[test]Add recall test (#18468)

Signed-off-by: zhuwenxing <wenxing.zhu@zilliz.com>
pull/18483/head
zhuwenxing 2022-08-01 16:20:33 +08:00 committed by GitHub
parent 8c030e3a32
commit 0420a8beb0
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
5 changed files with 216 additions and 0 deletions

View File

@ -6,6 +6,7 @@
**/docker-compose.yml
.idea
*.html
*.hdf5
.python-version
__pycache__

View File

@ -0,0 +1,18 @@
#!/bin/bash
# refer to https://github.com/yahoojapan/gongt/blob/master/assets/bench/download.sh
function check () {
if [ ! -e $1 ]; then
curl -LO $2
fi
# md5sum -c $1.md5
}
check fashion-mnist-784-euclidean.hdf5 http://vectors.erikbern.com/fashion-mnist-784-euclidean.hdf5
check glove-25-angular.hdf5 http://vectors.erikbern.com/glove-25-angular.hdf5
check glove-50-angular.hdf5 http://vectors.erikbern.com/glove-50-angular.hdf5
check glove-100-angular.hdf5 http://vectors.erikbern.com/glove-100-angular.hdf5
check glove-200-angular.hdf5 http://vectors.erikbern.com/glove-200-angular.hdf5
check mnist-784-euclidean.hdf5 http://vectors.erikbern.com/mnist-784-euclidean.hdf5
check nytimes-256-angular.hdf5 http://vectors.erikbern.com/nytimes-256-angular.hdf5
check sift-128-euclidean.hdf5 http://vectors.erikbern.com/sift-128-euclidean.hdf5

View File

@ -0,0 +1,130 @@
import h5py
import numpy as np
import time
from pathlib import Path
from pymilvus import (
connections,
FieldSchema, CollectionSchema, DataType,
Collection
)
def read_benchmark_hdf5(file_path):
f = h5py.File(file_path, 'r')
train = np.array(f["train"])
test = np.array(f["test"])
neighbors = np.array(f["neighbors"])
f.close()
return train, test, neighbors
dim = 128
TIMEOUT = 100
def milvus_recall_test(host='127.0.0.1'):
file_path = f"{str(Path(__file__).absolute().parent.parent.parent)}/assets/ann_hdf5/sift-128-euclidean.hdf5"
train, test, neighbors = read_benchmark_hdf5(file_path)
connections.connect(host=host, port="19530")
default_fields = [
FieldSchema(name="int64", dtype=DataType.INT64, is_primary=True),
FieldSchema(name="float", dtype=DataType.FLOAT),
FieldSchema(name="varchar", dtype=DataType.VARCHAR, max_length=65535),
FieldSchema(name="float_vector", dtype=DataType.FLOAT_VECTOR, dim=dim)
]
default_schema = CollectionSchema(
fields=default_fields, description="test collection")
collection = Collection(name="sift_128_euclidean", schema=default_schema)
nb = len(train)
batch_size = 50000
epoch = int(nb / batch_size)
t0 = time.time()
for i in range(epoch):
print("epoch:", i)
start = i * batch_size
end = (i + 1) * batch_size
if end > nb:
end = nb
data = [
[i for i in range(start, end)],
[np.float32(i) for i in range(start, end)],
[str(i) for i in range(start, end)],
train[start:end]
]
collection.insert(data)
t1 = time.time()
print(f"\nInsert {nb} vectors cost {t1 - t0:.4f} seconds")
t0 = time.time()
print(f"\nGet collection entities...")
print(collection.num_entities)
t1 = time.time()
print(f"\nGet collection entities cost {t1 - t0:.4f} seconds")
# create index
default_index = {"index_type": "IVF_SQ8",
"metric_type": "L2", "params": {"nlist": 64}}
print(f"\nCreate index...")
t0 = time.time()
collection.create_index(field_name="float_vector",
index_params=default_index)
t1 = time.time()
print(f"\nCreate index cost {t1 - t0:.4f} seconds")
# load collection
replica_number = 1
print(f"\nload collection...")
t0 = time.time()
collection.load(replica_number=replica_number)
t1 = time.time()
print(f"\nload collection cost {t1 - t0:.4f} seconds")
# search
topK = 100
nq = 10000
search_params = {"metric_type": "L2", "params": {"nprobe": 10}}
t0 = time.time()
print(f"\nSearch...")
# define output_fields of search result
res = collection.search(
test[:nq], "float_vector", search_params, topK, output_fields=["int64"], timeout=TIMEOUT
)
t1 = time.time()
print(f"search cost {t1 - t0:.4f} seconds")
result_ids = []
for hits in res:
result_id = []
for hit in hits:
result_id.append(hit.entity.get("int64"))
result_ids.append(result_id)
# calculate recall
true_ids = neighbors[:nq, :topK]
sum_radio = 0.0
for index, item in enumerate(result_ids):
# tmp = set(item).intersection(set(flat_id_list[index]))
assert len(item) == len(true_ids[index])
tmp = set(true_ids[index]).intersection(set(item))
sum_radio = sum_radio + len(tmp) / len(item)
recall = round(sum_radio / len(result_ids), 3)
assert recall >= 0.95
print(f"recall={recall}")
# query
expr = "int64 in [2,4,6,8]"
output_fields = ["int64", "float"]
res = collection.query(expr, output_fields, timeout=TIMEOUT)
sorted_res = sorted(res, key=lambda k: k['int64'])
for r in sorted_res:
print(r)
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser(description='config for recall test')
parser.add_argument('--host', type=str,
default="127.0.0.1", help='milvus server ip')
args = parser.parse_args()
host = args.host
milvus_recall_test(host)

View File

@ -0,0 +1,64 @@
import h5py
import numpy as np
import time
from pathlib import Path
from pymilvus import connections, Collection
def read_benchmark_hdf5(file_path):
f = h5py.File(file_path, 'r')
train = np.array(f["train"])
test = np.array(f["test"])
neighbors = np.array(f["neighbors"])
f.close()
return train, test, neighbors
dim = 128
TIMEOUT = 100
def search_test(host="127.0.0.1"):
file_path = f"{str(Path(__file__).absolute().parent.parent.parent)}/assets/ann_hdf5/sift-128-euclidean.hdf5"
train, test, neighbors = read_benchmark_hdf5(file_path)
connections.connect(host=host, port="19530")
collection = Collection(name="sift_128_euclidean")
nq = 10000
topK = 100
search_params = {"metric_type": "L2", "params": {"nprobe": 10}}
t0 = time.time()
print(f"\nSearch...")
# define output_fields of search result
res = collection.search(
test[:nq], "float_vector", search_params, topK, output_fields=["int64"], timeout=TIMEOUT
)
t1 = time.time()
print(f"search cost {t1 - t0:.4f} seconds")
result_ids = []
for hits in res:
result_id = []
for hit in hits:
result_id.append(hit.entity.get("int64"))
result_ids.append(result_id)
# calculate recall
true_ids = neighbors[:nq,:topK]
sum_radio = 0.0
for index, item in enumerate(result_ids):
# tmp = set(item).intersection(set(flat_id_list[index]))
assert len(item) == len(true_ids[index])
tmp = set(true_ids[index]).intersection(set(item))
sum_radio = sum_radio + len(tmp) / len(item)
recall = round(sum_radio / len(result_ids), 3)
assert recall >= 0.95
print(f"recall={recall}")
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser(description='config for recall test')
parser.add_argument('--host', type=str, default="127.0.0.1", help='milvus server ip')
args = parser.parse_args()
host = args.host
search_test(host)

View File

@ -33,3 +33,6 @@ protobuf==3.17.1
# for bulk load test
minio==7.1.5
# for benchmark
h5py==3.1.0