ArangoDB QA链

本笔记本展示了如何使用LLMs为ArangoDB数据库提供自然语言接口。

您可以通过ArangoDB Docker镜像运行本地ArangoDB实例：

docker run -p 8529:8529 -e ARANGO_ROOT_PASSWORD= arangodb/arangodb

另一种选择是使用ArangoDB Cloud Connector包来运行临时云实例：

pip install python-arango # ArangoDB Python驱动程序
pip install adb-cloud-connector # ArangoDB云实例提供程序
pip install openai
pip install langchain

# 实例化ArangoDB数据库
import json
from arango import ArangoClient
from adb_cloud_connector import get_temp_credentials

con = get_temp_credentials()

db = ArangoClient(hosts=con["url"]).db(
    con["dbName"], con["username"], con["password"], verify=True
)

print(json.dumps(con, indent=2))

    日志：请求新凭据...
    成功：获取新凭据
    {
      "dbName": "TUT3sp29s3pjf1io0h4cfdsq",
      "username": "TUTo6nkwgzkizej3kysgdyeo8",
      "password": "TUT9vx0qjqt42i9bq8uik4v9",
      "hostname": "tutorials.arangodb.cloud",
      "port": 8529,
      "url": "https://tutorials.arangodb.cloud:8529"
    }

# 实例化ArangoDB-LangChain图
from langchain.graphs import ArangoGraph

graph = ArangoGraph(db)

数据库填充 (Populating the Database)

我们将依靠Python驱动程序将我们的GameOfThrones数据导入到我们的数据库中。

if db.has_graph("GameOfThrones"):
    db.delete_graph("GameOfThrones", drop_collections=True)

db.create_graph(
    "GameOfThrones",
    edge_definitions=[
        {
            "edge_collection": "ChildOf",
            "from_vertex_collections": ["Characters"],
            "to_vertex_collections": ["Characters"],
        },
    ],
)

documents = [
    {
        "_key": "NedStark",
        "name": "Ned",
        "surname": "Stark",
        "alive": True,
        "age": 41,
        "gender": "male",
    },
    {
        "_key": "CatelynStark",
        "name": "Catelyn",
        "surname": "Stark",
        "alive": False,
        "age": 40,
        "gender": "female",
    },
    {
        "_key": "AryaStark",
        "name": "Arya",
        "surname": "Stark",
        "alive": True,
        "age": 11,
        "gender": "female",
    },
    {
        "_key": "BranStark",
        "name": "Bran",
        "surname": "Stark",
        "alive": True,
        "age": 10,
        "gender": "male",
    },
]

edges = [
    {"_to": "Characters/NedStark", "_from": "Characters/AryaStark"},
    {"_to": "Characters/NedStark", "_from": "Characters/BranStark"},
    {"_to": "Characters/CatelynStark", "_from": "Characters/AryaStark"},
    {"_to": "Characters/CatelynStark", "_from": "Characters/BranStark"},
]

db.collection("Characters").import_bulk(documents)
db.collection("ChildOf").import_bulk(edges)

    {'error': False,
     'created': 4,
     'errors': 0,
     'empty': 0,
     'updated': 0,
     'ignored': 0,
     'details': []}

获取和设置ArangoDB模式 (Getting & Setting the ArangoDB Schema)

在实例化ArangoDBGraph对象时，会生成一个初始的ArangoDB模式。如果您有兴趣查看或修改模式，以下是模式的获取器和设置器方法：

# 这里的模式应该是空的，
# 因为在ArangoDB数据导入之前已经初始化了`graph`。

import json

print(json.dumps(graph.schema, indent=4))

    {
        "Graph Schema": [],
        "Collection Schema": []
    }

graph.set_schema()

# 现在我们可以查看生成的模式

import json

print(json.dumps(graph.schema, indent=4))

    {
        "Graph Schema": [
            {
                "graph_name": "GameOfThrones",
                "edge_definitions": [
                    {
                        "edge_collection": "ChildOf",
                        "from_vertex_collections": [
                            "Characters"
                        ],
                        "to_vertex_collections": [
                            "Characters"
                        ]
                    }
                ]
            }
        ],
        "Collection Schema": [
            {
                "collection_name": "ChildOf",
                "collection_type": "edge",
                "edge_properties": [
                    {
                        "name": "_key",
                        "type": "str"
                    },
                    {
                        "name": "_id",
                        "type": "str"
                    },
                    {
                        "name": "_from",
                        "type": "str"
                    },
                    {
                        "name": "_to",
                        "type": "str"
                    },
                    {
                        "name": "_rev",
                        "type": "str"
                    }
                ],
                "example_edge": {
                    "_key": "266218884025",
                    "_id": "ChildOf/266218884025",
                    "_from": "Characters/AryaStark",
                    "_to": "Characters/NedStark",
                    "_rev": "_gVPKGSq---"
                }
            },
            {
                "collection_name": "Characters",
                "collection_type": "document",
                "document_properties": [
                    {
                        "name": "_key",
                        "type": "str"
                    },
                    {
                        "name": "_id",
                        "type": "str"
                    },
                    {
                        "name": "_rev",
                        "type": "str"
                    },
                    {
                        "name": "name",
                        "type": "str"
                    },
                    {
                        "name": "surname",
                        "type": "str"
                    },
                    {
                        "name": "alive",
                        "type": "bool"
                    },
                    {
                        "name": "age",
                        "type": "int"
                    },
                    {
                        "name": "gender",
                        "type": "str"
                    }
                ],
                "example_document": {
                    "_key": "NedStark",
                    "_id": "Characters/NedStark",
                    "_rev": "_gVPKGPi---",
                    "name": "Ned",
                    "surname": "Stark",
                    "alive": true,
                    "age": 41,
                    "gender": "male"
                }
            }
        ]
    }

查询ArangoDB数据库

我们现在可以使用ArangoDB图QA链来查询我们的数据

import os

os.environ["OPENAI_API_KEY"] = "your-key-here"

from langchain.chat_models import ChatOpenAI
from langchain.chains import ArangoGraphQAChain

chain = ArangoGraphQAChain.from_llm(
    ChatOpenAI(temperature=0), graph=graph, verbose=True
)

chain.run("Ned Stark还活着吗？")

    
    
    > 进入新的ArangoGraphQAChain链...
    AQL查询（1）：
    WITH Characters
    FOR character IN Characters
    FILTER character.name == "Ned" AND character.surname == "Stark"
    RETURN character.alive
    
    AQL结果：
    [True]
    
    > 完成链。





    '是的，Ned Stark还活着。'

chain.run("Arya Stark多大了？")

    
    
    > 进入新的ArangoGraphQAChain链...
    AQL查询（1）：
    WITH Characters
    FOR character IN Characters
    FILTER character.name == "Arya" && character.surname == "Stark"
    RETURN character.age
    
    AQL结果：
    [11]
    
    > 完成链。





    'Arya Stark今年11岁。'

chain.run("Arya Stark和Ned Stark有关系吗？")

    
    
    > 进入新的ArangoGraphQAChain链...
    AQL查询（1）：
    WITH Characters, ChildOf
    FOR v, e, p IN 1..1 OUTBOUND 'Characters/AryaStark' ChildOf
        FILTER p.vertices[-1]._key == 'NedStark'
        RETURN p
    
    AQL结果：
    [{'vertices': [{'_key': 'AryaStark', '_id': 'Characters/AryaStark', '_rev': '_gVPKGPi--B', 'name': 'Arya', 'surname': 'Stark', 'alive': True, 'age': 11, 'gender': 'female'}, {'_key': 'NedStark', '_id': 'Characters/NedStark', '_rev': '_gVPKGPi---', 'name': 'Ned', 'surname': 'Stark', 'alive': True, 'age': 41, 'gender': 'male'}], 'edges': [{'_key': '266218884025', '_id': 'ChildOf/266218884025', '_from': 'Characters/AryaStark', '_to': 'Characters/NedStark', '_rev': '_gVPKGSq---'}], 'weights': [0, 1]}]
    
    > 完成链.






    '是的，Arya Stark和Ned Stark有关系。根据从数据库检索到的信息，他们之间存在关系。Arya Stark是Ned Stark的孩子。'

chain.run("Arya Stark有死去的父母吗？")

    
    
    > 进入新的ArangoGraphQAChain链...
    AQL查询（1）：
    WITH Characters, ChildOf
    FOR v, e IN 1..1 OUTBOUND 'Characters/AryaStark' ChildOf
    FILTER v.alive == false
    RETURN e
    
    AQL结果：
    [{'_key': '266218884027', '_id': 'ChildOf/266218884027', '_from': 'Characters/AryaStark', '_to': 'Characters/CatelynStark', '_rev': '_gVPKGSu---'}]
    
    > 完成链.






    '是的，Arya Stark有一个已故的父母。父母是Catelyn Stark。'

链式修改器 (Chain Modifiers)

您可以修改以下 ArangoDBGraphQAChain 类的变量值，以修改链式查询结果的行为。

# 指定要返回的 AQL 查询结果的最大数量
chain.top_k = 10

# 指定是否在输出字典中返回 AQL 查询
chain.return_aql_query = True

# 指定是否在输出字典中返回 AQL JSON 结果
chain.return_aql_result = True

# 指定应进行的最大 AQL 生成尝试次数
chain.max_aql_generation_attempts = 5

# 指定一组 AQL 查询示例，这些示例将传递给 AQL 生成提示模板以促进少样本学习。
# 默认为空字符串。
chain.aql_examples = """
# Ned Stark 是否还活着？
RETURN DOCUMENT('Characters/NedStark').alive

# Arya Stark 是否是 Ned Stark 的孩子？
FOR e IN ChildOf
    FILTER e._from == "Characters/AryaStark" AND e._to == "Characters/NedStark"
    RETURN e
"""

chain.run("Ned Stark 是否还活着？")

# chain("Ned Stark 是否还活着？") # 返回包含 AQL 查询和 AQL 结果的字典

    
    
    > 进入新的 ArangoGraphQAChain 链...
    AQL 查询 (1):
    RETURN DOCUMENT('Characters/NedStark').alive
    
    AQL 结果:
    [True]
    
    > 链执行完毕.





    '是的，根据数据库中的信息，Ned Stark 还活着。'

chain.run("Bran Stark 是否是 Ned Stark 的孩子？")

    
    
    > 进入新的 ArangoGraphQAChain 链...
    AQL 查询 (1):
    FOR e IN ChildOf
        FILTER e._from == "Characters/BranStark" AND e._to == "Characters/NedStark"
        RETURN e
    
    AQL 结果:
    [{'_key': '266218884026', '_id': 'ChildOf/266218884026', '_from': 'Characters/BranStark', '_to': 'Characters/NedStark', '_rev': '_gVPKGSq--_'}]
    
    > 链执行完毕.





    '是的，根据 ArangoDB 数据库中的信息，Bran Stark 确实是 Ned Stark 的孩子。'