>The relationship between ef_construction and m is the reason ef_construction is required to be 2x the value of m >So, long-story short, keep the numbers relat

>The relationship between ef_construction and m is the reason ef_construction is required to be 2x the value of m >So, long-story short, keep the numbers relatively small because the quality improvement isn’t worth the performance hit. I don't know where the author got this from but this is patently false. In the tl;dr at the top of the article, he gives an efConstruction of 10, but this is quite low. In one of his citations of the paper, he omits an important part in "a reasonable quality index can be constructed for efConstruction=100" for an M=16 for 1m 128-length vectors. I've also read[1] (but not confirmed myself) that too low of an efConstruction results in abysmal performance. [1]: https://archive.ph/KQuVx#60% https://archive.ph/KQuVx#60%