Neo4j Schema Query Builder Integration #520

shahafp · 2023-09-14T11:46:09Z

Summary:

This pull request introduces the Neo4jQueryToolSpec class which provides an innovative way to query a Neo4j graph database based on a provided schema definition. It incorporates a language model to dynamically generate Cypher queries from user questions and is equipped with a self-healing mechanism for Cypher syntax errors.

Changes:

Added Neo4jQueryToolSpec class to enable schema-based querying.
Integrated a language model for dynamic and accurate Cypher query generation.
Implemented a self-healing mechanism to recover from Cypher syntax errors.
Included a README.md to guide users on installation, usage, features, and known limitations.

Testing:

Ensure the Neo4j database is set up and running. Initialize the Neo4jQueryToolSpec with appropriate database credentials and a valid language model, then run various queries to test its functionality and error-handling capabilities.

Impact:

This tool will significantly enhance the user's experience by facilitating intuitive and schema-aware querying of the Neo4j database, while also providing robust error handling.

jerryjliu

this is awesome! i'm really excited for this.

sorry for the delay in reviewing, a few comments. would be excited to help get this out

jerryjliu · 2023-09-18T06:41:01Z

llama_hub/tools/neo4j_db/README.md

@@ -0,0 +1,67 @@
+
+---


nit: remove these top-level lines

jerryjliu · 2023-09-18T06:42:47Z

llama_hub/tools/neo4j_db/base.py

+    This class is responsible for querying a Neo4j graph database based on a provided schema definition.
+    """
+
+    def __init__(self, url, user, password, llm):


can we use the native llamaindex llm abstractions? they include our native integrations (openai, anthropic, etc.) and also integrate with langchain llms:

https://gpt-index.readthedocs.io/en/latest/core_modules/model_modules/llms/root.html

jerryjliu · 2023-09-18T06:43:52Z

llama_hub/tools/neo4j_db/README.md

+from llama_hub.tools.neo4j_db.base import Neo4jQueryToolSpec
+from langchain.chat_models import ChatOpenAI
+
+llm = ChatOpenAI(


as mentioned below, would be good to have a native llamaindex demo e.g.

from llama_index.llms import OpenAI llm = OpenAI(model="gpt-4", ...) query_builder = Neo4jQueryToolSpec('url', 'user', 'password', llm)

jerryjliu

Actually a lot of this logic seems duplicated from our Neo4jGraphStore abstraction in the core LlamaIndex repo. Is there a way we can reuse code from there? E.g. define the tool spec, but just call Neo4jGraphStore under the hood instead of re-implementing the functions here

shahafp · 2023-09-18T07:28:09Z

Actually a lot of this logic seems duplicated from our Neo4jGraphStore abstraction in the core LlamaIndex repo. Is there a way we can reuse code from there? E.g. define the tool spec, but just call Neo4jGraphStore under the hood instead of re-implementing the functions here

Hey @jerryjliu
Thanks for your feedback!!
I will definitely will re-use your abstraction and update the pull :)

Use Neo4jGraphStore class instead of new graph database

shahafp · 2023-09-18T10:38:41Z

@jerryjliu I have changed the pull,
I didn't change the query function to the Neo4jGraphStore query because I needed the result keys, I can create another PR to the llama_index repo to add a flag to return the keys from the function if required.

Use Neo4jGraphStore class instead of new graph database

# Conflicts: # llama_hub/tools/neo4j_db/base.py

jerryjliu

thanks @shahafp, sorry for the delay, some final comments. maybe a general question here is do you have a notebook to show use cases?

the original intent of the tool spec was so that it could be used by agents, but curious how you've been using this so far!

llama_hub/tools/neo4j_db/base.py

add spec_functions fix string formatting

shahafp · 2023-09-23T17:33:13Z

thanks @shahafp, sorry for the delay, some final comments. maybe a general question here is do you have a notebook to show use cases?

the original intent of the tool spec was so that it could be used by agents, but curious how you've been using this so far!

Yeah, maybe I kind missed the whole point of the tool at first and thought perhaps it could be used also for RAG in the main repo of llama_index because it might help get information from the graph in a more accurate way.
anyway, I removed one of the functions.

I used this tool with the chat engine to get the information about the data from the graph and some reasoning.
Would you like me to add an example of usage with agent to the README file?

jerryjliu · 2023-09-23T18:37:03Z

Yeah, maybe I kind missed the whole point of the tool at first and thought perhaps it could be used also for RAG in the main repo of llama_index because it might help get information from the graph in a more accurate way.

no worries! it's good feedback for how we position these tools - maybe we should make it clear it can be used for both human and agent use :)

I used this tool with the chat engine to get the information about the data from the graph and some reasoning.

I'd actually be really interested if you're able to get an agent up and running with this

shahafp · 2023-09-23T20:17:24Z

I used this tool with the chat engine to get the information about the data from the graph and some reasoning.

I'd actually be really interested if you're able to get an agent up and running with this

Sure,
I used a simple agent chat with tools and queried the graph with it

llm_chat = OpenAI(
    openai_api_key="XXX-XXX",
    temperature=0,
    model='gpt-4',
)

gds_db = Neo4jQueryToolSpec(
    url="neo4j-url",
    user="neo4j-user",
    password="neo4j=password",
    llm=llm_chat,
    database='neo4j'
)

tools = gds_db.to_tool_list()
agent = OpenAIAgent.from_tools(tools, verbose=True)

# use agent
agent.chat("Where is JFK airport is located?")

Response:

'The port code JFK is located in New York, United States.'

Please let me know if this example is ok if so I will add it to the README file

jerryjliu · 2023-09-24T05:31:21Z

I used this tool with the chat engine to get the information about the data from the graph and some reasoning.

I'd actually be really interested if you're able to get an agent up and running with this

Sure, I used a simple agent chat with tools and queried the graph with it
llm_chat = OpenAI(
    openai_api_key="XXX-XXX",
    temperature=0,
    model='gpt-4',
)

gds_db = Neo4jQueryToolSpec(
    url="neo4j-url",
    user="neo4j-user",
    password="neo4j=password",
    llm=llm_chat,
    database='neo4j'
)

tools = gds_db.to_tool_list()
agent = OpenAIAgent.from_tools(tools, verbose=True)

# use agent
agent.chat("Where is JFK airport is located?")
Response:

'The port code JFK is located in New York, United States.'

Please let me know if this example is ok if so I will add it to the README file

yeah this works!

adriens · 2023-09-24T05:54:53Z

OMG the code seems so easy to use : do you think this feature could be released soon ? I'd like to use it at #nodes23 :

Btw, I already started to talk about llama_index on one of the datasets, but querying with llama_index could help compare with oter approaches.

What I like the most with llama_index is how simple code is 🤩 ... making it very helpful during demos.

shahafp · 2023-09-24T08:01:08Z

Thank you both for your feedback!
I have updated the README file with the agent initialization.
@jerryjliu please let me know if something else is missing 🙏

jerryjliu · 2023-09-24T19:04:32Z

@shahafp merged! https://llamahub.ai/l/tools-neo4j_db

what's your twitter handle? :) promoting soon

adriens · 2023-09-25T04:20:50Z

When will it be GA ?

shahafp · 2023-09-25T17:03:55Z

@shahafp merged! https://llamahub.ai/l/tools-neo4j_db

what's your twitter handle? :) promoting soon

That’s awesome!!! Thank you very much!
My twitter is @ParienteShahaf
I’m honored to be part of it 🤗

shahafpariente added 4 commits September 14, 2023 14:19

add new neo4j query schema based

b84f528

remove irrelevant imports

26c9253

change README file

a7a387e

change neo4j imports locations

25ed016

jerryjliu reviewed Sep 18, 2023

View reviewed changes

update README file

23a6890

Use Neo4jGraphStore class instead of new graph database

shahafpariente added 5 commits September 18, 2023 13:43

update README file

81daedf

Use Neo4jGraphStore class instead of new graph database

Merge remote-tracking branch 'origin/main'

7559df9

# Conflicts: # llama_hub/tools/neo4j_db/base.py

remove un-necessary code blocks

4d9d1aa

remove un-necessary code blocks

75cab53

add BaseToolSpec inheritance

305e5b5

jerryjliu reviewed Sep 23, 2023

View reviewed changes

llama_hub/tools/neo4j_db/base.py Show resolved Hide resolved

llama_hub/tools/neo4j_db/base.py Outdated Show resolved Hide resolved

llama_hub/tools/neo4j_db/base.py Outdated Show resolved Hide resolved

remove un-necessary function

d3e7d8d

add spec_functions fix string formatting

Update README file

cada481

Update README file

6519ee0

jerryjliu approved these changes Sep 24, 2023

View reviewed changes

jerryjliu merged commit 8f24610 into run-llama:main Sep 24, 2023
2 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Neo4j Schema Query Builder Integration #520

Neo4j Schema Query Builder Integration #520

shahafp commented Sep 14, 2023

jerryjliu left a comment

jerryjliu Sep 18, 2023

jerryjliu Sep 18, 2023

jerryjliu Sep 18, 2023

jerryjliu left a comment

shahafp commented Sep 18, 2023

shahafp commented Sep 18, 2023

jerryjliu left a comment

shahafp commented Sep 23, 2023 •

edited

Loading

jerryjliu commented Sep 23, 2023

shahafp commented Sep 23, 2023

jerryjliu commented Sep 24, 2023

adriens commented Sep 24, 2023

shahafp commented Sep 24, 2023

jerryjliu commented Sep 24, 2023

adriens commented Sep 25, 2023 •

edited

Loading

shahafp commented Sep 25, 2023

Neo4j Schema Query Builder Integration #520

Neo4j Schema Query Builder Integration #520

Conversation

shahafp commented Sep 14, 2023

Summary:

Changes:

Testing:

Impact:

jerryjliu left a comment

Choose a reason for hiding this comment

jerryjliu Sep 18, 2023

Choose a reason for hiding this comment

jerryjliu Sep 18, 2023

Choose a reason for hiding this comment

jerryjliu Sep 18, 2023

Choose a reason for hiding this comment

jerryjliu left a comment

Choose a reason for hiding this comment

shahafp commented Sep 18, 2023

shahafp commented Sep 18, 2023

jerryjliu left a comment

Choose a reason for hiding this comment

shahafp commented Sep 23, 2023 • edited Loading

jerryjliu commented Sep 23, 2023

shahafp commented Sep 23, 2023

jerryjliu commented Sep 24, 2023

adriens commented Sep 24, 2023

shahafp commented Sep 24, 2023

jerryjliu commented Sep 24, 2023

adriens commented Sep 25, 2023 • edited Loading

shahafp commented Sep 25, 2023

shahafp commented Sep 23, 2023 •

edited

Loading

adriens commented Sep 25, 2023 •

edited

Loading