"anns_field": "embedding",
"param": {
"metric_type": "L2",
"params": {
"nprobe": 8
}
},
"limit": 3,
"output_fields": ["release_year"],
"expression": "release_year in [1995, 2002]",
}
# ------
# Basic hybrid search entities
# ------
results = client.search_with_expression(collection_name, **search_param)
for entities in results:
for topk_film in entities:
current_entity = topk_film.entity
print("==")
print(f"- id: {topk_film.id}")
print(f"- title: {titles[topk_film.id]}")
print(f"- distance: {topk_film.distance}")
print(f"- release_year: {current_entity.release_year}")
# ------
# Basic delete index:
# You can drop index we create
# ------
client.drop_index(collection_name, "embedding")
],
}, orm=True)
assert c.has_collection(collection_name)
ids = c.insert(collection_name, [
{"name": "f1", "type": DataType.FLOAT_VECTOR, "values": [[1.1, 2.2, 3.3, 4.4], [5.5, 6.6, 7.7, 8.8]]},
{"name": "age", "type": DataType.FLOAT, "values": [3.45, 8.9]}
], orm=True)
c.flush([collection_name])
c.load_collection(collection_name)
#############################################################
search_params = {"metric_type": "L2", "params": {"nprobe": 1}}
results = c.search_with_expression(collection_name, [[1.1, 2.2, 3.3, 4.4]],
"f1", param=search_params, limit=2, output_fields=["id"])
print("search results: ", results[0][0].entity, " + ", results[0][1].entity)
#############################################################
ids_expr = ",".join(str(x) for x in ids.primary_keys)
expr = "id in [ " + ids_expr + " ] "
print(expr)
pprint(c.query(collection_name, expr))
def main():
# Specify server addr when create milvus client instance
# milvus client instance maintain a connection pool, param
# `pool_size` specify the max connection num.
milvus = Milvus(_HOST, _PORT)
# Create collection demo_collection if it dosen't exist.
collection_name = 'example_collection'
ok = milvus.has_collection(collection_name)
field_name = 'example_field'
id_name = "id"
if not ok:
fields = {"fields": [{
"name": field_name,
"type": DataType.FLOAT_VECTOR,
"metric_type": "L2",
"params": {"dim": _DIM},
"indexes": [{"metric_type": "L2"}]
},
{
"name": id_name,
"type": DataType.INT64,
"auto_id": True,
"is_primary": True,
}
]}
milvus.create_collection(collection_name=collection_name, fields=fields)
else:
milvus.drop_collection(collection_name=collection_name)
# Show collections in Milvus server
collections = milvus.list_collections()
print(collections)
# Describe demo_collection
stats = milvus.get_collection_stats(collection_name)
print(stats)
# 10000 vectors with 128 dimension
# element per dimension is float32 type
# vectors should be a 2-D array
vectors = [[random.random() for _ in range(_DIM)] for _ in range(10)]
print(vectors)
# You can also use numpy to generate random vectors:
# vectors = np.random.rand(10000, _DIM).astype(np.float32)
# Insert vectors into demo_collection, return status and vectors id list
entities = [{"name": field_name, "type": DataType.FLOAT_VECTOR, "values": vectors}]
res_ids = milvus.insert(collection_name=collection_name, entities=entities)
print("ids:", res_ids.primary_keys)
# Flush collection inserted data to disk.
milvus.flush([collection_name])
# present collection statistics info
stats = milvus.get_collection_stats(collection_name)
print(stats)
# create index of vectors, search more rapidly
index_param = {
"metric_type": "L2",
"index_type": "IVF_FLAT",
"params": {"nlist": 1024}
}
# Create ivflat index in demo_collection
# You can search vectors without creating index. however, Creating index help to
# search faster
print("Creating index: {}".format(index_param))
status = milvus.create_index(collection_name, field_name, index_param)
# execute vector similarity search
print("Searching ... ")
dsl = {"bool": {"must": [{"vector": {
field_name: {
"metric_type": "L2",
"query": vectors,
"topk": 10,
"params": {"nprobe": 16}
}
}}]}}
search_params = {"metric_type": "L2", "params": {"nprobe": 10}}
milvus.load_collection(collection_name)
results = milvus.search_with_expression(collection_name, vectors, field_name, param=search_params, limit=10,
output_fields=[id_name])
print("search results: ", results[0][0].entity)
# indicate search result
# also use by:
# `results.distance_array[0][0] == 0.0 or results.id_array[0][0] == ids[0]`
if results[0][0].distance == 0.0 or results[0][0].id == ids[0]:
print('Query result is correct')
else:
print('Query result isn\'t correct')
milvus.drop_index(collection_name, field_name)
milvus.release_collection(collection_name)
# Delete demo_collection
status = milvus.drop_collection(collection_name)
