|
| 1 | +# Copyright 2025 Google LLC |
| 2 | +# |
| 3 | +# Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | +# you may not use this file except in compliance with the License. |
| 5 | +# You may obtain a copy of the License at |
| 6 | +# |
| 7 | +# http://www.apache.org/licenses/LICENSE-2.0 |
| 8 | +# |
| 9 | +# Unless required by applicable law or agreed to in writing, software |
| 10 | +# distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | +# See the License for the specific language governing permissions and |
| 13 | +# limitations under the License. |
| 14 | +from __future__ import annotations |
| 15 | + |
| 16 | +import dataclasses |
| 17 | +import functools |
| 18 | +from typing import Any, Dict, Iterable, Optional, Sequence, TYPE_CHECKING |
| 19 | + |
| 20 | +from bigframes.core import blocks |
| 21 | +import bigframes.enums |
| 22 | + |
| 23 | +if TYPE_CHECKING: |
| 24 | + import bigframes.dataframe |
| 25 | + |
| 26 | + |
| 27 | +@dataclasses.dataclass(frozen=True) |
| 28 | +class InterchangeColumn: |
| 29 | + _dataframe: InterchangeDataFrame |
| 30 | + _pos: int |
| 31 | + |
| 32 | + @functools.cache |
| 33 | + def _arrow_column(self): |
| 34 | + # Conservatively downloads the whole underlying dataframe |
| 35 | + # This is much better if multiple columns end up being used, |
| 36 | + # but does incur a lot of overhead otherwise. |
| 37 | + return self._dataframe._arrow_dataframe().get_column(self._pos) |
| 38 | + |
| 39 | + def size(self) -> int: |
| 40 | + return self._arrow_column().size() |
| 41 | + |
| 42 | + @property |
| 43 | + def offset(self) -> int: |
| 44 | + return self._arrow_column().offset |
| 45 | + |
| 46 | + @property |
| 47 | + def dtype(self): |
| 48 | + return self._arrow_column().dtype |
| 49 | + |
| 50 | + @property |
| 51 | + def describe_categorical(self): |
| 52 | + raise TypeError(f"Column type {self.dtype} is not categorical") |
| 53 | + |
| 54 | + @property |
| 55 | + def describe_null(self): |
| 56 | + return self._arrow_column().describe_null |
| 57 | + |
| 58 | + @property |
| 59 | + def null_count(self): |
| 60 | + return self._arrow_column().null_count |
| 61 | + |
| 62 | + @property |
| 63 | + def metadata(self) -> Dict[str, Any]: |
| 64 | + return self._arrow_column().metadata |
| 65 | + |
| 66 | + def num_chunks(self) -> int: |
| 67 | + return self._arrow_column().num_chunks() |
| 68 | + |
| 69 | + def get_chunks(self, n_chunks: Optional[int] = None) -> Iterable: |
| 70 | + return self._arrow_column().get_chunks(n_chunks=n_chunks) |
| 71 | + |
| 72 | + def get_buffers(self): |
| 73 | + return self._arrow_column().get_buffers() |
| 74 | + |
| 75 | + |
| 76 | +@dataclasses.dataclass(frozen=True) |
| 77 | +class InterchangeDataFrame: |
| 78 | + """ |
| 79 | + Implements the dataframe interchange format. |
| 80 | +
|
| 81 | + Mostly implemented by downloading result to pyarrow, and using pyarrow interchange implementation. |
| 82 | + """ |
| 83 | + |
| 84 | + _value: blocks.Block |
| 85 | + |
| 86 | + version: int = 0 # version of the protocol |
| 87 | + |
| 88 | + def __dataframe__( |
| 89 | + self, nan_as_null: bool = False, allow_copy: bool = True |
| 90 | + ) -> InterchangeDataFrame: |
| 91 | + return self |
| 92 | + |
| 93 | + @classmethod |
| 94 | + def _from_bigframes(cls, df: bigframes.dataframe.DataFrame): |
| 95 | + block = df._block.with_column_labels( |
| 96 | + [str(label) for label in df._block.column_labels] |
| 97 | + ) |
| 98 | + return cls(block) |
| 99 | + |
| 100 | + # In future, could potentially rely on executor to refetch batches efficiently with caching, |
| 101 | + # but safest for now to just request a single execution and save the whole table. |
| 102 | + @functools.cache |
| 103 | + def _arrow_dataframe(self): |
| 104 | + arrow_table, _ = self._value.reset_index( |
| 105 | + replacement=bigframes.enums.DefaultIndexKind.NULL |
| 106 | + ).to_arrow(allow_large_results=False) |
| 107 | + return arrow_table.__dataframe__() |
| 108 | + |
| 109 | + @property |
| 110 | + def metadata(self): |
| 111 | + # Allows round-trip without materialization |
| 112 | + return {"bigframes.block": self._value} |
| 113 | + |
| 114 | + def num_columns(self) -> int: |
| 115 | + """ |
| 116 | + Return the number of columns in the DataFrame. |
| 117 | + """ |
| 118 | + return len(self._value.value_columns) |
| 119 | + |
| 120 | + def num_rows(self) -> Optional[int]: |
| 121 | + return self._value.shape[0] |
| 122 | + |
| 123 | + def num_chunks(self) -> int: |
| 124 | + return self._arrow_dataframe().num_chunks() |
| 125 | + |
| 126 | + def column_names(self) -> Iterable[str]: |
| 127 | + return [col for col in self._value.column_labels] |
| 128 | + |
| 129 | + def get_column(self, i: int) -> InterchangeColumn: |
| 130 | + return InterchangeColumn(self, i) |
| 131 | + |
| 132 | + # For single column getters, we download the whole dataframe still |
| 133 | + # This is inefficient in some cases, but more efficient in other |
| 134 | + def get_column_by_name(self, name: str) -> InterchangeColumn: |
| 135 | + col_id = self._value.resolve_label_exact(name) |
| 136 | + assert col_id is not None |
| 137 | + pos = self._value.value_columns.index(col_id) |
| 138 | + return InterchangeColumn(self, pos) |
| 139 | + |
| 140 | + def get_columns(self) -> Iterable[InterchangeColumn]: |
| 141 | + return [InterchangeColumn(self, i) for i in range(self.num_columns())] |
| 142 | + |
| 143 | + def select_columns(self, indices: Sequence[int]) -> InterchangeDataFrame: |
| 144 | + col_ids = [self._value.value_columns[i] for i in indices] |
| 145 | + new_value = self._value.select_columns(col_ids) |
| 146 | + return InterchangeDataFrame(new_value) |
| 147 | + |
| 148 | + def select_columns_by_name(self, names: Sequence[str]) -> InterchangeDataFrame: |
| 149 | + col_ids = [self._value.resolve_label_exact(name) for name in names] |
| 150 | + assert all(id is not None for id in col_ids) |
| 151 | + new_value = self._value.select_columns(col_ids) # type: ignore |
| 152 | + return InterchangeDataFrame(new_value) |
| 153 | + |
| 154 | + def get_chunks(self, n_chunks: Optional[int] = None) -> Iterable: |
| 155 | + return self._arrow_dataframe().get_chunks(n_chunks) |
0 commit comments