Actions26
- Delete Collection
- Get Collection
- Check Collection Exists
- Create Collection
- Update Collection
- Retrieve Point
- Retrieve Points
- Upsert Points
- Delete Points
- Update Vectors
- Delete Vectors
- Set Payload
- Overwrite Payload
- Delete Payload
- Clear Payload
- Batch Update Points
- Scroll Points
- Count Points
- Query Points
- Query Batch Points
- Query Points Groups
- Matrix Pairs
- Matrix Offsets
- Payload Facets
- Create Payload Index
- Delete Payload Index
Overview
The node interfaces with Qdrant, a vector search engine, to perform various operations on collections of points (vectors). The "Matrix Offsets" operation specifically finds neighboring points for sampled points within a collection based on vector similarity. This is useful in scenarios like recommendation systems, anomaly detection, or clustering where you want to analyze relationships or distances between vectors.
For example, given a collection of user embeddings, the node can sample some users and find their closest neighbors according to vector offsets, helping identify similar users or items.
Properties
| Name | Meaning |
|---|---|
| Collection Name | The name of the collection in Qdrant to query. |
| Filter | JSON object defining conditions to filter which points are considered for sampling and neighbor search. |
| Sample | Number of points to randomly select from the filtered set to find neighbors for. |
| Limit | Maximum number of neighbors to return per sampled point. |
| Using | Name of the vector field to use for querying; if empty, the default vector is used. |
| Shard Key | JSON specifying shards to restrict the search to; if null, all shards are searched. |
| Consistency | Read consistency level for the operation: Majority, Quorum, or All. |
| Timeout | Request timeout in seconds, overriding global timeout if set. |
Output
The output contains a JSON structure representing the matrix offsets results. This typically includes, for each sampled point, a list of its nearest neighbors along with their offset distances or similarity scores. The exact structure depends on Qdrant's response format but generally maps sampled points to arrays of neighbor points and associated metrics.
No binary data output is indicated for this operation.
Dependencies
- Requires an API key credential for authenticating with the Qdrant service.
- Needs the Qdrant service URL configured in credentials.
- Relies on Qdrant's REST API endpoints to perform the matrix offsets query.
Troubleshooting
- Timeouts: If the operation times out, consider increasing the Timeout property or checking network connectivity.
- Invalid Filter or Shard Key JSON: Malformed JSON in Filter or Shard Key properties will cause errors; ensure valid JSON syntax.
- Collection Not Found: Specifying a non-existent collection name will result in errors; verify collection existence before querying.
- Insufficient Permissions: Ensure the API key has permissions to read from the specified collection.
- Empty Results: If no neighbors are found, check that the filter criteria are not too restrictive and that the collection contains enough points.