[main < T225] Add LLM util

[katarinasupe]

Description

A module that contains procedures describing graphs in a format best suited for large language models (LLMs).

Example usage: Get raw graph schema: CALL llm_util.schema('raw') YIELD schema RETURN schema; Get prompt-ready graph schema: CALL llm_util.schema() YIELD schema RETURN schema; or CALL llm_util.schema('prompt_ready') YIELD schema RETURN schema;

TODO:

• [x] Update MAGE README

Changelog message: Now you can generate graph schema in a format best suited for large language models (LLMs).

Pull request type

• [x] Algorithm/Module

######################################

Reviewer checklist (the reviewer checks this part)

Module/Algorithm

• [ ] Core algorithm/module implementation
• [ ] Query module implementation
• [ ] Unit tests
• [ ] End-to-end tests
• [ ] Code documentation
• [ ] README short description
• [ ] Documentation on memgraph/docs

######################################

[Josipmrden]

@katarinasupe check https://www.notion.so/memgraph/Workflow-e99f310ca5ec463a95a3b5594c04aac6 for workflow on naming of branches.

In your case [main < T225] Add LLM util

[Josipmrden]

can we get any information by reusing the results from meta_util.schema? or we need to do the process of extracting the schema again manually?

[katarinasupe]

@Josipmrden I could get some info from meta util. My first approach was to edit meta_util and see what I can do, but that took more time than implementing a new module. Meta_util is also counting properties and how many nodes have those properties and that is an overkill for this module. To call meta_util from this module is, in my opinion, also an overkill. To get properties that exist, counts need to be done too.

[katarinasupe]

I updated the module, restructured a code a bit and applied @Josipmrden and Brett's suggestions. Here are the screenshots of new usage.

Here are my comments and I also refer to Brett's review here:

• I added raw and prompt-ready as two possible llm schema options. Using raw, users can get all they need to create their version of the prompt-ready schema. I hope this answers versioning mechanism for update or you had something else in mind? We can add arguments to input parts of the output string, but if the user can easily concatenate those strings to the raw output, then I am not sure if it makes sense? But, I do get your point when it comes to testing. Maybe some specific parts of the prompt-ready string should be set as arguments?
• The id property came from the dataset; hence it is something user-defined, not an internal id. Do you think we should exclude any id property from all datasets?
• I changed the usage. Let me know if it's better: Get raw graph schema: CALL llm_util.schema('raw') YIELD schema RETURN schema; Get prompt-ready graph schema: CALL llm_util.schema() YIELD schema RETURN schema; or CALL llm_util.schema('prompt_ready') YIELD schema RETURN schema;
• I will have to test with multiple labels. For now, if node has multiple labels, for example, Person and Student, its label is :Person:Student and hence, there is a label in the dataset :Person:Student. If there is any other node in the dataset with only :Person label, then we will have also that kind of node. It will be something like this: Node name: Person:Student Node name: Person I need to check the exact output, put let me know what are your suggestions for multiple labels. For relationships, we would have [:Person]-[:FRIENDS_WITH]->(:Person:Student), [:Person]-[:FRIENDS_WITH]->(:Person) if there is at least one relationship of both types.
• Nodes can have multiple labels, and relationships can have only one type.
• I changed type to rel in code. I agree it can be confusing.

[katarinasupe]

I tested it on Europe gas pipelines dataset from Memgraph Lab (which does not have a pretty schema).

Here is the output:

Node properties are the following:
Node name: 'NODE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}]
Node name: 'BORDER_POINT:INTER_CONNECTION_POINT:NODE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'pipe_name', 'type': 'str'}]
Node name: 'NODE:STORAGE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'max_cap_store2pipe_M_m3_per_d', 'type': 'float'}, {'property': 'max_cap_pipe2store_M_m3_per_d', 'type': 'float'}, {'property': 'max_power_MW', 'type': 'float'}, {'property': 'min_storage_pressure_bar', 'type': 'int'}, {'property': 'max_workingGas_M_m3', 'type': 'float'}, {'property': 'max_storage_pressure_bar', 'type': 'int'}]
Node name: 'COMPRESSOR:NODE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'max_power_MW', 'type': 'float'}, {'property': 'max_cap_M_m3_per_d', 'type': 'float'}, {'property': 'max_pressure_bar', 'type': 'float'}]
Node name: 'NODE:PRODUCTION', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'max_production_M_m3_per_d', 'type': 'float'}]
Node name: 'ENTRY_POINT:NODE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}]
Node name: 'BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'pipe_name', 'type': 'str'}, {'property': 'max_power_MW', 'type': 'float'}, {'property': 'max_cap_M_m3_per_d', 'type': 'float'}, {'property': 'max_pressure_bar', 'type': 'int'}]
Node name: 'BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE:STORAGE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'pipe_name', 'type': 'str'}, {'property': 'max_cap_store2pipe_M_m3_per_d', 'type': 'float'}, {'property': 'max_cap_pipe2store_M_m3_per_d', 'type': 'float'}, {'property': 'max_power_MW', 'type': 'float'}, {'property': 'min_storage_pressure_bar', 'type': 'int'}, {'property': 'max_workingGas_M_m3', 'type': 'float'}, {'property': 'max_storage_pressure_bar', 'type': 'int'}, {'property': 'max_cap_M_m3_per_d', 'type': 'float'}, {'property': 'max_pressure_bar', 'type': 'float'}]
Node name: 'BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'pipe_name', 'type': 'str'}, {'property': 'max_power_MW', 'type': 'float'}, {'property': 'max_cap_M_m3_per_d', 'type': 'float'}, {'property': 'max_pressure_bar', 'type': 'float'}]
Node name: 'COMPRESSOR:NODE:STORAGE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'max_cap_store2pipe_M_m3_per_d', 'type': 'float'}, {'property': 'max_cap_pipe2store_M_m3_per_d', 'type': 'float'}, {'property': 'max_power_MW', 'type': 'float'}, {'property': 'min_storage_pressure_bar', 'type': 'float'}, {'property': 'max_workingGas_M_m3', 'type': 'float'}, {'property': 'max_storage_pressure_bar', 'type': 'float'}, {'property': 'max_cap_M_m3_per_d', 'type': 'float'}, {'property': 'max_pressure_bar', 'type': 'float'}]
Node name: 'BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE:STORAGE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'pipe_name', 'type': 'str'}, {'property': 'max_cap_store2pipe_M_m3_per_d', 'type': 'float'}, {'property': 'max_cap_pipe2store_M_m3_per_d', 'type': 'float'}, {'property': 'max_power_MW', 'type': 'float'}, {'property': 'min_storage_pressure_bar', 'type': 'float'}, {'property': 'max_workingGas_M_m3', 'type': 'float'}, {'property': 'max_storage_pressure_bar', 'type': 'float'}, {'property': 'max_cap_M_m3_per_d', 'type': 'float'}, {'property': 'max_pressure_bar', 'type': 'float'}]
Node name: 'COMPRESSOR:LNG:NODE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'max_cap_store2pipe_M_m3_per_d', 'type': 'float'}, {'property': 'max_power_MW', 'type': 'float'}, {'property': 'max_workingGas_M_m3', 'type': 'float'}, {'property': 'max_cap_M_m3_per_d', 'type': 'float'}, {'property': 'max_pressure_bar', 'type': 'float'}]
Node name: 'BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:NODE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'pipe_name', 'type': 'str'}]
Node name: 'BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:LNG:NODE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'pipe_name', 'type': 'str'}, {'property': 'max_cap_store2pipe_M_m3_per_d', 'type': 'float'}, {'property': 'max_workingGas_M_m3', 'type': 'float'}]
Node name: 'BORDER_POINT:INTER_CONNECTION_POINT:NODE:STORAGE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'pipe_name', 'type': 'str'}, {'property': 'max_cap_store2pipe_M_m3_per_d', 'type': 'float'}, {'property': 'max_cap_pipe2store_M_m3_per_d', 'type': 'float'}, {'property': 'max_power_MW', 'type': 'float'}, {'property': 'min_storage_pressure_bar', 'type': 'int'}, {'property': 'max_workingGas_M_m3', 'type': 'float'}, {'property': 'max_storage_pressure_bar', 'type': 'int'}]
Node name: 'LNG:NODE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'max_cap_store2pipe_M_m3_per_d', 'type': 'float'}, {'property': 'max_workingGas_M_m3', 'type': 'float'}]
Node name: 'ENTRY_POINT:NODE:STORAGE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'max_cap_store2pipe_M_m3_per_d', 'type': 'float'}, {'property': 'max_cap_pipe2store_M_m3_per_d', 'type': 'float'}, {'property': 'max_power_MW', 'type': 'float'}, {'property': 'min_storage_pressure_bar', 'type': 'float'}, {'property': 'max_workingGas_M_m3', 'type': 'float'}, {'property': 'max_storage_pressure_bar', 'type': 'float'}]
Node name: 'ENTRY_POINT:LNG:NODE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'max_cap_store2pipe_M_m3_per_d', 'type': 'float'}, {'property': 'max_workingGas_M_m3', 'type': 'float'}]
Node name: 'BORDER_POINT:INTER_CONNECTION_POINT:LNG:NODE', Node properties: [{'property': 'name', 'type': 'str'}, {'property': 'country_code', 'type': 'str'}, {'property': 'node_id', 'type': 'str'}, {'property': 'lat', 'type': 'float'}, {'property': 'lng', 'type': 'float'}, {'property': 'elevation_m', 'type': 'int'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'pipe_name', 'type': 'str'}, {'property': 'max_cap_store2pipe_M_m3_per_d', 'type': 'float'}, {'property': 'max_workingGas_M_m3', 'type': 'float'}]

Relationship properties are the following:
Relationship Name: 'PIPE', Relationship Properties: [{'property': 'name', 'type': 'str'}, {'property': 'operator_name', 'type': 'str'}, {'property': 'start_year', 'type': 'int'}, {'property': 'end_year', 'type': 'int'}, {'property': 'max_cap_M_m3_per_d', 'type': 'float'}, {'property': 'max_pressure_bar', 'type': 'int'}, {'property': 'pipe_id', 'type': 'str'}, {'property': 'diameter_mm', 'type': 'int'}, {'property': 'length_km', 'type': 'float'}, {'property': 'num_compressor', 'type': 'int'}]

The relationships are the following:
['(:NODE)-[:PIPE]->(:NODE)']
['(:BORDER_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:NODE)']
['(:NODE:STORAGE)-[:PIPE]->(:BORDER_POINT:INTER_CONNECTION_POINT:NODE)']
['(:NODE:STORAGE)-[:PIPE]->(:NODE)']
['(:NODE:STORAGE)-[:PIPE]->(:NODE:STORAGE)']
['(:NODE)-[:PIPE]->(:NODE:STORAGE)']
['(:NODE)-[:PIPE]->(:BORDER_POINT:INTER_CONNECTION_POINT:NODE)']
['(:NODE)-[:PIPE]->(:COMPRESSOR:NODE)']
['(:COMPRESSOR:NODE)-[:PIPE]->(:NODE)']
['(:COMPRESSOR:NODE)-[:PIPE]->(:BORDER_POINT:INTER_CONNECTION_POINT:NODE)']
['(:NODE:STORAGE)-[:PIPE]->(:COMPRESSOR:NODE)']
['(:NODE:STORAGE)-[:PIPE]->(:NODE:PRODUCTION)']
['(:COMPRESSOR:NODE)-[:PIPE]->(:NODE:STORAGE)']
['(:NODE)-[:PIPE]->(:NODE:PRODUCTION)']
['(:NODE:PRODUCTION)-[:PIPE]->(:NODE)']
['(:NODE)-[:PIPE]->(:ENTRY_POINT:NODE)']
['(:NODE)-[:PIPE]->(:COMPRESSOR:NODE:STORAGE)']
['(:ENTRY_POINT:NODE)-[:PIPE]->(:NODE)']
['(:NODE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE)']
['(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:NODE)']
['(:COMPRESSOR:NODE)-[:PIPE]->(:COMPRESSOR:NODE:STORAGE)']
['(:COMPRESSOR:NODE)-[:PIPE]->(:COMPRESSOR:NODE)']
['(:COMPRESSOR:NODE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE)']
['(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:COMPRESSOR:NODE)']
['(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:BORDER_POINT:INTER_CONNECTION_POINT:NODE)']
['(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE:STORAGE)-[:PIPE]->(:NODE)']
['(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE:STORAGE)-[:PIPE]->(:COMPRESSOR:NODE)']
['(:BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:COMPRESSOR:NODE:STORAGE)']
['(:BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:COMPRESSOR:NODE)']
['(:COMPRESSOR:NODE:STORAGE)-[:PIPE]->(:NODE)']
['(:COMPRESSOR:NODE:STORAGE)-[:PIPE]->(:COMPRESSOR:NODE)']
['(:COMPRESSOR:NODE)-[:PIPE]->(:LNG:NODE)']
['(:BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:NODE)']
['(:COMPRESSOR:NODE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE:STORAGE)']
['(:BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE:STORAGE)-[:PIPE]->(:NODE)']
['(:COMPRESSOR:NODE)-[:PIPE]->(:ENTRY_POINT:LNG:NODE)']
['(:COMPRESSOR:NODE:STORAGE)-[:PIPE]->(:BORDER_POINT:INTER_CONNECTION_POINT:NODE)']
['(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE:STORAGE)']
['(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:NODE:STORAGE)']
['(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE:STORAGE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE)']
['(:BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:LNG:NODE)']
['(:BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:ENTRY_POINT:NODE)']
['(:COMPRESSOR:NODE)-[:PIPE]->(:COMPRESSOR:LNG:NODE)']
['(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)']
['(:COMPRESSOR:NODE:STORAGE)-[:PIPE]->(:NODE:STORAGE)']
['(:COMPRESSOR:LNG:NODE)-[:PIPE]->(:COMPRESSOR:NODE)']
['(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:LNG:NODE)']
['(:COMPRESSOR:NODE:STORAGE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE:STORAGE)']
['(:COMPRESSOR:NODE:STORAGE)-[:PIPE]->(:ENTRY_POINT:NODE:STORAGE)']
['(:COMPRESSOR:NODE:STORAGE)-[:PIPE]->(:COMPRESSOR:NODE:STORAGE)']
['(:COMPRESSOR:NODE:STORAGE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)']
['(:COMPRESSOR:NODE)-[:PIPE]->(:BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)']
['(:BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE:STORAGE)-[:PIPE]->(:NODE:STORAGE)']
['(:COMPRESSOR:NODE)-[:PIPE]->(:ENTRY_POINT:NODE)']
['(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:LNG:NODE)']
['(:BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:NODE)']
['(:BORDER_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:COMPRESSOR:NODE)']
['(:BORDER_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE)']
['(:BORDER_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:BORDER_POINT:INTER_CONNECTION_POINT:NODE)']
['(:BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE:STORAGE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE)']
['(:BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:COMPRESSOR:NODE)']
['(:BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE)']
['(:BORDER_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:COMPRESSOR:NODE:STORAGE)']
['(:BORDER_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)']
['(:BORDER_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE:STORAGE)']
['(:BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:LNG:NODE)-[:PIPE]->(:NODE)']
['(:BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:LNG:NODE)-[:PIPE]->(:ENTRY_POINT:NODE:STORAGE)']
['(:BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)']
['(:LNG:NODE)-[:PIPE]->(:NODE)']
['(:COMPRESSOR:LNG:NODE)-[:PIPE]->(:NODE)']
['(:LNG:NODE)-[:PIPE]->(:BORDER_POINT:INTER_CONNECTION_POINT:NODE)']
['(:LNG:NODE)-[:PIPE]->(:COMPRESSOR:NODE:STORAGE)']
['(:LNG:NODE)-[:PIPE]->(:NODE:STORAGE)']
['(:LNG:NODE)-[:PIPE]->(:COMPRESSOR:NODE)']
['(:ENTRY_POINT:NODE:STORAGE)-[:PIPE]->(:COMPRESSOR:NODE:STORAGE)']
['(:NODE:STORAGE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE)']
['(:NODE:STORAGE)-[:PIPE]->(:BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)']
['(:NODE:STORAGE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE:STORAGE)']
['(:BORDER_POINT:INTER_CONNECTION_POINT:NODE:STORAGE)-[:PIPE]->(:COMPRESSOR:NODE)']
['(:BORDER_POINT:INTER_CONNECTION_POINT:NODE:STORAGE)-[:PIPE]->(:BORDER_POINT:INTER_CONNECTION_POINT:NODE:STORAGE)']
['(:BORDER_POINT:INTER_CONNECTION_POINT:NODE:STORAGE)-[:PIPE]->(:NODE:STORAGE)']
['(:NODE:STORAGE)-[:PIPE]->(:COMPRESSOR:NODE:STORAGE)']
['(:ENTRY_POINT:NODE:STORAGE)-[:PIPE]->(:BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)']
['(:ENTRY_POINT:NODE:STORAGE)-[:PIPE]->(:NODE)']
['(:ENTRY_POINT:NODE:STORAGE)-[:PIPE]->(:BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:LNG:NODE)']
['(:ENTRY_POINT:NODE:STORAGE)-[:PIPE]->(:COMPRESSOR:NODE)']
['(:ENTRY_POINT:NODE)-[:PIPE]->(:NODE:STORAGE)']
['(:ENTRY_POINT:NODE)-[:PIPE]->(:COMPRESSOR:NODE)']
['(:NODE)-[:PIPE]->(:BORDER_POINT:INTER_CONNECTION_POINT:NODE:STORAGE)']
['(:NODE)-[:PIPE]->(:BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)']
['(:BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:ENTRY_POINT:NODE:STORAGE)']
['(:BORDER_POINT:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:COMPRESSOR:NODE:STORAGE)']
['(:NODE:STORAGE)-[:PIPE]->(:LNG:NODE)']
['(:NODE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE)']
['(:LNG:NODE)-[:PIPE]->(:LNG:NODE)']
['(:BORDER_POINT:INTER_CONNECTION_POINT:LNG:NODE)-[:PIPE]->(:COMPRESSOR:NODE)']
['(:NODE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:INTER_CONNECTION_POINT:NODE:STORAGE)']
['(:NODE)-[:PIPE]->(:BORDER_POINT:COMPRESSOR:ENTRY_POINT:INTER_CONNECTION_POINT:NODE:STORAGE)']
['(:BORDER_POINT:INTER_CONNECTION_POINT:NODE)-[:PIPE]->(:LNG:NODE)']
['(:NODE)-[:PIPE]->(:LNG:NODE)']

[brettdbrewer]

• yes, I think having the 'raw" output can address my versioning question. A developer can always fall back to it and parse it how they want. And that is certainly easier than maintaining multiple 'prompt-ready' versions. My concern was simply around how fluid this space is, so making sure any API can adapt with that change.
• Re id properties: If they are user defined, then I assume you'd want to retain them in this schema as a user's natural language query may reference id's specifically in that use case.
• I like using the function param to determine raw or prompt-ready
• Regarding nodes having multiple labels: I'd want to test some variations so see which performs better on complex queries. I assume you'd want the Node Name to be representative as possible to how it would be used in a Cypher query. Btw, it appears that sometimes the parent is first and sometimes it is second? e.g. NODE:STORAGE, COMPRESSOR:NODE. Am I just reading it wrong?
• Regarding nodes with multiple labels in relationship strings: yes, I think we'll need to be explicit in having each of the permutations (because you don't know how a user will query it). I assume each would be its own row in the Relationships section of the prompt-ready string.
• Because this is such a new area, I highly recommend having a set of unit tests that test validity of LLM output across multiple LLMs and multiple databases for the prompt-ready string format. It will also be highly informative imho as I have constantly found edge cases and extreme quality diffs between LLMs and their versions.

[katarinasupe]

Thank you @brettdbrewer for the comments, I will check the ones on Discord too and copy anything useful to reply here.

• There is no parent connection regarding labels on nodes, again just a user defined thing - they defined label NODE:STORAGE or COMPRESSOR:NODE and that is how it was stored. This is a user's error because it would make more sense to have it logically in the right order.
• I will add tests definitely, will ask you more about that if needed, since I am not that familiar with different LLMs

[katarinasupe]

@brettdbrewer raised concerns regarding the capitalization of properties. Once we discuss this, tolower() might be added to all string properties.

[brettdbrewer]

I don't think tolower() will be needed as the issue is with property values, not property/node/relationship names. Said another way, the issue isn't with the schema definition, it is in the Cypher query generated by the LLM. E.g. for the query "Who was killed by magic?" in the GoT dataset, the LLM would need to create the following Cypher query "MATCH (killer:Character)-[k:KILLED]->(victim:Character) WHERE toLower(k.method) = 'magic' RETURN victim.name" to get the right answer since "Magic" is capitalized in the dataset for that relationship property. I think we can divorce this issue from the schema definition and address it as a solution problem any developer will have to (potentially) solve in their solution.

[katarinasupe]

I added a 'fix' for multiple labels according to @brettdbrewer's suggestion. Here is what it looks like for the above example now:

Besides that, I restructured the code a bit to follow the MAGE codebase and improved the docstring according to the documentation. I will update the documentation based on these changes.