Skip to content

M
milvus

milvus / milvus
9 个月 前同步成功

通知 260
Star 22476
Fork 2472
M
milvus

前往新版Gitcode，体验更适合开发者的 AI 搜索 >>

未验证 提交 bafcbf5a 编写于 作者: B binbin 提交者： GitHub
浏览文件
操作

Update range search cases (#26372) 

Signed-off-by: Nbinbin lv <binbin.lv@zilliz.com>
上级 8b549142
隐藏空白更改
内联 并排
Showing with 51 addition and 40 deletion
tests/python_client/common/common_func.py
浏览文件 @ bafcbf5a
......@@ -864,7 +864,7 @@ def gen_search_param(index_type, metric_type="L2"):
ivf_search_params = {"metric_type": metric_type, "params": {}}
search_params.append(ivf_search_params)
else:
for nprobe in [64,]:
for nprobe in [64]:
ivf_search_params = {"metric_type": metric_type, "params": {"nprobe": nprobe}}
search_params.append(ivf_search_params)
elif index_type in ["BIN_FLAT", "BIN_IVF_FLAT"]:
......@@ -876,7 +876,7 @@ def gen_search_param(index_type, metric_type="L2"):
binary_search_params = {"metric_type": metric_type, "params": {"nprobe": nprobe}}
search_params.append(binary_search_params)
elif index_type in ["HNSW"]:
for ef in [64, 32768]:
for ef in [64, 1500, 32768]:
hnsw_search_param = {"metric_type": metric_type, "params": {"ef": ef}}
search_params.append(hnsw_search_param)
elif index_type == "ANNOY":
......@@ -884,12 +884,14 @@ def gen_search_param(index_type, metric_type="L2"):
annoy_search_param = {"metric_type": metric_type, "params": {"search_k": search_k}}
search_params.append(annoy_search_param)
elif index_type == "DISKANN":
for search_list in [20, 30]:
for search_list in [20, 300, 1500]:
diskann_search_param = {"metric_type": metric_type, "params": {"search_list": search_list}}
search_params.append(diskann_search_param)
else:
log.error("Invalid index_type.")
raise Exception("Invalid index_type.")
log.debug(search_params)
return search_params
......
tests/python_client/common/common_type.py
浏览文件 @ bafcbf5a
......@@ -11,6 +11,7 @@ default_nb_medium = 5000
default_top_k = 10
default_nq = 2
default_limit = 10
max_limit = 16384
default_search_params = {"metric_type": "COSINE", "params": {"nprobe": 10}}
default_search_ip_params = {"metric_type": "IP", "params": {"nprobe": 10}}
default_search_binary_params = {"metric_type": "JACCARD", "params": {"nprobe": 10}}
......
tests/python_client/testcases/test_search.py
浏览文件 @ bafcbf5a
......@@ -31,6 +31,7 @@ default_nb_medium = ct.default_nb_medium
default_nq = ct.default_nq
default_dim = ct.default_dim
default_limit = ct.default_limit
max_limit = ct.max_limit
default_search_exp = "int64 >= 0"
default_search_string_exp = "varchar >= \"0\""
default_search_mix_exp = "int64 >= 0 && varchar >= \"0\""
......@@ -3258,7 +3259,8 @@ class TestCollectionSearch(TestcaseBase):
zip(ct.all_index_types[:6],
ct.default_index_params[:6]))
@pytest.mark.parametrize("metrics", ct.float_metrics)
def test_search_output_field_vector_after_different_index_metrics(self, index, params, metrics):
@pytest.mark.parametrize("limit", [20, 1200])
def test_search_output_field_vector_after_different_index_metrics(self, index, params, metrics, limit):
"""
target: test search with output vector field after different index
method: 1. create a collection and insert data
......@@ -3282,15 +3284,23 @@ class TestCollectionSearch(TestcaseBase):
collection_w.load()
# 3. search with output field vector
search_params = cf.gen_search_param(index, metrics)[0]
collection_w.search(vectors[:1], default_search_field,
search_params, default_limit, default_search_exp,
output_fields=[field_name],
check_task=CheckTasks.check_search_results,
check_items={"nq": 1,
"limit": default_limit,
"original_entities": _vectors,
"output_fields": [field_name]})
search_params = cf.gen_search_param(index, metrics)
for search_param in search_params:
log.info(search_param)
if index == "HNSW":
limit = search_param["params"]["ef"]
if limit > max_limit:
limit = default_nb
if index == "DISKANN":
limit = search_param["params"]["search_list"]
collection_w.search(vectors[:1], default_search_field,
search_param, limit, default_search_exp,
output_fields=[field_name],
check_task=CheckTasks.check_search_results,
check_items={"nq": 1,
"limit": limit,
"original_entities": _vectors,
"output_fields": [field_name]})
@pytest.mark.tags(CaseLabel.L2)
@pytest.mark.parametrize("index", ["HNSW", "BIN_FLAT", "BIN_IVF_FLAT"])
......@@ -5943,7 +5953,7 @@ class TestCollectionRangeSearch(TestcaseBase):
vectors = np.array(_vectors[0]).tolist()
vectors = [vectors[i][-1] for i in range(nq)]
# 3. range search
range_search_params = {"metric_type": "COSINE", "params": {"nprobe": 10, "radius": radius,
range_search_params = {"metric_type": "COSINE", "params": {"radius": radius,
"range_filter": range_filter}}
search_res = collection_w.search(vectors[:nq], default_search_field,
range_search_params, default_limit,
......@@ -5972,7 +5982,7 @@ class TestCollectionRangeSearch(TestcaseBase):
vectors = np.array(_vectors[0]).tolist()
vectors = [vectors[i][-1] for i in range(default_nq)]
# 3. range search with L2
range_search_params = {"metric_type": "COSINE", "params": {"nprobe": 10, "range_filter": 1}}
range_search_params = {"metric_type": "COSINE", "params": {"range_filter": 1}}
collection_w.search(vectors[:default_nq], default_search_field,
range_search_params, default_limit,
default_search_exp,
......@@ -5981,7 +5991,7 @@ class TestCollectionRangeSearch(TestcaseBase):
"ids": insert_ids,
"limit": default_limit})
# 4. range search with IP
range_search_params = {"metric_type": "IP", "params": {"nprobe": 10, "range_filter": 1}}
range_search_params = {"metric_type": "IP", "params": {"range_filter": 1}}
collection_w.search(vectors[:default_nq], default_search_field,
range_search_params, default_limit,
default_search_exp,
......@@ -6004,7 +6014,7 @@ class TestCollectionRangeSearch(TestcaseBase):
vectors = np.array(_vectors[0]).tolist()
vectors = [vectors[i][-1] for i in range(default_nq)]
# 3. range search with L2
range_search_params = {"metric_type": "L2", "params": {"nprobe": 10, "radius": 0}}
range_search_params = {"metric_type": "L2", "params": {"radius": 0}}
collection_w.search(vectors[:default_nq], default_search_field,
range_search_params, default_limit,
default_search_exp,
......@@ -6013,7 +6023,7 @@ class TestCollectionRangeSearch(TestcaseBase):
"ids": [],
"limit": 0})
# 4. range search with IP
range_search_params = {"metric_type": "IP", "params": {"nprobe": 10, "radius": 0}}
range_search_params = {"metric_type": "IP", "params": {"radius": 0}}
collection_w.search(vectors[:default_nq], default_search_field,
range_search_params, default_limit,
default_search_exp,
......@@ -6034,7 +6044,7 @@ class TestCollectionRangeSearch(TestcaseBase):
vectors = np.array(_vectors[0]).tolist()
vectors = [vectors[i][-1] for i in range(default_nq)]
# 3. range search with L2
range_search_params = {"metric_type": "COSINE", "params": {"nprobe": 10}, "radius": 0, "range_filter": 1}
range_search_params = {"metric_type": "COSINE", "radius": 0, "range_filter": 1}
collection_w.search(vectors[:default_nq], default_search_field,
range_search_params, default_limit,
default_search_exp,
......@@ -6043,7 +6053,7 @@ class TestCollectionRangeSearch(TestcaseBase):
"ids": insert_ids,
"limit": default_limit})
# 4. range search with IP
range_search_params = {"metric_type": "IP", "params": {"nprobe": 10}, "radius": 1, "range_filter": 0}
range_search_params = {"metric_type": "IP", "radius": 1, "range_filter": 0}
collection_w.search(vectors[:default_nq], default_search_field,
range_search_params, default_limit,
default_search_exp,
......@@ -6297,11 +6307,11 @@ class TestCollectionRangeSearch(TestcaseBase):
enable_dynamic_field)[0:5]
# 2. search for original data after load
vectors = [[random.random() for _ in range(dim)] for _ in range(nq)]
range_search_params = {"metric_type": "COSINE", "params": {"nprobe": 10, "radius": 0,
range_search_params = {"metric_type": "COSINE", "params": {"radius": 0,
"range_filter": 1000}}
log.info("test_range_search_new_data: searching for original data after load")
collection_w.search(vectors[:nq], default_search_field,
default_search_params, limit,
range_search_params, limit,
default_search_exp,
check_task=CheckTasks.check_search_results,
check_items={"nq": nq,
......@@ -6316,8 +6326,6 @@ class TestCollectionRangeSearch(TestcaseBase):
# 4. search for new data without load
# Using bounded staleness, maybe we could not search the "inserted" entities,
# since the search requests arrived query nodes earlier than query nodes consume the insert requests.
range_search_params = {"metric_type": "COSINE", "params": {"nprobe": 10, "radius": 0,
"range_filter": 1000}}
collection_w.search(vectors[:nq], default_search_field,
range_search_params, limit,
default_search_exp,
......@@ -6351,7 +6359,7 @@ class TestCollectionRangeSearch(TestcaseBase):
# 3. load and range search
collection_w.load()
vectors = [[random.random() for _ in range(dim)] for _ in range(default_nq)]
range_search_params = {"metric_type": "L2", "params": {"nprobe": 10, "radius": 1000,
range_search_params = {"metric_type": "L2", "params": {"radius": 1000,
"range_filter": 0}}
collection_w.search(vectors[:default_nq], default_search_field,
range_search_params, default_limit,
......@@ -6428,6 +6436,8 @@ class TestCollectionRangeSearch(TestcaseBase):
for search_param in search_params:
search_param["params"]["radius"] = 1000
search_param["params"]["range_filter"] = 0
if index.startswith("IVF_"):
search_param["params"].pop("nprobe")
log.info("Searching with search params: {}".format(search_param))
collection_w.search(vectors[:default_nq], default_search_field,
search_param, default_limit,
......@@ -6469,6 +6479,8 @@ class TestCollectionRangeSearch(TestcaseBase):
for search_param in search_params:
search_param["params"]["radius"] = 0
search_param["params"]["range_filter"] = 1000
if index.startswith("IVF_"):
search_param["params"].pop("nprobe")
log.info("Searching with search params: {}".format(search_param))
collection_w.search(vectors[:default_nq], default_search_field,
search_param, default_limit,
......@@ -6503,8 +6515,7 @@ class TestCollectionRangeSearch(TestcaseBase):
limit_check = par[1].num_entities
else:
limit_check = limit
range_search_params = {"metric_type": "L2", "params": {"nprobe": 128, "radius": 1000,
"range_filter": 0}}
range_search_params = {"metric_type": "L2", "params": {"radius": 1000, "range_filter": 0}}
collection_w.search(vectors[:default_nq], default_search_field,
range_search_params, limit, default_search_exp,
[par[1].name], _async=_async,
......@@ -6539,8 +6550,7 @@ class TestCollectionRangeSearch(TestcaseBase):
distance_0 = cf.jaccard(query_raw_vector[0], binary_raw_vector[0])
distance_1 = cf.jaccard(query_raw_vector[0], binary_raw_vector[1])
# 4. search and compare the distance
search_params = {"metric_type": "JACCARD", "params": {"nprobe": 10, "radius": 1000,
"range_filter": 0}}
search_params = {"metric_type": "JACCARD", "params": {"radius": 1000, "range_filter": 0}}
res = collection_w.search(binary_vectors[:nq], "binary_vector",
search_params, default_limit, "int64 >= 0",
_async=_async,
......@@ -6574,8 +6584,7 @@ class TestCollectionRangeSearch(TestcaseBase):
# 3. compute the distance
query_raw_vector, binary_vectors = cf.gen_binary_vectors(3000, default_dim)
# 4. range search
search_params = {"metric_type": "JACCARD", "params": {"nprobe": 10, "radius": -1,
"range_filter": -10}}
search_params = {"metric_type": "JACCARD", "params": {"radius": -1, "range_filter": -10}}
collection_w.search(binary_vectors[:default_nq], "binary_vector",
search_params, default_limit,
check_task=CheckTasks.check_search_results,
......@@ -6616,8 +6625,7 @@ class TestCollectionRangeSearch(TestcaseBase):
distance_0 = cf.hamming(query_raw_vector[0], binary_raw_vector[0])
distance_1 = cf.hamming(query_raw_vector[0], binary_raw_vector[1])
# 4. search and compare the distance
search_params = {"metric_type": "HAMMING", "params": {"nprobe": 10, "radius": 1000,
"range_filter": 0}}
search_params = {"metric_type": "HAMMING", "params": {"radius": 1000, "range_filter": 0}}
res = collection_w.search(binary_vectors[:nq], "binary_vector",
search_params, default_limit, "int64 >= 0",
_async=_async,
......@@ -6701,7 +6709,7 @@ class TestCollectionRangeSearch(TestcaseBase):
if radius > distance_single >= range_filter:
limit += 1
# 5. range search and compare the distance
search_params = {"metric_type": "TANIMOTO", "params": {"nprobe": 10, "radius": radius,
search_params = {"metric_type": "TANIMOTO", "params": {"radius": radius,
"range_filter": range_filter}}
res = collection_w.search(binary_vectors[:1], "binary_vector",
search_params, default_limit, "int64 >= 0",
......@@ -6737,7 +6745,7 @@ class TestCollectionRangeSearch(TestcaseBase):
# 3. compute the distance
query_raw_vector, binary_vectors = cf.gen_binary_vectors(3000, default_dim)
# 4. range search
search_params = {"metric_type": "JACCARD", "params": {"nprobe": 10, "radius": -1, "range_filter": -10}}
search_params = {"metric_type": "JACCARD", "params": {"radius": -1, "range_filter": -10}}
collection_w.search(binary_vectors[:default_nq], "binary_vector",
search_params, default_limit,
check_task=CheckTasks.check_search_results,
......@@ -6763,7 +6771,7 @@ class TestCollectionRangeSearch(TestcaseBase):
# 4. search
log.info("test_range_search_binary_without_flush: searching collection %s" % collection_w.name)
binary_vectors = cf.gen_binary_vectors(default_nq, default_dim)[1]
search_params = {"metric_type": metrics, "params": {"nprobe": 10, "radius": 1000,
search_params = {"metric_type": metrics, "params": {"radius": 1000,
"range_filter": 0}}
collection_w.search(binary_vectors[:default_nq], "binary_vector",
search_params, default_limit,
......@@ -6811,7 +6819,7 @@ class TestCollectionRangeSearch(TestcaseBase):
# 3. search with expression
log.info("test_range_search_with_expression: searching with expression: %s" % expression)
vectors = [[random.random() for _ in range(dim)] for _ in range(default_nq)]
range_search_params = {"metric_type": "L2", "params": {"nprobe": 10, "radius": 1000,
range_search_params = {"metric_type": "L2", "params": {"radius": 1000,
"range_filter": 0}}
search_res, _ = collection_w.search(vectors[:default_nq], default_search_field,
range_search_params, nb, expression,
......@@ -6844,7 +6852,7 @@ class TestCollectionRangeSearch(TestcaseBase):
enable_dynamic_field)[0:4]
# 2. search
log.info("test_range_search_with_output_field: Searching collection %s" % collection_w.name)
range_search_params = {"metric_type": "COSINE", "params": {"nprobe": 10, "radius": 0,
range_search_params = {"metric_type": "COSINE", "params": {"radius": 0,
"range_filter": 1000}}
res = collection_w.search(vectors[:default_nq], default_search_field,
range_search_params, default_limit,
......@@ -6877,7 +6885,7 @@ class TestCollectionRangeSearch(TestcaseBase):
def search(collection_w):
vectors = [[random.random() for _ in range(dim)] for _ in range(nq)]
range_search_params = {"metric_type": "COSINE", "params": {"nprobe": 10, "radius": 0,
range_search_params = {"metric_type": "COSINE", "params": {"radius": 0,
"range_filter": 1000}}
collection_w.search(vectors[:nq], default_search_field,
range_search_params, default_limit,
......@@ -6955,7 +6963,7 @@ class TestCollectionRangeSearch(TestcaseBase):
nums = 5000
vectors = [[random.random() for _ in range(dim)] for _ in range(nums)]
range_search_params = {"metric_type": "L2", "params": {"nprobe": 10, "radius": 1000,
range_search_params = {"metric_type": "L2", "params": {"radius": 1000,
"range_filter": 0}}
search_res, _ = collection_w.search(vectors, default_search_field,
range_search_params, default_limit, expression,
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或