Input training set too big (max size is 65536)

[zhouxiaoxu]

I am installing FAISS onto Red Hat Enterprise Linux Server release 6.3. I installed openblas and ran make and tests/demo_ivfpq_indexing and got the following info:
Input training set too big (max size is 65536), sampling 65536 / 100000 vectors
Is that mean I can't use more than 65536 vectors ？
My os is 64 bit operating system,when I use "uname -a" command, it's show below:
Linux snisspreweb02 2.6.32-279.el6.x86_64 #1 SMP Wed Jun 13 18:24:36 EDT 2012 x86_64 x86_64 x86_64 GNU/Linux

[zhouxiaoxu]

more information: I using faiss with cpu

[mdouze]

Hi

This just means that the system found that 65536 vectors are enough. If you want to use more vectors, please set max_points_per_centroid in the clustering object to some large value like here

https://github.com/facebookresearch/faiss/blob/master/benchs/kmeans_mnist.py#L54