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)
if not ok:
param = {
'collection_name': collection_name,
'dimension': _DIM,
'index_file_size': _INDEX_FILE_SIZE, # optional
'metric_type': MetricType.L2 # optional
}
milvus.create_collection(param)
# Show collections in Milvus server
_, collections = milvus.list_collections()
# Describe demo_collection
_, collection = milvus.get_collection_info(collection_name)
print(collection)
# 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
status, ids = milvus.insert(collection_name=collection_name,
records=vectors)
if not status.OK():
print("Insert failed: {}".format(status))
# Flush collection inserted data to disk.
milvus.flush([collection_name])
# Get demo_collection row count
status, result = milvus.count_entities(collection_name)
# present collection statistics info
_, info = milvus.get_collection_stats(collection_name)
print(info)
# Obtain raw vectors by providing vector ids
status, result_vectors = milvus.get_entity_by_id(collection_name, ids[:10])
# create index of vectors, search more rapidly
index_param = {'nlist': 2048}
# 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, IndexType.IVF_FLAT,
index_param)
# describe index, get information of index
status, index = milvus.get_index_info(collection_name)
print(index)
# Use the top 10 vectors for similarity search
query_vectors = vectors[0:10]
# execute vector similarity search
search_param = {"nprobe": 16}
print("Searching ... ")
param = {
'collection_name': collection_name,
'query_records': query_vectors,
'top_k': 1,
'params': search_param,
}
status, results = milvus.search(**param)
if status.OK():
# 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')
# print results
print(results)
else:
print("Search failed. ", status)
# Delete demo_collection
status = milvus.drop_collection(collection_name)
#建立分区
milvus.create_partition('test01', 'tag01')
#nlist聚类为多少簇
ivf_param = {'nlist': 16}
milvus.create_index('test01', IndexType.IVF_FLAT, ivf_param) #可以不写这个，默认方式IndexType.FLAT
# 随机生成一批向量数据
vectors_array = np.random.rand(num_vec,vec_dim)
vectors_list = vectors_array.tolist()
vectors_list.append([1 for _ in range(vec_dim)])
ids_list = [i for i in range(len(vectors_list))]
#单次插入的数据量不能大于 256 MB，插入后存在缓存区，缓存区大小由参数index_file_size决定，默认1024M
milvus.insert(collection_name='test01', records=vectors_list, partition_tag="tag01",ids=ids_list)
#一些信息
print(milvus.list_collections())
print(milvus.get_collection_info('test01'))
print(milvus.get_index_info('test01'))
t2 = time.time()
print("create cost:",t2-t1)
# # 创建查询向量
query_vec_list = [[1 for _ in range(vec_dim)]]
# 进行查询, 注意这里的参数nprobe和建立索引时的参数nlist 会因为索引类型不同而影响到查询性能和查询准确率
#IVF_FLAT下查询多少个簇，不能超过nlist
search_param = {'nprobe': 10}
#现在数据还在内存，需要数据落盘，保存到数据库中去，不然查不到数据
milvus.flush(collection_name_array=['test01'])
results = milvus.search(collection_name='test01', query_records=query_vec_list, top_k=10,params=search_param)
print(results)
print("search cost:",time.time()-t2)
#
# sudo docker run -d --name milvus_cpu \
def main():
milvus = Milvus(uri=uri)
param = {
'collection_name': collection_name,
'dimension': _DIM,
'index_file_size': 32,
#'metric_type': MetricType.IP
'metric_type': MetricType.L2
}
# show collections in Milvus server
_, collections = milvus.list_collections()
# 创建 collection
milvus.create_collection(param)
# 创建 collection partion
milvus.create_partition(collection_name, partition_tag)
print(f'collections in Milvus: {collections}')
# Describe demo_collection
_, collection = milvus.get_collection_info(collection_name)
print(f'descript demo_collection: {collection}')
# build fake vectors
vectors = [[random.random() for _ in range(_DIM)] for _ in range(10)]
vectors1 = [[random.random() for _ in range(_DIM)] for _ in range(10)]
status, id = milvus.insert(collection_name=collection_name,
records=vectors,
ids=list(range(10)),
partition_tag=partition_tag)
print(f'status: {status} | id: {id}')
if not status.OK():
print(f"insert failded: {status}")
status1, id1 = milvus.insert(collection_name=collection_name,
records=vectors1,
ids=list(range(10, 20)),
partition_tag=partition_tag)
print(f'status1: {status1} | id1: {id1}')
ids_deleted = list(range(10))
status_delete = milvus.delete_entity_by_id(collection_name=collection_name,
id_array=ids_deleted)
if status_delete.OK():
print(f'delete successful')
# Flush collection insered data to disk
milvus.flush([collection_name])
# Get demo_collection row count
status, result = milvus.count_entities(collection_name)
print(f"demo_collection row count: {result}")
# Obtain raw vectors by providing vector ids
status, result_vectors = milvus.get_entity_by_id(collection_name,
list(range(10, 20)))
# create index of vectors, search more repidly
index_param = {'nlist': 2}
# create ivflat index in demo_collection
status = milvus.create_index(collection_name, IndexType.IVF_FLAT,
index_param)
if status.OK():
print(f"create index ivf_flat succeeed")
# use the top 10 vectors for similarity search
query_vectors = vectors1[0:2]
# execute vector similariy search
search_param = {"nprobe": 16}
param = {
'collection_name': collection_name,
'query_records': query_vectors,
'top_k': 1,
'params': search_param
}
status, results = milvus.search(**param)
if status.OK():
if results[0][0].distance == 0.0:
print('query result is correct')
else:
print('not correct')
print(results)
else:
print(f'search failed: {status}')
# 清除已经存在的collection
milvus.drop_collection(collection_name=collection_name)
milvus.close()
"params": {
"dim": 32
}
},
],
"segment_row_limit":
16384,
"auto_id":
False
}
client.create_collection(collection_name, collection_param)
client.create_partition(collection_name, "Movie")
print("--------get collection info--------")
collection = client.get_collection_info(collection_name)
pprint(collection)
partitions = client.list_partitions(collection_name)
print("\n----------list partitions----------")
pprint(partitions)
import codecs
with codecs.open("movie_vectors.txt", "r", encoding='utf-8',
errors='ignore') as f:
lines = f.readlines()
ids = [int(line.split(":")[0]) for line in lines]
embeddings = []
for line in lines:
line = line.strip().split(":")[1][1:-1]
str_nums = line.split(",")
def main():
# Specify server addr when create milvus client instance
milvus = Milvus(_HOST, _PORT)
# Create collection demo_collection if it dosen't exist.
collection_name = 'example_async_collection_'
status, ok = milvus.has_collection(collection_name)
if not ok:
param = {
'collection_name': collection_name,
'dimension': _DIM,
'index_file_size': 128, # optional
'metric_type': MetricType.L2 # optional
}
status = milvus.create_collection(param)
if not status.OK():
print("Create collection failed: {}".format(status.message), file=sys.stderr)
print("exiting ...", file=sys.stderr)
sys.exit(1)
# Show collections in Milvus server
_, collections = milvus.list_collections()
# Describe demo_collection
_, collection = milvus.get_collection_info(collection_name)
print(collection)
# 10000 vectors with 16 dimension
# element per dimension is float32 type
# vectors should be a 2-D array
vectors = [[random.random() for _ in range(_DIM)] for _ in range(100000)]
# You can also use numpy to generate random vectors:
# `vectors = np.random.rand(10000, 16).astype(np.float32)`
def _insert_callback(status, ids):
if status.OK():
print("Insert successfully")
else:
print("Insert failed.", status.message)
# Insert vectors into demo_collection, adding callback function
insert_future = milvus.insert(collection_name=collection_name, records=vectors, _async=True,
_callback=_insert_callback)
# Or invoke result() to get results:
# insert_future = milvus.insert(collection_name=collection_name, records=vectors, _async=True)
# status, ids = insert_future.result()
insert_future.done()
# Flush collection inserted data to disk.
def _flush_callback(status):
if status.OK():
print("Flush successfully")
else:
print("Flush failed.", status.message)
flush_future = milvus.flush([collection_name], _async=True, _callback=_flush_callback)
# Or invoke result() to get results:
# flush_future = milvus.flush([collection_name], _async=True)
# status = flush_future.result()
flush_future.done()
def _compact_callback(status):
if status.OK():
print("Compact successfully")
else:
print("Compact failed.", status.message)
compact_furure = milvus.compact(collection_name, _async=True, _cakkback=_compact_callback)
# Or invoke result() to get results:
# compact_future = milvus.compact(collection_name, _async=True)
# status = compact_future.result()
compact_furure.done()
# Get demo_collection row count
status, result = milvus.count_entities(collection_name)
# present collection info
_, info = milvus.get_collection_stats(collection_name)
print(info)
# create index of vectors, search more rapidly
index_param = {
'nlist': 2048
}
def _index_callback(status):
if status.OK():
print("Create index successfully")
else:
print("Create index failed.", status.message)
# 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))
index_future = milvus.create_index(collection_name, IndexType.IVF_FLAT, index_param, _async=True,
_callback=_index_callback)
# Or invoke result() to get results:
# index_future = milvus.create_index(collection_name, IndexType.IVF_FLAT, index_param, _async=True)
# status = index_future.result()
index_future.done()
# describe index, get information of index
status, index = milvus.get_index_info(collection_name)
print(index)
# Use the top 10 vectors for similarity search
query_vectors = vectors[0:10]
# execute vector similarity search
search_param = {
"nprobe": 16
}
print("Searching ... ")
def _search_callback(status, results):
# if status.OK():
# print("Search successfully")
# else:
# print("Search failed.", status.message)
if status.OK():
# 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')
# print results
print(results)
else:
print("Search failed. ", status)
param = {
'collection_name': collection_name,
'query_records': query_vectors,
'top_k': 1,
'params': search_param,
"_async": True,
"_callback": _search_callback
}
search_future = milvus.search(**param)
# Or invoke result() to get results:
#
# param = {
# 'collection_name': collection_name,
# 'query_records': query_vectors,
# 'top_k': 1,
# 'params': search_param,
# "_async": True,
# }
# search_future = milvus.search(param)
# status, results = index_future.result()
search_future.done()
# Delete demo_collection
status = milvus.drop_collection(collection_name)
def main():
milvus = Milvus(_HOST, _PORT)
num = random.randint(1, 100)
# Create collection demo_collection if it dosen't exist.
collection_name = 'example_hybrid_collection_{}'.format(num)
collection_param = {
"fields": [{
"field": "A",
"type": DataType.INT32
}, {
"field": "B",
"type": DataType.INT32
}, {
"field": "C",
"type": DataType.INT64
}, {
"field": "Vec",
"type": DataType.FLOAT_VECTOR,
"params": {
"dimension": 128,
"metric_type": "L2"
}
}],
"segment_size":
100
}
milvus.create_collection(collection_name, collection_param)
# milvus.create_partition(collection_name, "p_01", timeout=1800)
# pars = milvus.list_partitions(collection_name)
# ok = milvus.has_partition(collection_name, "p_01", timeout=1800)
# assert ok
# ok = milvus.has_partition(collection_name, "p_02")
# assert not ok
# for p in pars:
# if p == "_default":
# continue
# milvus.drop_partition(collection_name, p)
# milvus.drop_collection(collection_name)
# sys.exit(0)
A_list = [random.randint(0, 255) for _ in range(num)]
vec = [[random.random() for _ in range(128)] for _ in range(num)]
hybrid_entities = [{
"field": "A",
"values": A_list,
"type": DataType.INT32
}, {
"field": "B",
"values": A_list,
"type": DataType.INT32
}, {
"field": "C",
"values": A_list,
"type": DataType.INT64
}, {
"field": "Vec",
"values": vec,
"type": DataType.FLOAT_VECTOR
}]
ids = milvus.insert(collection_name, hybrid_entities)
milvus.flush([collection_name])
print("Flush ... ")
time.sleep(3)
milvus.delete_entity_by_id(collection_name, ids[:10])
# print("Get entity be id start ...... ")
# entities = milvus.get_entity_by_id(collection_name, ids[:1])
# et = entities.dict()
print("Create index ......")
milvus.create_index(collection_name, "Vec", "ivf_flat", {
"index_type": "IVF_FLAT",
"nlist": 100
})
print("Create index done.")
info = milvus.get_collection_info(collection_name)
query_hybrid = {
"bool": {
"must": [{
"term": {
"A": [1, 2, 5]
}
}, {
"range": {
"B": {
"GT": 1,
"LT": 100
}
}
}, {
"vector": {
"Vec": {
"topk": 10,
"query": vec[:1],
"params": {
"nprobe": 10
}
}
}
}],
},
}
# print("Start searach ..", flush=True)
# results = milvus.search(collection_name, query_hybrid)
# print(results)
#
# for r in list(results):
# print("ids", r.ids)
# print("distances", r.distances)
results = milvus.search(collection_name, query_hybrid, fields=["B"])
for r in list(results):
print("ids", r.ids)
print("distances", r.distances)
for rr in r:
print(rr.entity.get("B"))
# for result in results:
# for r in result:
# print(f"{r}")
# itertor entity id
# for result in results:
# for r in result:
# # get distance
# dis = r.distance
# id_ = r.id
# # obtain all field name
# fields = r.entity.fields
# for f in fields:
# # get field value by field name
# # fv = r.entity.
# fv = r.entity.value_of_field(f)
# print(fv)
milvus.drop_collection(collection_name)
def main():
# Connect to Milvus server
# You may need to change _HOST and _PORT accordingly
param = {'host': _HOST, 'port': _PORT}
# You can create a instance specified server addr and
# invoke rpc method directly
client = Milvus(**param)
# Create collection demo_collection if it dosen't exist.
collection_name = 'demo_partition_collection'
partition_tag = "random"
# create collection
param = {
'collection_name': collection_name,
'dimension': _DIM,
'index_file_size': _INDEX_FILE_SIZE, # optional
'metric_type': MetricType.L2 # optional
}
client.create_collection(param)
# Show collections in Milvus server
_, collections = client.list_collections()
# Describe collection
_, collection = client.get_collection_info(collection_name)
print(collection)
# create partition
client.create_partition(collection_name, partition_tag=partition_tag)
# display partitions
_, partitions = client.list_partitions(collection_name)
# 10000 vectors with 16 dimension
# element per dimension is float32 type
# vectors should be a 2-D array
vectors = [[random.random() for _ in range(_DIM)] for _ in range(10000)]
# You can also use numpy to generate random vectors:
# `vectors = np.random.rand(10000, 16).astype(np.float32).tolist()`
# Insert vectors into partition of collection, return status and vectors id list
status, ids = client.insert(collection_name=collection_name, records=vectors, partition_tag=partition_tag)
# Wait for 6 seconds, until Milvus server persist vector data.
time.sleep(6)
# Get demo_collection row count
status, num = client.count_entities(collection_name)
# create index of vectors, search more rapidly
index_param = {
'nlist': 2048
}
# Create ivflat index in demo_collection
# You can search vectors without creating index. however, Creating index help to
# search faster
status = client.create_index(collection_name, IndexType.IVF_FLAT, index_param)
# describe index, get information of index
status, index = client.get_index_info(collection_name)
print(index)
# Use the top 10 vectors for similarity search
query_vectors = vectors[0:10]
# execute vector similarity search, search range in partition `partition1`
search_param = {
"nprobe": 10
}
param = {
'collection_name': collection_name,
'query_records': query_vectors,
'top_k': 1,
'partition_tags': ["random"],
'params': search_param
}
status, results = client.search(**param)
if status.OK():
# 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')
# print results
print(results)
# Drop partition. You can also invoke `drop_collection()`, so that all of partitions belongs to
# designated collections will be deleted.
status = client.drop_partition(collection_name, partition_tag)
# Delete collection. All of partitions of this collection will be dropped.
status = client.drop_collection(collection_name)
class MyMilvus():
def __init__(self, name, host, port, collection_param):
self.host = host
self.port = port
self.client = Milvus(host, port)
self.collection_name = name
self.collection_param = collection_param
# self.collection_param = {
# "fields": [
# {"name": "release_year", "type": DataType.INT32},
# {"name": "embedding", "type": DataType.FLOAT_VECTOR, "params": {"dim": 8}},
# ],
# "segment_row_limit": 4096,
# "auto_id": False
# }
def create_collection(self):
if self.collection_name not in self.client.list_collections():
# self.client.drop_collection(self.collection_name)
self.client.create_collection(self.collection_name, self.collection_param)
# ------
# Basic create index:
# Now that we have a collection in Milvus with `segment_row_limit` 4096, we can create index or
# insert entities.
#
# We can call `create_index` BEFORE we insert any entities or AFTER. However Milvus won't actually
# start build index task if the segment row count is smaller than `segment_row_limit`. So if we want
# to make Milvus build index, we need to insert number of entities larger than `segment_row_limit`.
#
# We are going to use data in `films.csv` so you can checkout the structure. And we need to group
# data with same fields together, so here is a example of how we obtain the data in files and transfer
# them into what we need.
# ------
ids = [] # ids
titles = [] # titles
release_years = [] # release year
embeddings = [] # embeddings
films = []
with open('films.csv', 'r') as csvfile:
reader = csv.reader(csvfile)
films = [film for film in reader]
for film in films:
ids.append(int(film[0]))
titles.append(film[1])
release_years.append(int(film[2]))
embeddings.append(list(map(float, film[3][1:][:-1].split(','))))
hybrid_entities = [
{"name": "release_year", "values": release_years, "type": DataType.INT32},
{"name": "embedding", "values": embeddings, "type": DataType.FLOAT_VECTOR},
]
# ------
# Basic insert:
# After preparing the data, we are going to insert them into our collection.
# The number of films inserted should be 8657.
# ------
ids = self.client.insert(self.collection_name, hybrid_entities, ids)
self.client.flush([self.collection_name])
after_flush_counts = self.client.count_entities(self.collection_name)
print(" > There are {} films in collection `{}` after flush".format(after_flush_counts, self.collection_name))
# ------
# Basic create index:
# Now that we have inserted all the films into Milvus, we are going to build index with these data.
#
# While building index, we have to indicate which `field` to build index for, the `index_type`,
# `metric_type` and params for the specific index type. In our case, we want to build a `IVF_FLAT`
# index, so the specific params are "nlist". See pymilvus documentation
# (https://milvus-io.github.io/milvus-sdk-python/pythondoc/v0.3.0/index.html) for `index_type` we
# support and the params accordingly.
#
# If there are already index for a collection and you call `create_index` with different params, the
# older index will be replaced by new one.
# ------
self.client.create_index(self.collection_name, "embedding",
{"index_type": "IVF_FLAT", "metric_type": "L2", "params": {"nlist": 100}})
# ------
# Basic create index:
# We can get the detail of the index by `get_collection_info`.
# ------
info = self.client.get_collection_info(self.collection_name)
pprint(info)
# ------
# Basic hybrid search entities:
# If we want to use index, the specific index params need to be provided, in our case, the "params"
# should be "nprobe", if no "params" given, Milvus will complain about it and raise a exception.
# ------
# query_embedding = [random.random() for _ in range(8)]
# query_hybrid = {
# "bool": {
# "must": [
# {
# "term": {"release_year": [2002, 1995]}
# },
# {
# "vector": {
# "embedding": {"topk": 3,
# "query": [query_embedding],
# "metric_type": "L2",
# "params": {"nprobe": 8}}
# }
# }
# ]
# }
# }
# ------
# Basic hybrid search entities
# ------
# results = client.search(collection_name, query_hybrid, fields=["release_year", "embedding"])
# for entities in results:
# for topk_film in entities:
# current_entity = topk_film.entity
# print("==")
# print("- id: {}".format(topk_film.id))
# print("- title: {}".format(titles[topk_film.id]))
# print("- distance: {}".format(topk_film.distance))
#
# print("- release_year: {}".format(current_entity.release_year))
# print("- embedding: {}".format(current_entity.embedding))
# ------
# Basic delete index:
# You can drop index for a field.
# ------
self.client.drop_index(self.collection_name, "embedding")
if self.collection_name in self.client.list_collections():
self.client.drop_collection(self.collection_name)