Faiss: AssertionError: assert x.flags.contiguous

[gr8Adakron]

Sample data inside train feature:
[0.42307538 2.3160305 0.24694297 ... 1.0617049 0.09734876 1.0471938 ]

my train feature sample size:
3122708 X 2048

Error in part:

import faiss                                   # make faiss available
index_sim = faiss.IndexFlatL2(2048)            # build the index
print(index_sim.is_trained)
index_sim.add(tr_features_np)                  # add vectors to the index
print(index_sim.ntotal)

---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
<ipython-input-7-418d31a6035c> in <module>()
      2 index_sim = faiss.IndexFlatL2(2048)   # build the index
      3 print(index_sim.is_trained)
----> 4 index_sim.add(tr_features_np)                  # add vectors to the index
      5 print(index_sim.ntotal)

~/anaconda3/lib/python3.6/site-packages/faiss/__init__.py in replacement_add(self, x)
     95 
     96     def replacement_add(self, x):
---> 97         assert x.flags.contiguous
     98         n, d = x.shape
     99         assert d == self.d

Any Help!

[Enet4]

Adding a numpy array of vectors to the index only works with contiguous arrays. The one that you have might have been manipulated in some way as to no longer be C-contiguous.

You can try np.ascontiguousarray:

index_sim.add(np.ascontiguousarray(tr_features_np))

---

I also noticed that maybe those asserts in faiss.py should be removed, because swig_ptr already checks for C-contiguous arrays and yields a better error message when this happens: ValueError: array is not C-contiguous.

[gr8Adakron]

Thanks, it worked!

[gr8Adakron]

One more doubt please, can you please tell me how to _save_ this variable _index_sim_ in some file, after adding the train feature.

Because if everytime I am running _.add_ function it take like 5 to 10 minutes. so I just want to save this variable somewhere and load it later when I want to use it.

The data_type of variable _index_sim_ is: _faiss.swigfaiss.IndexFlatL2_

Thanx!

[Enet4]

Admittedly, it was a bit hard for me to find the subject on the wiki pages, but here it is. You are looking for faiss.write_index and faiss.read_index:

faiss.write_index(index_sim, "sim.index") # save
index_sim = faiss.read_index("sim.index") # load

[gr8Adakron]

Thank you! so much. Have a great day.

[mdouze]

@Enet4: changed the chapter name in the wiki to make it easier to find.

[Enet4]

@mdouze Yes, it seems to be much better now. :+1: