• Math/algorithmic questions
• How can I get constructive criticism about my data?
• Why do I get weird results with brute force search on vectors with large components?
• Analyzing accuracy issues with IndexIVFPQ
• Why is multi-level IVF search not supported in Faiss?
• How can I reproduce results from the original PQ paper?
• How can I index 2D or 3D data?
• Questions on the API
• Can searches and additions be performed concurrently from several threads?
• Why are searches with a single query vector slow?
• In C++, do I need to keep a reference to the coarse quantizer around for an IndexIVFFlat/IndexIVFPQ/IndexIVFPQR index?
• How can I distribute index building on several machines?
• Does Faiss support string ids?
• Searching duplicate vectors is slow
• Reproducibility
• Questions about training
• Can I ignore "WARNING clustering XXX points to YYY centroids"?
• How many training points do I need for k-means?
• How can I do unbiased sampling from a large set of vectors?
• Can I use k-means with an inner-product index?
• Can I do k-means with something else than L2 assignment?
• How can I warm-start the centroid training for an IVF index?
• Can I used an index trained on some kind of data to index some other type of data of the same dimension?
• Is re-training an index supported?
• Extracting information from a C++ Index object
• Other questions
• Why do I get "wrong number or type of arguments" errors in Python?
• How can diversity be enforced on search results?
• Is it possible to dynamically exclude vectors based on some criterion?
• What does it mean when a search returns -1 ids?
• Is PQ encoding with more or less than 8 bits per components supported?
• How can I set nprobe on the sub-indexes of an IndexShards or IndexReplicas?
• How can I set nprobe on an "opaque" index?
• Why does the RAM usage not go down when I delete an index?
• Why are the number of openmp threads not being set correctly when calling faiss.omp_set_num_threads(n)?