Why is the search result always 0 when calculating binary vectors using Hamming distance？

[yesletgo]

[xiaofan-luan]

@yesletgo
did you check IDs?
maybe they are all [1,1,1,1....] so distance is all 0

[yesletgo]

Isn't the top k sorted by hummin distance? There are many in the library that are not [1,1,1...], so there should be some with scores other than 0

[yhmo]

Hamming distance:
0000 and 1111 is 4
1111 and 1111 is 0

Hamming distance = 0 means the two vectors are the same.

There are 500000000 vectors in the collection. Do you know how many [1,1,1...] in the collection? I guess there are millions of [1,1,1...] in the collection.
The search result is sorted by hamming distance, from zero to non-zero. So, it returns the top 100 items with distance =0. The top 100 vectors are all [1,1,1...].
If you set topk=1000000, maybe there are some distances not 0, but it is impossible because milvus only accepts the max topk 16384.

[yesletgo]

Thank you. I mistakenly thought it was sorted in descending order based on the Hamming distance value, but currently it seems to be sorted in ascending order.