ClassifyDescriptor
Classify one finite float32 query vector using its nearest neighbors’ stored labels. Requires objects.read permission and one input blob; no output blob is produced.
engine (Flat/IVF/HNSW/ACORN) and positive k_neighbors are required. with_class optionally restricts one descriptor class. Omission searches compatible class indexes using the first common metric in L2/IP/CS order. This command does not accept a separate metric, pagination or result projection.
The most frequent stored string label wins. Ties use the best distance for a label, then its name. Missing/non-string labels do not vote. Success returns {"status":0,"label":"Alice"} under ClassifyDescriptor. With no usable vote, the label is the literal string "None".
See Descriptor classes and vector indexes for class compatibility and exact tie behavior, and Responses for errors.
[{
"ClassifyDescriptor": {
"engine": "IVF",
"with_class": "party_faces",
"k_neighbors": 10
}
}]
[{
"ClassifyDescriptor": {
"label": "bar",
"status": 0
}
}]