Support for filters in Vector and VectorCypher retrievers #29
Conversation
# Conflicts: # src/neo4j_genai/retrievers/hybrid.py # src/neo4j_genai/retrievers/vector.py
src/neo4j_genai/filters.py
Outdated
| self.node_alias = node_alias | ||
|
|
||
| def lhs(self, field): | ||
| return f"{self.node_alias}.`{field}`" |
There was a problem hiding this comment.
- Prefix any field name backtick with a backtick
- If you find non alphabetic characters in the field name, wrap the whole field name with backticks
name -> name
1name -> `1name`
na`me -> `na``me`
Example of valid Cypher:
MATCH (n:`Hell``o`) WHERE n.`prop``erty` = true RETURN n LIMIT 25
More info: https://neo4j.com/docs/cypher-manual/current/syntax/naming/#symbolic-names-escaping-rules
There was a problem hiding this comment.
Interesting, thanks! Wdyt about always wrapping the field name in backticks, whether it contains special characters or not? In your example it means that
name -> `name`
There was a problem hiding this comment.
I'd rather see that we don't do that. It makes the queries more noisy when debugging and teaches bad practices.
There was a problem hiding this comment.
I understand your point, and at the same time I'm a bit scared that we miss some special characters and produce invalid query "just" to make it nice. Let me check if we can find something really robust.
There was a problem hiding this comment.
See here, if you think the test cases are covering enough use cases: 2b89aff#diff-00b9f6996b286b97ed1b966f8596505071f2876cac256d97e31bb0de4449b198R53-R69
There was a problem hiding this comment.
Yeah, I see your concern. But if anyone should be able to do the right thing it should be us :)
This should give us confidence: https://github.com/neo4j/cypher-builder/blob/main/src/utils/escape.ts#L54-L63
There was a problem hiding this comment.
Fair enough, used the same regex to test the field validity.
# Conflicts: # src/neo4j_genai/retrievers/vector.py
Add support for pre-filtering for the Vector and VectorCypher retrievers