How to get (reconstruct) a set of vectors that was associated with a single id?

[volkov-maxim]

Hi!
I had associate some sets of vectors with ids, but can't find approach to get back this sets from index.
In case with 2 ids I had got only 2 vectors.

Code for reproduction:

import faiss
import numpy as np

d = 16
k = 2
quantiser = faiss.IndexFlatL2(d)
index = faiss.IndexIVFFlat(quantiser, d, k)
ids = range(1, 3)

index.train(faiss.randn((10, d), 123))
print(index.is_trained)

# Create 1st index.
index.add(faiss.randn((10, d), 345))
index.add(faiss.randn((10, d), 567))
print(index.ntotal)

ref_recons1 = index.reconstruct_n(0, 20)

# Create 2nd index.
index.reset()
index.add_with_ids(faiss.randn((10, d), 345), np.repeat(ids[0], 10))
index.add_with_ids(faiss.randn((10, d), 567), np.repeat(ids[1], 10))

index.set_direct_map_type(faiss.DirectMap.Hashtable)
ref_recons2 = index.reconstruct_n(0, 20)

# Compare items.
for i in range(index.ntotal):
    for j in range(index.ntotal):
        if np.all(ref_recons1[i] == ref_recons2[j]):
            print("i=", i, ", j=", j)

Output:

True
20
i= 9 , j= 1
i= 17 , j= 2

I'm expect 2 sets of vectors:

• 10 vectors associated with id=1
• 10 vectors associated with id=2

But, only 2 items are equal and I dont't understand why.
Please help me to implement reconstruction sets of vectors that associated with ids.

[satymish]

Hi @volkov-maxim when I run the code you wrote, I get the following output

True
20
i= 0 , j= 0
i= 3 , j= 3
i= 4 , j= 4
i= 5 , j= 5
i= 6 , j= 6
i= 7 , j= 7
i= 8 , j= 8
i= 9 , j= 1
i= 9 , j= 9
i= 10 , j= 10
i= 11 , j= 11
i= 12 , j= 12
i= 13 , j= 13
i= 14 , j= 14
i= 15 , j= 15
i= 16 , j= 16
i= 17 , j= 2
i= 17 , j= 17
i= 18 , j= 18
i= 19 , j= 19