pyobvector 0.2.15__tar.gz → 0.2.17__tar.gz

{pyobvector-0.2.15 → pyobvector-0.2.17}/PKG-INFO

@@ -1,7 +1,8 @@
-Metadata-Version: 2.3
+Metadata-Version: 2.4
 Name: pyobvector
-Version: 0.2.15
+Version: 0.2.17
 Summary: A python SDK for OceanBase Vector Store, based on SQLAlchemy, compatible with Milvus API.
+License-File: LICENSE
 Author: shanhaikang.shk
 Author-email: shanhaikang.shk@oceanbase.com
 Requires-Python: >=3.9,<4.0
@@ -11,12 +12,13 @@ Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
-Requires-Dist: aiomysql (>=0.2.0,<0.3.0)
+Classifier: Programming Language :: Python :: 3.14
+Requires-Dist: aiomysql (>=0.3.2,<0.4.0)
 Requires-Dist: numpy (>=1.17.0,<2.0.0)
 Requires-Dist: pydantic (>=2.7.0,<3)
 Requires-Dist: pymysql (>=1.1.1,<2.0.0)
 Requires-Dist: sqlalchemy (>=1.4,<=3)
-Requires-Dist: sqlglot (>=26.0.1,<27.0.0)
+Requires-Dist: sqlglot (>=26.0.1)
 Description-Content-Type: text/markdown
 
 # pyobvector
@@ -36,7 +38,7 @@ poetry install
 - install with pip:
 
 ```shell
-pip install pyobvector==0.2.14
+pip install pyobvector==0.2.17
 ```
 
 ## Build Doc
@@ -174,19 +176,75 @@ client.insert(test_collection_name, data=data1)
 - do ann search:
 
 ```python
-# perform ann search
+# perform ann search with basic column selection
 res = self.client.ann_search(
     test_collection_name, 
     vec_data=[0,0,0], 
     vec_column_name='embedding',
     distance_func=l2_distance,
     topk=5,
-    output_column_names=['id']
+    output_column_names=['id']  # Legacy parameter
 )
 # For example, the result will be:
 # [(112,), (111,), (10,), (11,), (12,)]
+
+# perform ann search with SQLAlchemy expressions (recommended)
+from sqlalchemy import Table, text, func
+
+table = Table(test_collection_name, client.metadata_obj, autoload_with=client.engine)
+res = self.client.ann_search(
+    test_collection_name, 
+    vec_data=[0,0,0], 
+    vec_column_name='embedding',
+    distance_func=l2_distance,
+    topk=5,
+    output_columns=[
+        table.c.id,
+        table.c.meta,
+        (table.c.id + 1000).label('id_plus_1000'),
+        text("JSON_EXTRACT(meta, '$.key') as extracted_key")
+    ]
+)
+# For example, the result will be:
+# [(112, '{"key": "value"}', 1112, 'value'), ...]
+
+# perform ann search with distance threshold (filter results by distance)
+res = self.client.ann_search(
+    test_collection_name, 
+    vec_data=[0,0,0], 
+    vec_column_name='embedding',
+    distance_func=l2_distance,
+    with_dist=True,
+    topk=10,
+    output_column_names=['id'],
+    distance_threshold=0.5  # Only return results where distance <= 0.5
+)
+# Only returns results with distance <= 0.5
+# For example, the result will be:
+# [(10, 0.0), (11, 0.0), ...]  # Only includes results with distance <= 0.5
 ```
 
+#### ann_search Parameters
+
+The `ann_search` method supports flexible output column selection through the `output_columns` parameter:
+
+- **`output_columns`** (recommended): Accepts SQLAlchemy Column objects, expressions, or a mix of both
+  - Column objects: `table.c.id`, `table.c.name`
+  - Expressions: `(table.c.age + 10).label('age_plus_10')`
+  - JSON queries: `text("JSON_EXTRACT(meta, '$.key') as extracted_key")`
+  - String functions: `func.concat(table.c.name, ' (', table.c.age, ')').label('name_age')`
+
+- **`output_column_names`** (legacy): Accepts list of column name strings
+  - Example: `['id', 'name', 'meta']`
+
+- **Parameter Priority**: `output_columns` takes precedence over `output_column_names` when both are provided
+
+- **`distance_threshold`** (optional): Filter results by distance threshold
+  - Type: `Optional[float]`
+  - Only returns results where `distance <= threshold`
+  - Example: `distance_threshold=0.5` returns only results with distance <= 0.5
+  - Use case: Quality control for similarity search, only return highly similar results
+
 - If you want to use pure `SQLAlchemy` API with `OceanBase` dialect, you can just get an `SQLAlchemy.engine` via `client.engine`. The engine can also be created as following:
 
 ```python

{pyobvector-0.2.15 → pyobvector-0.2.17}/README.md

@@ -15,7 +15,7 @@ poetry install
 - install with pip:
 
 ```shell
-pip install pyobvector==0.2.14
+pip install pyobvector==0.2.17
 ```
 
 ## Build Doc
@@ -153,19 +153,75 @@ client.insert(test_collection_name, data=data1)
 - do ann search:
 
 ```python
-# perform ann search
+# perform ann search with basic column selection
 res = self.client.ann_search(
     test_collection_name, 
     vec_data=[0,0,0], 
     vec_column_name='embedding',
     distance_func=l2_distance,
     topk=5,
-    output_column_names=['id']
+    output_column_names=['id']  # Legacy parameter
 )
 # For example, the result will be:
 # [(112,), (111,), (10,), (11,), (12,)]
+
+# perform ann search with SQLAlchemy expressions (recommended)
+from sqlalchemy import Table, text, func
+
+table = Table(test_collection_name, client.metadata_obj, autoload_with=client.engine)
+res = self.client.ann_search(
+    test_collection_name, 
+    vec_data=[0,0,0], 
+    vec_column_name='embedding',
+    distance_func=l2_distance,
+    topk=5,
+    output_columns=[
+        table.c.id,
+        table.c.meta,
+        (table.c.id + 1000).label('id_plus_1000'),
+        text("JSON_EXTRACT(meta, '$.key') as extracted_key")
+    ]
+)
+# For example, the result will be:
+# [(112, '{"key": "value"}', 1112, 'value'), ...]
+
+# perform ann search with distance threshold (filter results by distance)
+res = self.client.ann_search(
+    test_collection_name, 
+    vec_data=[0,0,0], 
+    vec_column_name='embedding',
+    distance_func=l2_distance,
+    with_dist=True,
+    topk=10,
+    output_column_names=['id'],
+    distance_threshold=0.5  # Only return results where distance <= 0.5
+)
+# Only returns results with distance <= 0.5
+# For example, the result will be:
+# [(10, 0.0), (11, 0.0), ...]  # Only includes results with distance <= 0.5
 ```
 
+#### ann_search Parameters
+
+The `ann_search` method supports flexible output column selection through the `output_columns` parameter:
+
+- **`output_columns`** (recommended): Accepts SQLAlchemy Column objects, expressions, or a mix of both
+  - Column objects: `table.c.id`, `table.c.name`
+  - Expressions: `(table.c.age + 10).label('age_plus_10')`
+  - JSON queries: `text("JSON_EXTRACT(meta, '$.key') as extracted_key")`
+  - String functions: `func.concat(table.c.name, ' (', table.c.age, ')').label('name_age')`
+
+- **`output_column_names`** (legacy): Accepts list of column name strings
+  - Example: `['id', 'name', 'meta']`
+
+- **Parameter Priority**: `output_columns` takes precedence over `output_column_names` when both are provided
+
+- **`distance_threshold`** (optional): Filter results by distance threshold
+  - Type: `Optional[float]`
+  - Only returns results where `distance <= threshold`
+  - Example: `distance_threshold=0.5` returns only results with distance <= 0.5
+  - Use case: Quality control for similarity search, only return highly similar results
+
 - If you want to use pure `SQLAlchemy` API with `OceanBase` dialect, you can just get an `SQLAlchemy.engine` via `client.engine`. The engine can also be created as following:
 
 ```python

{pyobvector-0.2.15 → pyobvector-0.2.17}/pyobvector/__init__.py

@@ -14,6 +14,7 @@ In this mode, you can regard `pyobvector` as an extension of SQLAlchemy.
 * IndexParams           A list of IndexParam to create vector index in batch
 * DataType              Specify field type in collection schema for MilvusLikeClient
 * VECTOR                An extended data type in SQLAlchemy for ObVecClient
+* SPARSE_VECTOR         An extended data type in SQLAlchemy for ObVecClient
 * VectorIndex           An extended index type in SQLAlchemy for ObVecClient
 * FtsIndex              Full Text Search Index
 * FieldSchema           Clas to define field schema in collection for MilvusLikeClient
@@ -43,6 +44,7 @@ from .client import *
 from .schema import (
     ARRAY,
     VECTOR,
+    SPARSE_VECTOR,
     POINT,
     VectorIndex,
     OceanBaseDialect,
@@ -70,6 +72,7 @@ __all__ = [
     "DataType",
     "ARRAY",
     "VECTOR",
+    "SPARSE_VECTOR",
     "POINT",
     "VectorIndex",
     "FtsIndex",

{pyobvector-0.2.15 → pyobvector-0.2.17}/pyobvector/client/collection_schema.py

@@ -79,14 +79,14 @@ class FieldSchema:
             if "max_length" not in self.kwargs:
                 raise VarcharFieldParamException(
                     code=ErrorCode.INVALID_ARGUMENT,
-                    message=ExceptionsMessage.VarcharFieldMissinglengthParam,
+                    message=ExceptionsMessage.VarcharFieldMissingLengthParam,
                 )
             self.type_params["length"] = self.kwargs["max_length"]
         elif self.dtype == DataType.ARRAY:
             if "element_type" not in self.kwargs:
                 raise ArrayFieldParamException(
                     code=ErrorCode.INVALID_ARGUMENT,
-                    message=ExceptionsMessage.ArrayFiledMissingElementType,
+                    message=ExceptionsMessage.ArrayFieldMissingElementType,
                 )
             if self.kwargs["element_type"] in (
                 DataType.ARRAY,
@@ -95,7 +95,7 @@ class FieldSchema:
             ):
                 raise ArrayFieldParamException(
                     code=ErrorCode.INVALID_ARGUMENT,
-                    message=ExceptionsMessage.ArrayFiledInvalidElementType,
+                    message=ExceptionsMessage.ArrayFieldInvalidElementType,
                 )
 
             self.type_params["item_type"] = convert_datatype_to_sqltype(
@@ -147,9 +147,9 @@ class CollectionSchema:
         """Add field to collection.
 
         Args:
-        :param field_name (string) : new field name
-        :param datatype (DataType) : field data type
-        :param kwargs : parameters for data type
+            field_name (string): new field name
+            datatype (DataType): field data type
+            **kwargs: parameters for data type
         """
         field = FieldSchema(field_name, datatype, **kwargs)
         cur_idx = len(self.fields)

{pyobvector-0.2.15 → pyobvector-0.2.17}/pyobvector/client/exceptions.py

@@ -101,14 +101,14 @@ class ExceptionsMessage:
     )
     PrimaryFieldType = "Param primary_field must be int or str type."
     VectorFieldMissingDimParam = "Param 'dim' must be set for vector field."
-    VarcharFieldMissinglengthParam = "Param 'max_length' must be set for varchar field."
-    ArrayFiledMissingElementType = "Param 'element_type' must be set for array field."
-    ArrayFiledInvalidElementType = (
+    VarcharFieldMissingLengthParam = "Param 'max_length' must be set for varchar field."
+    ArrayFieldMissingElementType = "Param 'element_type' must be set for array field."
+    ArrayFieldInvalidElementType = (
         "Param 'element_type' can not be array/vector/varchar."
     )
     CollectionNotExists = "Collection does not exist."
     MetricTypeParamTypeInvalid = "MetricType param type should be string."
-    MetricTypeValueInvalid = "MetricType should be 'l2'/'ip' in ann search."
+    MetricTypeValueInvalid = "MetricType should be 'l2'/'ip'/'neg_ip'/'cosine' in ann search."
     UsingInIDsWhenMultiPrimaryKey = "Using 'ids' when table has multi primary key."
     ClusterVersionIsLow = (
         "OceanBase Vector Store is not supported because cluster version is below 4.3.3.0."

{pyobvector-0.2.15 → pyobvector-0.2.17}/pyobvector/client/fts_index_param.py

@@ -18,13 +18,12 @@ class FtsIndexParam:
         self.field_names = field_names
         self.parser_type = parser_type
 
-    def param_str(self) -> str:
-        if self.parser_type is None:
-            return None
+    def param_str(self) -> str | None:
         if self.parser_type == FtsParser.IK:
             return "ik"
         if self.parser_type == FtsParser.NGRAM:
             return "ngram"
+        return None
 
     def __iter__(self):
         yield "index_name", self.index_name

{pyobvector-0.2.15 → pyobvector-0.2.17}/pyobvector/client/index_param.py

@@ -9,7 +9,7 @@ class VecIndexType(Enum):
     IVFFLAT = 2
     IVFSQ = 3
     IVFPQ = 4
-
+    DAAT = 5
 
 class IndexParam:
     """Vector index parameters.
@@ -31,6 +31,7 @@ class IndexParam:
     IVFFLAT_ALGO_NAME = "ivf_flat"
     IVFSQ_ALGO_NAME = "ivf_sq8"
     IVFPQ_ALGO_NAME = "ivf_pq"
+    DAAT_ALGO_NAME = "daat"
 
     def __init__(
         self, index_name: str, field_name: str, index_type: Union[VecIndexType, str], **kwargs
@@ -57,6 +58,11 @@ class IndexParam:
         return self.index_type in [
             IndexParam.IVFPQ_ALGO_NAME,
         ]
+    
+    def is_index_type_sparse_vector(self):
+        return self.index_type in [
+            IndexParam.DAAT_ALGO_NAME,
+        ]
 
     def _get_vector_index_type_str(self):
         """Parse vector index type to string."""
@@ -71,6 +77,8 @@ class IndexParam:
                 return IndexParam.IVFSQ_ALGO_NAME
             elif self.index_type == VecIndexType.IVFPQ:
                 return IndexParam.IVFPQ_ALGO_NAME
+            elif self.index_type == VecIndexType.DAAT:
+                return IndexParam.DAAT_ALGO_NAME
             raise ValueError(f"unsupported vector index type: {self.index_type}")
         assert isinstance(self.index_type, str)
         index_type = self.index_type.lower()
@@ -80,6 +88,7 @@ class IndexParam:
             IndexParam.IVFFLAT_ALGO_NAME,
             IndexParam.IVFSQ_ALGO_NAME,
             IndexParam.IVFPQ_ALGO_NAME,
+            IndexParam.DAAT_ALGO_NAME,
         ]:
             raise ValueError(f"unsupported vector index type: {self.index_type}")
         return index_type
@@ -124,15 +133,19 @@ class IndexParam:
                     ob_params['ef_construction'] = params['efConstruction']
                 if 'efSearch' in params:
                     ob_params['ef_search'] = params['efSearch']
+        
+        if self.is_index_type_sparse_vector() and ob_params['distance'] != 'inner_product':
+            raise ValueError("Metric type should be 'inner_product' for sparse vector index.")
         return ob_params
 
     def param_str(self):
         """Parse vector index parameters to string."""
         ob_param = self._parse_kwargs()
         partial_str = ",".join([f"{k}={v}" for k, v in ob_param.items()])
-        if len(partial_str) > 0:
-            partial_str += ","
-        partial_str += f"type={self.index_type}"
+        if not self.is_index_type_sparse_vector():
+            if len(partial_str) > 0:
+                partial_str += ","
+            partial_str += f"type={self.index_type}"
         return partial_str
 
     def __iter__(self):
@@ -165,10 +178,10 @@ class IndexParams:
         """Add `IndexParam` to `IndexParams`
         
         Args:
-        :param field_name (string) : vector index built on which field
-        :param index_type (VecIndexType) :
-                vector index algorithms (Only HNSW supported)
-        :param index_name (string) : vector index name
+            field_name (string): vector index built on which field
+            index_type (VecIndexType): vector index algorithms (Only HNSW supported)
+            index_name (string): vector index name
+            **kwargs: additional parameters for different index types
         """
         index_param = IndexParam(index_name, field_name, index_type, **kwargs)
         pair_key = (field_name, index_name)