[BUG] Brute force python save_load test fail randomly #704
Assignees
Labels
bug
Something isn't working
Comments
|
I found one query distances with different to samples are very close occasionally: E array([[7356, 6315, 3579, 2860, 3790, 1747, 9880, 2584, 4067, 2456],
E [7520, 6982, 5815, 929, 2325, 2250, 1439, 9866, 8289, 4179],
E [5404, 6802, 2551, 5991, 1892, 1024, 2428, 6699, 2606, 7739],
E [5774, 8981, 6457, 7510, 8410, 9036, 9010, 2026, 2463, 8314],
E [1369, 9634, 338, 8415, 9194, 3971, 1448, 9331, 1082, 1115],
E [7754, 1067, 7113, 3789, 1614, 6765, 7434, 541, 8518, 2720],
E [8104, 8804, 6654, 1776, 4759, 4700, 9157, 744, 9911, 9321],
E [8735, 3937, 5238, 5307, 9597, 4964, 8644, 3688, 3545, 8054],
E [5948, 1218, 8975, 3946, 7593, 2818, 9065, 1631, 5087, 9053],
E [6484, 8709, 7806, 9591, 6222, 3882, 3098, 5923, 1283, 2439],
E [8394, 3987, 7022, 7663, 2785, 4737, 2056, 4049, 8812, 7542],
E [4657, 7893, 332, 8275, 6701, 6561, 3746, 2839, 9894, 5603],
E [8009, 2137, 2013, 1234, 9554, 249, 4547, 9061, 8111, 6968],
E [5326, 512, 9282, 4583, 3257, 654, 8371, 7634, 2745, 3021],
E [ 543, 6044, 5574, 4648, 3479, 7515, 2366, 8803, 5344, 7147],
E [5148, 3850, 4997, 2023, 2537, 2392, 3603, 8942, 8239, 3979],
E [4257, 4958, 3282, 1352, 3671, 1301, 1036, 3618, 4553, 3117],
E [4014, 7698, 4968, 7374, 6330, 3304, 3840, 2815, 4923, 9442],
E [7005, 1510, 9989, 9899, 5163, 7446, 9105, 6494, 8157, 5861],
E [ 593, 5258, 3633, 2708, 7588, 2196, 7364, 7779, 7036, 5612],
E [2448, 1863, 100, 3025, 1347, 4543, 7865, 4251, 251, 6939],
E [ 339, 7729, 8217, 5146, 5502, 2968, 8170, 7342, 5736, 3631],
E [9443, 9989, 5761, 9708, 9475, 9838, 2443, 641, 6118, 4325],
E [6073, 1128, 2027, 1558, 8969, 59, 7697, 3103, 4433, 7554],
E [4006, 9111, 8843, 4222, 8114, 509, 402, 3178, 5061, 1806],
E [ 609, 7838, 5915, 8270, 871, 9903, 7710, 8054, 29, 2906],
E [5312, 9984, 1499, 6424, 8703, 3908, 1291, 6405, 9563, 2469],
E [8868, 6144, 8657, 6008, 2755, 2112, 6999, 2860, 5169, 4849],
E [9227, 6329, 7620, 1352, 7751, 8577, 5430, 4524, 8615, 8882],
E [6497, 6595, 3627, 7979, 1118, 3082, 3075, 1519, 7416, 9992],
E [4499, 7839, 3740, 6926, 2839, 5849, 4634, 9760, 6142, 9553],
E [6603, 2519, 4735, 2744, 6931, 9085, 4633, 8498, 2755, 4033],
E [ 175, 2110, 7, 1412, 1798, 1481, 8614, 1371, 910, 4410],
E [5100, 9721, 7463, 6862, 5599, 2337, 7774, 1352, 8642, 9823],
E [7126, 6686, 5524, 5590, 3049, 1207, 1657, 1299, 8063, 1240],
E [4630, 4522, 6362, 3991, 4703, 404, 4806, 3987, 8907, 2668],
E [1806, 3429, 5282, 2632, 9513, 6054, 1176, 2030, 8223, 4034],
E [1682, 2312, 8311, 4145, 2564, 9377, 9194, 7276, 6398, 8011],
E [3354, 2311, 3432, 3160, 8459, 2196, 2540, 7471, 9730, 346],
E [5835, 4479, 2361, 398, 8541, 2573, 4455, 9601, 6844, 2813],
E [8434, 2691, 8829, 7797, 9241, 119, 2212, 7958, 5488, 5151],
E [5737, 9222, 2020, 1136, 4479, 2518, 9206, 3687, 4255, 107],
E [6184, 6582, 3734, 3876, 9532, 730, 8907, 7171, 3682, 7437],
E [3581, 9124, 3998, 3021, 3323, 2810, 7940, 6185, 7069, 9403],
E - [8882, 9900, 8757, 107, 8479, 5171, 9484, 9775, 3933, 2111],
E ? ------
E + [8882, 9900, 8757, 107, 8479, 9484, 5171, 9775, 3933, 2111],
E ? ++++++ ++++++
[3.3649712 3.8245468 4.1913643 4.262558 4.3329716 4.3587875 4.3587875
4.4026165 4.408066 4.4113426]It can be reproduced by the code: import numpy as np
dataset_5171 = np.array( [9.32142138e-01, 8.72813582e-01, 7.71189928e-01, 5.85802495e-01,
2.38001600e-01, 7.06278622e-01, 8.78981054e-01, 4.03263688e-01,
4.75592285e-01, 1.45135760e-01, 3.21364909e-01, 6.87913895e-01,
8.30989003e-01, 2.19078675e-01, 6.42280996e-01, 4.20177460e-01,
8.86989295e-01, 3.68644297e-01, 9.93755877e-01, 6.99248761e-02,
9.09213126e-02, 2.24599376e-01, 8.28429461e-01, 3.59329879e-01,
5.71323633e-01, 6.59366027e-02, 4.18484896e-01, 8.31805766e-01,
2.35715076e-01, 8.27406108e-01, 8.21960211e-01, 4.06000704e-01,
4.96624112e-01, 4.26898181e-01, 2.14131534e-01, 6.58396363e-01,
3.63032669e-01, 9.13158238e-01, 8.36311519e-01, 2.79704154e-01,
4.96733159e-01, 9.39586386e-02, 8.37650478e-01, 7.54839361e-01,
4.97722834e-01, 5.02949297e-01, 8.90806139e-01, 5.37597716e-01,
7.28471994e-01, 7.22921133e-01], dtype=float)
dataset_9484 = np.array( [9.81457651e-01, 5.93166947e-01, 1.56116426e-01, 7.87705481e-01,
6.65047050e-01, 6.26070380e-01, 1.89543471e-01, 8.99864018e-01,
4.08117533e-01, 2.98274849e-02, 8.38999808e-01, 9.41052794e-01,
7.70968139e-01, 1.60131723e-01, 9.21209812e-01, 4.37662721e-01,
7.82714367e-01, 9.02948081e-01, 9.07859057e-02, 4.08284068e-01,
4.53266650e-01, 6.05524331e-03, 9.13958311e-01, 7.49397278e-01,
4.31984991e-01, 1.09489582e-01, 4.56626981e-01, 3.62299412e-01,
1.01631679e-01, 6.10530853e-01, 9.31025982e-01, 7.03031659e-01,
4.18992788e-01, 7.79154778e-01, 6.04481623e-03, 1.96646154e-01,
8.79877806e-01, 4.18404818e-01, 1.33909434e-01, 6.40034139e-01,
6.06465399e-01, 2.39080772e-01, 7.01063633e-01, 7.51568615e-01,
9.95701730e-01, 5.62685132e-01, 4.60158437e-01, 2.16199681e-01,
4.73403454e-01, 6.09784663e-01], dtype=float)
query = np.array([
6.98716998e-01, 7.96642601e-01, 4.80482608e-01, 4.85115111e-01,
4.14034247e-01, 6.72952533e-01, 2.37295032e-01, 7.65312135e-01,
9.61143553e-01, 4.36596006e-01, 6.77114785e-01, 4.44995224e-01,
6.83229864e-01, 3.40059429e-01, 6.11794770e-01, 6.90815866e-01,
9.56425905e-01, 8.06521237e-01, 9.50217426e-01, 9.30451825e-02,
1.24300353e-01, 6.84115410e-01, 5.58719575e-01, 9.52109814e-01,
7.08390355e-01, 2.19248887e-02, 7.92785466e-01, 8.06614876e-01,
2.92044669e-01, 8.83844793e-01, 8.60216200e-01, 9.07594740e-01,
2.09351406e-01, 9.41360295e-01, 4.94379848e-02, 4.60062832e-01,
8.38162124e-01, 5.27859628e-01, 3.16174507e-01, 2.18997210e-01,
5.47150195e-01, 1.86117634e-01, 6.49071097e-01, 6.42057657e-01,
4.42520112e-01, 7.30912447e-01, 5.05176723e-01, 2.86207590e-02,
5.84955156e-01, 3.34418684e-01], dtype=float)
def squared_euclidean_distance(vec1, vec2):
return np.sum((vec1 - vec2) ** 2)
print("Squared Euclidean Distance (5171, Query):", squared_euclidean_distance(dataset_5171, query))
print("Squared Euclidean Distance (9484, Query):", squared_euclidean_distance(dataset_9484, query))
# Squared Euclidean Distance (5171, Query): 4.358781377215945
# Squared Euclidean Distance (9484, Query): 4.358787387166914 |
|
The current conclusion is a random failure caused because the distances are too close. There's still a reminded suspect point as to why the precision looks like a downgrade. As we can see, the Python code shows the differences in the 6th decimal place. But anyway, it does not seem to be a serious issue. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
The log shows the set of neighbors is correct, but some order of them is not overturned
This could be reproduced with very high possibility by: https://github.com/rhdong/cuvs/tree/rhdong/bf-py-test-fail-reproduce
The text was updated successfully, but these errors were encountered: