Personalised queries #31
Conversation
![orca-security-eu[bot]](https://avatars.githubusercontent.com/in/276244?s=60&v=4){kind=small}
There was a problem hiding this comment.
Orca Security Scan Summary
| Status | Check | Issues by priority | |
|---|---|---|---|
 Passed |
Infrastructure as Code |  0  0  0  0 |
View in Orca |
| Passed |
SAST | 0 0 0 0 |
View in Orca |
| Passed |
Secrets | 0 0 0 0 |
View in Orca |
| Passed |
Vulnerabilities | 0 0 0 0 |
View in Orca |
danmichaeljones
force-pushed
the
personalised_query
branch
from
f9ea440 to
547006d
Compare
danmichaeljones
force-pushed
the
personalised_query
branch
from
bff0042 to
4fc03ac
Compare
danmichaeljones
force-pushed
the
personalised_query
branch
from
34b4b11 to
d31606d
Compare
danmichaeljones
force-pushed
the
personalised_query
branch
from
d31606d to
3446521
Compare
danmichaeljones
force-pushed
the
personalised_query
branch
from
fe4ffbf to
c5d8b5b
Compare
danmichaeljones
force-pushed
the
personalised_query
branch
from
11b94bd to
700b913
Compare
| @@ -439,3 +441,29 @@ def exists( | |||
| ) | |||
|
|
|||
| return response.json()["persona_collection_exists"] | |||
|
|
|||
| def query( | |||
There was a problem hiding this comment.
In order to match the collections.query API, this method just returns a configured instance of PersonalizedQuery which implements all the query methods
| self.recent_interactions_count = recent_interactions_count | ||
| self.decay_rate = decay_rate | ||
|
|
||
| def near_text( |
There was a problem hiding this comment.
This is where all the query methods are implemented. The only difference between each method is what model to use for the query parameters - these models are a tagged union and all have a query_method field which disambiguates them on the backend.
| from weaviate.collections.classes.internal import ReturnProperties, ReturnReferences | ||
|
|
||
|
|
||
| class NearTextQueryParameters(BaseModel): |
There was a problem hiding this comment.
Finally, these are the query parameter models (all with query_method fields). The serialise_X annotations use pydantic's custom serialiser hooks (PlainSerializer) to turn that field into a dict the backend can deserialise back into that object.
| overfetch_factor: float = 1.5, | ||
| recent_interactions_count: int = 100, | ||
| decay_rate: float = 0.1, | ||
| ): |
There was a problem hiding this comment.
Can we add a return type hint here?
There was a problem hiding this comment.
Sure, done! 👍
| ) | ||
|
|
||
|
|
||
| class PersonalizedQuery: |
There was a problem hiding this comment.
I'm trying to figure out which new classes in this PR would need to be exported in weaviate_agents/classes/__init__.py.
The motivation for this being that with the extras structure we do in the main python client weaviate-client[agents] we have essentially a set of stubs inside that repo ( https://github.com/weaviate/weaviate-python-client/tree/main/weaviate/agents). These stubs allow us to not require changes in the main client for every change we do in this one but still allow for the types to exported if the agents extra is installed.
So for any types in weaviate_agents/personalization/classes/__init__.py that might be used for typing reasons etc in future we might need to bring those up one level.
There was a problem hiding this comment.
I've tried to follow the existing exports, so have just added PersonalizedQueryResponse (the output from all the query methods). I've skipped PersonalizedQuery (the class which implements all the query methods) since that's not a plain model like the rest of these exports (and is importable from personalization_agent.py like the PersonalizationAgent class) but can add this if you think it's needed.
There was a problem hiding this comment.
There were also two definitions of Usage in personalization and query that had all the same fields/types but were different definitions so wouldn't have been recognised as the same types - I've moved these into a shared core classes file, but all the external export paths should be the same.
| json=self._get_request_data(query_parameters), | ||
| timeout=self.timeout, | ||
| ) | ||
| response_data = response.raise_for_status().json() |
There was a problem hiding this comment.
raise_for_status only returns the status code message to the user.
And we often return detailed error messages in response body.
So most flexible way to handle errors is like we do in other places:
if response.is_error:
raise Exception(response.text)
There was a problem hiding this comment.
Oops, I thought these were equivalent, didn't realise that raise_for_status swallows the error text - changed 👍
Closes https://github.com/weaviate/agents/issues/416.
This adds the personlized queries to the PersonalizationAgent client. The
query()method on the PA returns an instance ofPersonalizedQuerywhich implements methods matching the existing collections query methods.See the corresponding backend changes here.