Merge branch 'main' into garrettwu-ml

Author: sycai

bigframes/bigquery/__init__.py

@@ -43,6 +43,7 @@
     st_regionstats,
     st_simplify,
 )
+from bigframes.bigquery._operations.io import load_data
 from bigframes.bigquery._operations.json import (
     json_extract,
     json_extract_array,
@@ -107,6 +108,8 @@
     struct,
     # table ops
     create_external_table,
+    # io ops
+    load_data,
 ]
 
 _module = sys.modules[__name__]
@@ -160,6 +163,8 @@
     "struct",
     # table ops
     "create_external_table",
+    # io ops
+    "load_data",
     # Modules / SQL namespaces
     "ai",
     "ml",

bigframes/bigquery/_operations/ai.py

@@ -19,7 +19,7 @@
 from __future__ import annotations
 
 import json
-from typing import Any, Iterable, List, Literal, Mapping, Tuple, Union
+from typing import Any, Dict, Iterable, List, Literal, Mapping, Optional, Tuple, Union
 
 import pandas as pd
 
@@ -28,6 +28,7 @@
 from bigframes import series, session
 from bigframes.core import convert
 from bigframes.core.logging import log_adapter
+import bigframes.core.sql.literals
 from bigframes.ml import core as ml_core
 from bigframes.operations import ai_ops, output_schemas
 
@@ -388,6 +389,113 @@ def generate_double(
     return series_list[0]._apply_nary_op(operator, series_list[1:])
 
 
+@log_adapter.method_logger(custom_base_name="bigquery_ai")
+def generate_embedding(
+    model_name: str,
+    data: Union[dataframe.DataFrame, series.Series, pd.DataFrame, pd.Series],
+    *,
+    output_dimensionality: Optional[int] = None,
+    task_type: Optional[str] = None,
+    start_second: Optional[float] = None,
+    end_second: Optional[float] = None,
+    interval_seconds: Optional[float] = None,
+    trial_id: Optional[int] = None,
+) -> dataframe.DataFrame:
+    """
+    Creates embeddings that describe an entity—for example, a piece of text or an image.
+
+    **Examples:**
+
+        >>> import bigframes.pandas as bpd
+        >>> import bigframes.bigquery as bbq
+        >>> df = bpd.DataFrame({"content": ["apple", "bear", "pear"]})
+        >>> bbq.ai.generate_embedding(
+        ...     "project.dataset.model_name",
+        ...     df
+        ... ) # doctest: +SKIP
+
+    Args:
+        model_name (str):
+            The name of a remote model from Vertex AI, such as the
+            multimodalembedding@001 model.
+        data (bigframes.pandas.DataFrame or bigframes.pandas.Series):
+            The data to generate embeddings for. If a Series is provided, it is
+            treated as the 'content' column.  If a DataFrame is provided, it
+            must contain a 'content' column, or you must rename the column you
+            wish to embed to 'content'.
+        output_dimensionality (int, optional):
+            An INT64 value that specifies the number of dimensions to use when
+            generating embeddings. For example, if you specify 256 AS
+            output_dimensionality, then the embedding output column contains a
+            256-dimensional embedding for each input value. To find the
+            supported range of output dimensions, read about the available
+            `Google text embedding models <https://docs.cloud.google.com/vertex-ai/generative-ai/docs/embeddings/get-text-embeddings#google-models>`_.
+        task_type (str, optional):
+            A STRING literal that specifies the intended downstream application to
+            help the model produce better quality embeddings. For a list of
+            supported task types and how to choose which one to use, see `Choose an
+            embeddings task type <http://docs.cloud.google.com/vertex-ai/generative-ai/docs/embeddings/task-types>`_.
+        start_second (float, optional):
+            The second in the video at which to start the embedding. The default value is 0.
+        end_second (float, optional):
+            The second in the video at which to end the embedding. The default value is 120.
+        interval_seconds (float, optional):
+            The interval to use when creating embeddings. The default value is 16.
+        trial_id (int, optional):
+            An INT64 value that identifies the hyperparameter tuning trial that
+            you want the function to evaluate. The function uses the optimal
+            trial by default. Only specify this argument if you ran
+            hyperparameter tuning when creating the model.
+
+    Returns:
+        bigframes.pandas.DataFrame:
+            A new DataFrame with the generated embeddings. See the `SQL
+            reference for AI.GENERATE_EMBEDDING
+            <https://docs.cloud.google.com/bigquery/docs/reference/standard-sql/bigqueryml-syntax-ai-generate-embedding#output>`_
+            for details.
+    """
+    if isinstance(data, (pd.DataFrame, pd.Series)):
+        data = bpd.read_pandas(data)
+
+    if isinstance(data, series.Series):
+        data = data.copy()
+        data.name = "content"
+        data_df = data.to_frame()
+    elif isinstance(data, dataframe.DataFrame):
+        data_df = data
+    else:
+        raise ValueError(f"Unsupported data type: {type(data)}")
+
+    # We need to get the SQL for the input data to pass as a subquery to the TVF
+    source_sql = data_df.sql
+
+    struct_fields: Dict[str, bigframes.core.sql.literals.STRUCT_VALUES] = {}
+    if output_dimensionality is not None:
+        struct_fields["OUTPUT_DIMENSIONALITY"] = output_dimensionality
+    if task_type is not None:
+        struct_fields["TASK_TYPE"] = task_type
+    if start_second is not None:
+        struct_fields["START_SECOND"] = start_second
+    if end_second is not None:
+        struct_fields["END_SECOND"] = end_second
+    if interval_seconds is not None:
+        struct_fields["INTERVAL_SECONDS"] = interval_seconds
+    if trial_id is not None:
+        struct_fields["TRIAL_ID"] = trial_id
+
+    # Construct the TVF query
+    query = f"""
+        SELECT *
+        FROM AI.GENERATE_EMBEDDING(
+            MODEL `{model_name}`,
+            ({source_sql}),
+            {bigframes.core.sql.literals.struct_literal(struct_fields)})
+        )
+    """
+
+    return data_df._session.read_gbq(query)
+
+
 @log_adapter.method_logger(custom_base_name="bigquery_ai")
 def if_(
     prompt: PROMPT_TYPE,

bigframes/bigquery/_operations/io.py

@@ -0,0 +1,94 @@
+# Copyright 2026 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import annotations
+
+from typing import Mapping, Optional, Union
+
+import pandas as pd
+
+from bigframes.bigquery._operations.table import _get_table_metadata
+import bigframes.core.logging.log_adapter as log_adapter
+import bigframes.core.sql.io
+import bigframes.session
+
+
+@log_adapter.method_logger(custom_base_name="bigquery_io")
+def load_data(
+    table_name: str,
+    *,
+    write_disposition: str = "INTO",
+    columns: Optional[Mapping[str, str]] = None,
+    partition_by: Optional[list[str]] = None,
+    cluster_by: Optional[list[str]] = None,
+    table_options: Optional[Mapping[str, Union[str, int, float, bool, list]]] = None,
+    from_files_options: Mapping[str, Union[str, int, float, bool, list]],
+    with_partition_columns: Optional[Mapping[str, str]] = None,
+    connection_name: Optional[str] = None,
+    session: Optional[bigframes.session.Session] = None,
+) -> pd.Series:
+    """
+    Loads data into a BigQuery table.
+    See the `BigQuery LOAD DATA DDL syntax
+    <https://docs.cloud.google.com/bigquery/docs/reference/standard-sql/load-statements>`_
+    for additional reference.
+    Args:
+        table_name (str):
+            The name of the table in BigQuery.
+        write_disposition (str, default "INTO"):
+            Whether to replace the table if it already exists ("OVERWRITE") or append to it ("INTO").
+        columns (Mapping[str, str], optional):
+            The table's schema.
+        partition_by (list[str], optional):
+            A list of partition expressions to partition the table by. See https://docs.cloud.google.com/bigquery/docs/reference/standard-sql/load-statements#partition_expression.
+        cluster_by (list[str], optional):
+            A list of columns to cluster the table by.
+        table_options (Mapping[str, Union[str, int, float, bool, list]], optional):
+            The table options.
+        from_files_options (Mapping[str, Union[str, int, float, bool, list]]):
+            The options for loading data from files.
+        with_partition_columns (Mapping[str, str], optional):
+            The table's partition columns.
+        connection_name (str, optional):
+            The connection to use for the table.
+        session (bigframes.session.Session, optional):
+            The session to use. If not provided, the default session is used.
+    Returns:
+        pandas.Series:
+            A Series with object dtype containing the table metadata. Reference
+            the `BigQuery Table REST API reference
+            <https://cloud.google.com/bigquery/docs/reference/rest/v2/tables#Table>`_
+            for available fields.
+    """
+    import bigframes.pandas as bpd
+
+    sql = bigframes.core.sql.io.load_data_ddl(
+        table_name=table_name,
+        write_disposition=write_disposition,
+        columns=columns,
+        partition_by=partition_by,
+        cluster_by=cluster_by,
+        table_options=table_options,
+        from_files_options=from_files_options,
+        with_partition_columns=with_partition_columns,
+        connection_name=connection_name,
+    )
+
+    if session is None:
+        bpd.read_gbq_query(sql)
+        session = bpd.get_global_session()
+    else:
+        session.read_gbq_query(sql)
+
+    return _get_table_metadata(bqclient=session.bqclient, table_name=table_name)

bigframes/bigquery/ai.py

@@ -22,6 +22,7 @@
     generate,
     generate_bool,
     generate_double,
+    generate_embedding,
     generate_int,
     if_,
     score,
@@ -33,6 +34,7 @@
     "generate",
     "generate_bool",
     "generate_double",
+    "generate_embedding",
     "generate_int",
     "if_",
     "score",

bigframes/core/sql/io.py

@@ -0,0 +1,87 @@
+# Copyright 2026 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import annotations
+
+from typing import Mapping, Optional, Union
+
+
+def load_data_ddl(
+    table_name: str,
+    *,
+    write_disposition: str = "INTO",
+    columns: Optional[Mapping[str, str]] = None,
+    partition_by: Optional[list[str]] = None,
+    cluster_by: Optional[list[str]] = None,
+    table_options: Optional[Mapping[str, Union[str, int, float, bool, list]]] = None,
+    from_files_options: Mapping[str, Union[str, int, float, bool, list]],
+    with_partition_columns: Optional[Mapping[str, str]] = None,
+    connection_name: Optional[str] = None,
+) -> str:
+    """Generates the LOAD DATA DDL statement."""
+    statement = ["LOAD DATA"]
+    statement.append(write_disposition)
+    statement.append(table_name)
+
+    if columns:
+        column_defs = ", ".join([f"{name} {typ}" for name, typ in columns.items()])
+        statement.append(f"({column_defs})")
+
+    if partition_by:
+        statement.append(f"PARTITION BY {', '.join(partition_by)}")
+
+    if cluster_by:
+        statement.append(f"CLUSTER BY {', '.join(cluster_by)}")
+
+    if table_options:
+        opts = []
+        for key, value in table_options.items():
+            if isinstance(value, str):
+                value_sql = repr(value)
+                opts.append(f"{key} = {value_sql}")
+            elif isinstance(value, bool):
+                opts.append(f"{key} = {str(value).upper()}")
+            elif isinstance(value, list):
+                list_str = ", ".join([repr(v) for v in value])
+                opts.append(f"{key} = [{list_str}]")
+            else:
+                opts.append(f"{key} = {value}")
+        options_str = ", ".join(opts)
+        statement.append(f"OPTIONS ({options_str})")
+
+    opts = []
+    for key, value in from_files_options.items():
+        if isinstance(value, str):
+            value_sql = repr(value)
+            opts.append(f"{key} = {value_sql}")
+        elif isinstance(value, bool):
+            opts.append(f"{key} = {str(value).upper()}")
+        elif isinstance(value, list):
+            list_str = ", ".join([repr(v) for v in value])
+            opts.append(f"{key} = [{list_str}]")
+        else:
+            opts.append(f"{key} = {value}")
+    options_str = ", ".join(opts)
+    statement.append(f"FROM FILES ({options_str})")
+
+    if with_partition_columns:
+        part_defs = ", ".join(
+            [f"{name} {typ}" for name, typ in with_partition_columns.items()]
+        )
+        statement.append(f"WITH PARTITION COLUMNS ({part_defs})")
+
+    if connection_name:
+        statement.append(f"WITH CONNECTION `{connection_name}`")
+
+    return " ".join(statement)