name}' does not hold any column with "
+                             f"a supported data type")
+
+        primary_keys = [
+            column.name for column in self._source_column_dict.values()
+            if column.is_primary_key
+        ]
+        if len(primary_keys) == 1:  # NOTE No composite keys yet.
+            if primary_key is not None and primary_key != primary_keys[0]:
+                raise ValueError(f"Found duplicate primary key "
+                                 f"definition '{primary_key}' and "
+                                 f"'{primary_keys[0]}' in table '{name}'")
+            primary_key = primary_keys[0]
+
+        unique_keys = [
+            column.name for column in self._source_column_dict.values()
+            if column.is_unique_key
+        ]
+        if primary_key is None and len(unique_keys) == 1:
+            primary_key = unique_keys[0]
+
         self._columns: Dict[str, Column] = {}
-        for column_name in df.columns:
+        for column_name in columns or list(self._source_column_dict.keys()):
             self.add_column(column_name)
 
         if primary_key is not None:
+            if primary_key not in self:
+                self.add_column(primary_key)
             self.primary_key = primary_key
 
         if time_column is not None:
+            if time_column not in self:
+                self.add_column(time_column)
             self.time_column = time_column
 
         if end_time_column is not None:
+            if end_time_column not in self:
+                self.add_column(end_time_column)
             self.end_time_column = end_time_column
 
     @property
     def name(self) -> str:
-        r"""The name of the table."""
+        r"""The name of this table."""
         return self._name
 
     # Data column #############################################################
@@ -200,24 +136,35 @@ class LocalTable:
             raise KeyError(f"Column '{name}' already exists in table "
                            f"'{self.name}'")
 
-        if name not in self._data.columns:
-            raise KeyError(f"Column '{name}' does not exist in the underyling "
-                           f"data frame")
+        if name not in self._source_column_dict:
+            raise KeyError(f"Column '{name}' does not exist in the underlying "
+                           f"source table")
 
         try:
-            dtype = utils.to_dtype(self._data[name])
+            dtype = self._source_column_dict[name].dtype
         except Exception as e:
-            raise RuntimeError(f"Data type inference for column '{name}' in "
-                               f"table '{self.name}' failed. Consider "
-                               f"changing the data type of the column or "
-                               f"removing it from the table.") from e
+            raise RuntimeError(f"Could not obtain data type for column "
+                               f"'{name}' in table '{self.name}'. Change "
+                               f"the data type of the column in the source "
+                               f"table or remove it from the table.") from e
+
         try:
-            stype = utils.infer_stype(self._data[name], name, dtype)
+            ser = self._sample_df[name]
+            if contains_id(ser, name, dtype):
+                stype = Stype.ID
+            elif contains_timestamp(ser, name, dtype):
+                stype = Stype.timestamp
+            elif contains_multicategorical(ser, name, dtype):
+                stype = Stype.multicategorical
+            elif contains_categorical(ser, name, dtype):
+                stype = Stype.categorical
+            else:
+                stype = dtype.default_stype
         except Exception as e:
-            raise RuntimeError(f"Semantic type inference for column '{name}' "
-                               f"in table '{self.name}' failed. Consider "
-                               f"changing the data type of the column or "
-                               f"removing it from the table.") from e
+            raise RuntimeError(f"Could not obtain semantic type for column "
+                               f"'{name}' in table '{self.name}'. Change "
+                               f"the data type of the column in the source "
+                               f"table or remove it from the table.") from e
 
         self._columns[name] = Column(
             name=name,
@@ -432,12 +379,20 @@ class LocalTable:
         })
 
     def print_metadata(self) -> None:
-        r"""Prints the :meth:`~LocalTable.metadata` of the table."""
-        if in_notebook():
+        r"""Prints the :meth:`~metadata` of this table."""
+        num_rows_repr = ''
+        if self._num_rows is not None:
+            num_rows_repr = ' ({self._num_rows:,} rows)'
+
+        if in_snowflake_notebook():
+            import streamlit as st
+            md_repr = f"### 🏷️ Metadata of Table `{self.name}`{num_rows_repr}"
+            st.markdown(md_repr)
+            st.dataframe(self.metadata, hide_index=True)
+        elif in_notebook():
             from IPython.display import Markdown, display
-            display(
-                Markdown(f"### 🏷️ Metadata of Table `{self.name}` "
-                         f"({len(self._data):,} rows)"))
+            md_repr = f"### 🏷️ Metadata of Table `{self.name}`{num_rows_repr}"
+            display(Markdown(md_repr))
             df = self.metadata
             try:
                 if hasattr(df.style, 'hide'):
@@ -447,8 +402,7 @@ class LocalTable:
             except ImportError:
                 print(df.to_string(index=False))  # missing jinja2
         else:
-            print(f"🏷️ Metadata of Table '{self.name}' "
-                  f"({len(self._data):,} rows):")
+            print(f"🏷️ Metadata of Table '{self.name}'{num_rows_repr}")
             print(self.metadata.to_string(index=False))
 
     def infer_metadata(self, verbose: bool = True) -> Self:
@@ -478,9 +432,9 @@ class LocalTable:
                 column.name for column in self.columns if is_candidate(column)
             ]
 
-            if primary_key := utils.detect_primary_key(
+            if primary_key := infer_primary_key(
                     table_name=self.name,
-                    df=self._data,
+                    df=self._sample_df,
                     candidates=candidates,
             ):
                 self.primary_key = primary_key
@@ -493,7 +447,10 @@ class LocalTable:
                 if column.stype == Stype.timestamp
                 and column.name != self._end_time_column
             ]
-            if time_column := utils.detect_time_column(self._data, candidates):
+            if time_column := infer_time_column(
+                    df=self._sample_df,
+                    candidates=candidates,
+            ):
                 self.time_column = time_column
                 logs.append(f"time column '{time_column}'")
 
@@ -543,3 +500,46 @@ class LocalTable:
                 f'  time_column={self._time_column},\n'
                 f'  end_time_column={self._end_time_column},\n'
                 f')')
+
+    # Abstract Methods ########################################################
+
+    @cached_property
+    def _source_column_dict(self) -> Dict[str, SourceColumn]:
+        return {col.name: col for col in self._get_source_columns()}
+
+    @abstractmethod
+    def _get_source_columns(self) -> List[SourceColumn]:
+        pass
+
+    @cached_property
+    def _source_foreign_key_dict(self) -> Dict[str, SourceForeignKey]:
+        fkeys = self._get_source_foreign_keys()
+        # NOTE Drop all keys that link to different primary keys in the same
+        # table since we don't support composite keys yet:
+        table_pkeys: Dict[str, Set[str]] = defaultdict(set)
+        for fkey in fkeys:
+            table_pkeys[fkey.dst_table].add(fkey.primary_key)
+        return {
+            fkey.name: fkey
+            for fkey in fkeys if len(table_pkeys[fkey.dst_table]) == 1
+        }
+
+    @abstractmethod
+    def _get_source_foreign_keys(self) -> List[SourceForeignKey]:
+        pass
+
+    @cached_property
+    def _sample_df(self) -> pd.DataFrame:
+        return self._get_sample_df()
+
+    @abstractmethod
+    def _get_sample_df(self) -> pd.DataFrame:
+        pass
+
+    @cached_property
+    def _num_rows(self) -> Optional[int]:
+        return self._get_num_rows()
+
+    @abstractmethod
+    def _get_num_rows(self) -> Optional[int]:
+        pass