New Feature
-
Upsert function
Support users to update or insert data through the upsert interface. Known limitations, self-incrementing id does not support upsert; upsert is an internal implementation of delete + insert, so there will be a certain loss in performance. If you know clearly that the scene is writing data, please continue to use insert.
-
Range Search function
Support users to query by specifying the distance of search through input parameters, and return all the results within a certain distance from the target vector. For example:
// add radius and range_filter to params in search_params
search_params = {"params": {"nprobe": 10, "radius": 10, "range_filter" : 20}, "metric_type": "L2"}
res = collection.search(
vectors, "float_vector", search_params, topK,
"int64 > 100", output_fields=["int64", "float"]
)
In this example, vectors with distances between 10 and 20 are returned. Note that the difference