langchain/libs/community/langchain_community/vectorstores/baiduvectordb.py

438 lines
15 KiB
Python
Raw Normal View History

"""Wrapper around the Baidu vector database."""
from __future__ import annotations
import json
import logging
import time
from typing import Any, Dict, Iterable, List, Optional, Tuple
import numpy as np
from langchain_core.documents import Document
from langchain_core.embeddings import Embeddings
from langchain_core.utils import guard_import
from langchain_core.vectorstores import VectorStore
from langchain_community.vectorstores.utils import maximal_marginal_relevance
logger = logging.getLogger(__name__)
class ConnectionParams:
"""Baidu VectorDB Connection params.
See the following documentation for details:
https://cloud.baidu.com/doc/VDB/s/6lrsob0wy
Attribute:
endpoint (str) : The access address of the vector database server
that the client needs to connect to.
api_key (str): API key for client to access the vector database server,
which is used for authentication.
account (str) : Account for client to access the vector database server.
connection_timeout_in_mills (int) : Request Timeout.
"""
def __init__(
self,
endpoint: str,
api_key: str,
account: str = "root",
connection_timeout_in_mills: int = 50 * 1000,
):
self.endpoint = endpoint
self.api_key = api_key
self.account = account
self.connection_timeout_in_mills = connection_timeout_in_mills
class TableParams:
"""Baidu VectorDB table params.
See the following documentation for details:
https://cloud.baidu.com/doc/VDB/s/mlrsob0p6
"""
def __init__(
self,
dimension: int,
replication: int = 3,
partition: int = 1,
index_type: str = "HNSW",
metric_type: str = "L2",
params: Optional[Dict] = None,
):
self.dimension = dimension
self.replication = replication
self.partition = partition
self.index_type = index_type
self.metric_type = metric_type
self.params = params
class BaiduVectorDB(VectorStore):
"""Baidu VectorDB as a vector store.
In order to use this you need to have a database instance.
See the following documentation for details:
https://cloud.baidu.com/doc/VDB/index.html
"""
field_id: str = "id"
field_vector: str = "vector"
field_text: str = "text"
field_metadata: str = "metadata"
index_vector: str = "vector_idx"
def __init__(
self,
embedding: Embeddings,
connection_params: ConnectionParams,
table_params: TableParams = TableParams(128),
database_name: str = "LangChainDatabase",
table_name: str = "LangChainTable",
drop_old: Optional[bool] = False,
):
pymochow = guard_import("pymochow")
configuration = guard_import("pymochow.configuration")
auth = guard_import("pymochow.auth.bce_credentials")
self.mochowtable = guard_import("pymochow.model.table")
self.mochowenum = guard_import("pymochow.model.enum")
self.embedding_func = embedding
self.table_params = table_params
config = configuration.Configuration(
credentials=auth.BceCredentials(
connection_params.account, connection_params.api_key
),
endpoint=connection_params.endpoint,
connection_timeout_in_mills=connection_params.connection_timeout_in_mills,
)
self.vdb_client = pymochow.MochowClient(config)
db_list = self.vdb_client.list_databases()
db_exist: bool = False
for db in db_list:
if database_name == db.database_name:
db_exist = True
break
if db_exist:
self.database = self.vdb_client.database(database_name)
else:
self.database = self.vdb_client.create_database(database_name)
try:
self.table = self.database.describe_table(table_name)
if drop_old:
self.database.drop_table(table_name)
self._create_table(table_name)
except pymochow.exception.ServerError:
self._create_table(table_name)
def _create_table(self, table_name: str) -> None:
schema = guard_import("pymochow.model.schema")
index_type = None
for k, v in self.mochowenum.IndexType.__members__.items():
if k == self.table_params.index_type:
index_type = v
if index_type is None:
raise ValueError("unsupported index_type")
metric_type = None
for k, v in self.mochowenum.MetricType.__members__.items():
if k == self.table_params.metric_type:
metric_type = v
if metric_type is None:
raise ValueError("unsupported metric_type")
if self.table_params.params is None:
params = schema.HNSWParams(m=16, efconstruction=200)
else:
params = schema.HNSWParams(
m=self.table_params.params.get("M", 16),
efconstruction=self.table_params.params.get("efConstruction", 200),
)
fields = []
fields.append(
schema.Field(
self.field_id,
self.mochowenum.FieldType.STRING,
primary_key=True,
partition_key=True,
auto_increment=False,
not_null=True,
)
)
fields.append(
schema.Field(
self.field_vector,
self.mochowenum.FieldType.FLOAT_VECTOR,
dimension=self.table_params.dimension,
not_null=True,
)
)
fields.append(schema.Field(self.field_text, self.mochowenum.FieldType.STRING))
fields.append(
schema.Field(self.field_metadata, self.mochowenum.FieldType.STRING)
)
indexes = []
indexes.append(
schema.VectorIndex(
index_name=self.index_vector,
index_type=index_type,
field=self.field_vector,
metric_type=metric_type,
params=params,
)
)
self.table = self.database.create_table(
table_name=table_name,
replication=self.table_params.replication,
partition=self.mochowtable.Partition(
partition_num=self.table_params.partition
),
schema=schema.Schema(fields=fields, indexes=indexes),
)
while True:
time.sleep(1)
table = self.database.describe_table(table_name)
if table.state == self.mochowenum.TableState.NORMAL:
break
@property
def embeddings(self) -> Embeddings:
return self.embedding_func
@classmethod
def from_texts(
cls,
texts: List[str],
embedding: Embeddings,
metadatas: Optional[List[dict]] = None,
connection_params: Optional[ConnectionParams] = None,
table_params: Optional[TableParams] = None,
database_name: str = "LangChainDatabase",
table_name: str = "LangChainTable",
drop_old: Optional[bool] = False,
**kwargs: Any,
) -> BaiduVectorDB:
"""Create a table, indexes it with HNSW, and insert data."""
if len(texts) == 0:
raise ValueError("texts is empty")
if connection_params is None:
raise ValueError("connection_params is empty")
try:
embeddings = embedding.embed_documents(texts[0:1])
except NotImplementedError:
embeddings = [embedding.embed_query(texts[0])]
dimension = len(embeddings[0])
if table_params is None:
table_params = TableParams(dimension=dimension)
else:
table_params.dimension = dimension
vector_db = cls(
embedding=embedding,
connection_params=connection_params,
table_params=table_params,
database_name=database_name,
table_name=table_name,
drop_old=drop_old,
)
vector_db.add_texts(texts=texts, metadatas=metadatas)
return vector_db
def add_texts(
self,
texts: Iterable[str],
metadatas: Optional[List[dict]] = None,
batch_size: int = 1000,
**kwargs: Any,
) -> List[str]:
"""Insert text data into Baidu VectorDB."""
texts = list(texts)
try:
embeddings = self.embedding_func.embed_documents(texts)
except NotImplementedError:
embeddings = [self.embedding_func.embed_query(x) for x in texts]
if len(embeddings) == 0:
logger.debug("Nothing to insert, skipping.")
return []
pks: list[str] = []
total_count = len(embeddings)
for start in range(0, total_count, batch_size):
# Grab end index
rows = []
end = min(start + batch_size, total_count)
for id in range(start, end, 1):
metadata = "{}"
if metadatas is not None:
metadata = json.dumps(metadatas[id])
row = self.mochowtable.Row(
id="{}-{}-{}".format(time.time_ns(), hash(texts[id]), id),
vector=[float(num) for num in embeddings[id]],
text=texts[id],
metadata=metadata,
)
rows.append(row)
pks.append(str(id))
self.table.upsert(rows=rows)
# need rebuild vindex after upsert
self.table.rebuild_index(self.index_vector)
while True:
time.sleep(2)
index = self.table.describe_index(self.index_vector)
if index.state == self.mochowenum.IndexState.NORMAL:
break
return pks
def similarity_search(
self,
query: str,
k: int = 4,
param: Optional[dict] = None,
expr: Optional[str] = None,
**kwargs: Any,
) -> List[Document]:
"""Perform a similarity search against the query string."""
res = self.similarity_search_with_score(
query=query, k=k, param=param, expr=expr, **kwargs
)
return [doc for doc, _ in res]
def similarity_search_with_score(
self,
query: str,
k: int = 4,
param: Optional[dict] = None,
expr: Optional[str] = None,
**kwargs: Any,
) -> List[Tuple[Document, float]]:
"""Perform a search on a query string and return results with score."""
# Embed the query text.
embedding = self.embedding_func.embed_query(query)
res = self._similarity_search_with_score(
embedding=embedding, k=k, param=param, expr=expr, **kwargs
)
return res
def similarity_search_by_vector(
self,
embedding: List[float],
k: int = 4,
param: Optional[dict] = None,
expr: Optional[str] = None,
**kwargs: Any,
) -> List[Document]:
"""Perform a similarity search against the query string."""
res = self._similarity_search_with_score(
embedding=embedding, k=k, param=param, expr=expr, **kwargs
)
return [doc for doc, _ in res]
def _similarity_search_with_score(
self,
embedding: List[float],
k: int = 4,
param: Optional[dict] = None,
expr: Optional[str] = None,
**kwargs: Any,
) -> List[Tuple[Document, float]]:
"""Perform a search on a query string and return results with score."""
ef = 10 if param is None else param.get("ef", 10)
anns = self.mochowtable.AnnSearch(
vector_field=self.field_vector,
vector_floats=[float(num) for num in embedding],
params=self.mochowtable.HNSWSearchParams(ef=ef, limit=k),
filter=expr,
)
res = self.table.search(anns=anns)
rows = [[item] for item in res.rows]
# Organize results.
ret: List[Tuple[Document, float]] = []
if rows is None or len(rows) == 0:
return ret
for row in rows:
for result in row:
row_data = result.get("row", {})
meta = row_data.get(self.field_metadata)
if meta is not None:
meta = json.loads(meta)
doc = Document(
page_content=row_data.get(self.field_text), metadata=meta
)
pair = (doc, result.get("score", 0.0))
ret.append(pair)
return ret
def max_marginal_relevance_search(
self,
query: str,
k: int = 4,
fetch_k: int = 20,
lambda_mult: float = 0.5,
param: Optional[dict] = None,
expr: Optional[str] = None,
**kwargs: Any,
) -> List[Document]:
"""Perform a search and return results that are reordered by MMR."""
embedding = self.embedding_func.embed_query(query)
return self._max_marginal_relevance_search(
embedding=embedding,
k=k,
fetch_k=fetch_k,
lambda_mult=lambda_mult,
param=param,
expr=expr,
**kwargs,
)
def _max_marginal_relevance_search(
self,
embedding: list[float],
k: int = 4,
fetch_k: int = 20,
lambda_mult: float = 0.5,
param: Optional[dict] = None,
expr: Optional[str] = None,
**kwargs: Any,
) -> List[Document]:
"""Perform a search and return results that are reordered by MMR."""
ef = 10 if param is None else param.get("ef", 10)
anns = self.mochowtable.AnnSearch(
vector_field=self.field_vector,
vector_floats=[float(num) for num in embedding],
params=self.mochowtable.HNSWSearchParams(ef=ef, limit=k),
filter=expr,
)
res = self.table.search(anns=anns, retrieve_vector=True)
# Organize results.
documents: List[Document] = []
ordered_result_embeddings = []
rows = [[item] for item in res.rows]
if rows is None or len(rows) == 0:
return documents
for row in rows:
for result in row:
row_data = result.get("row", {})
meta = row_data.get(self.field_metadata)
if meta is not None:
meta = json.loads(meta)
doc = Document(
page_content=row_data.get(self.field_text), metadata=meta
)
documents.append(doc)
ordered_result_embeddings.append(row_data.get(self.field_vector))
# Get the new order of results.
new_ordering = maximal_marginal_relevance(
np.array(embedding), ordered_result_embeddings, k=k, lambda_mult=lambda_mult
)
# Reorder the values and return.
ret = []
for x in new_ordering:
# Function can return -1 index
if x == -1:
break
else:
ret.append(documents[x])
return ret