ResultView

ResultView is the common return type for search, query, head, tail, range, and data retrieval APIs.

Result shapes

Search results contain IDs and distances:

result = collection.search(query, k=3, return_fields=True)

print(result.ids)        # np.ndarray[int64]
print(result.distances)  # np.ndarray[float32]
print(result.fields)     # list[dict]

Data results contain vectors:

result = collection.query_vectors(filter_ids=[1, 2, 3])

print(result.ids)
print(result.vectors)    # np.ndarray[float32], shape (n, dim)
print(result.fields)

Query results contain IDs and optional fields:

result = collection.query(where="lang = 'en'")

print(result.ids)
print(result.fields)

The main result types are:

Result type	Produced by	Main attributes
search	search, batch_search, search_range, text_search, search_sparse, hybrid_search	ids, distances, optional fields
query	query	ids, optional fields
data	query_vectors, head, tail, remote read_by_only_id	vectors, ids, fields

distances may be lower-is-better distances or higher-is-better scores depending on the metric. See the indexing and search tutorials for metric semantics.

Tuple unpacking

Search results unpack as (ids, distances, fields):

ids, distances, fields = collection.search(query, k=3, return_fields=True)

Data results unpack as (vectors, ids, fields):

vectors, ids, fields = collection.query_vectors(filter_ids=[1, 2])

Prefer attributes when clarity matters.

Row iteration

ResultView is not a row iterator. Convert to rows first:

for row in result.to_list():
    print(row)

to_list() returns dictionaries such as:

[
    {"id": 1, "distance": 0.0, "title": "intro"},
    {"id": 2, "distance": 0.01, "title": "guide"},
]

Conversion helpers

result.to_tuple()
result.to_numpy()
result.to_dict()
result.to_list()
result.to_json()

Typical use:

ids = result.ids.tolist()
rows = result.to_list()
payload = result.to_json()

Use to_numpy() for numerical post-processing and to_list() for application responses or debug prints.

Optional conversions:

result.to_pandas()  # requires pandas
result.to_polars()  # requires polars
result.to_arrow()   # requires pyarrow

Indexing by component

Use string keys to access components:

result["ids"]
result["distances"]
result["fields"]
result["vectors"]

Integer row indexing is intentionally not the main access path. Use result.to_list()[0] when you need the first row as a dict.

Empty results

Empty results are still valid ResultView objects:

result = collection.query(where="lang = 'missing'")

print(len(result))       # 0
print(bool(result))      # False
print(result.ids)        # empty int64 array
print(result.to_list())  # []

This lets application code handle "no match" without special exception paths.