-
Notifications
You must be signed in to change notification settings - Fork 1
/
make_groundtruth.py
72 lines (48 loc) · 1.83 KB
/
make_groundtruth.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
import argparse
import logging
import time
import resource
import numpy as np
import faiss
from preprocess_datasets import DATASETS, get_dataset
def knn_ground_truth(X, k):
print("knn_ground_truth queries size %s k=%d" % (X.shape, k))
t0 = time.time()
_, d = X.shape
index = faiss.IndexFlat(d, faiss.METRIC_L2)
index.add(X)
index.train(X)
D, I = index.search(X, k)
return D, I
def usbin_write(ids, dist, fname):
ids = np.ascontiguousarray(ids, dtype="int32")
dist = np.ascontiguousarray(dist, dtype="float32")
assert ids.shape == dist.shape
f = open(fname, "wb")
n, d = dist.shape
np.array([n, d], dtype='int32').tofile(f)
ids.tofile(f)
dist.tofile(f)
def write_data(X, fname):
f = open(fname, "wb")
X = np.ascontiguousarray(X, dtype="float64")
X.tofile(f)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
def aa(*args, **kwargs):
group.add_argument(*args, **kwargs)
group = parser.add_argument_group('dataset options')
aa('--dataset', choices=DATASETS.keys(), required=True)
group = parser.add_argument_group('computation options')
# determined from ds
# aa('--range_search', action="store_true", help="do range search instead of kNN search")
aa('--k', default=100, type=int, help="number of nearest kNN neighbors to search")
args = parser.parse_args()
ds = get_dataset(args.dataset, "gaussian")
print(ds)
for query_type in ('train', 'test', 'validation'):
D, I = knn_ground_truth(np.array(ds[query_type]).astype(np.float32), k=args.k)
print(f"writing index matrix of size {I.shape}")
# write in the usbin format
usbin_write(I, D, args.dataset + f".{query_type}.knn")
write_data(np.array(ds[query_type]).astype(np.float64), args.dataset + f".{query_type}.data")