luyuncheng

> Hi @luyuncheng, > > Thanks for the PR! I will take a look some time early next week. My concern about the "non-optimized" indices is that we will end...

Previously, this PR using string to parse the bit vector, this may occupy too much memory! As i checked the code: [space_bit_hamming.h](https://github.com/nmslib/nmslib/blob/master/similarity_search/include/space/space_bit_vector.h) in nmslib, i think use a int32 array...

+1, As [ml-supervised-workflow](https://www.elastic.co/guide/en/machine-learning/7.11/ml-supervised-workflow.html) shows. may be we can use some workflow in faiss training

> all segments use the same trained models > without relying on storing the raw vectors in Lucene LGTM, i am wondering the data to be trained stored in the...

> @luyuncheng I plan to simplify this, but it's not easy for the current version, sorry. That's there's no quick-and-easy fix. BTW, which space are you using? I see you...

> @luyuncheng yes, turns out it is somewhat easier for floating-point data. In principle, I can do the same for bit Hamming (and surely will do with time), but at...

have you solved this problem? I met the same problem

hi @jmazanec15 @navneet1v @heemin32 , I have an idea about this topic. in some scenarios, we want to `reduce the disk usage` and `io throughput` for the source field. so,...

> If you remove the fields from _soruce then you cannot do reindexing and update by query. For more details you can refer this doc: https://www.elastic.co/guide/en/elasticsearch/reference/7.10/mapping-source-field.html#:~:text=Think%20before%20disabling%20the%20_source%20field @navneet1v if we would...

> When I tried running the similar query some time back, I got the issue that field doesn't support Sortedbinarydocvalues. So I am wondering we might need to just ensure...