Quantization compresses a set of full-size vectors to an equally-sized set of representative quantized vectors. Each quantized vector is a fraction of the size of the original full-size vector that it represents. While quantization loses information about the original vector, the quantized form can still be used to estimate the distance between the original vector and a user-provided query vector.
The Quantizer and QuantizedVectorSet interfaces are defined, which abstract quantization so that we can swap different algorithms in/out. While a future PR will include a "real" quantization algorithm, this PR implements the "UnQuantizer", which trivially implements the Quantizer interface by storing the vectors as-is.
cockroach-teamcity
commented
11 hours ago 
Quantization compresses a set of full-size vectors to an equally-sized set of representative quantized vectors. Each quantized vector is a fraction of the size of the original full-size vector that it represents. While quantization loses information about the original vector, the quantized form can still be used to estimate the distance between the original vector and a user-provided query vector.
The Quantizer and QuantizedVectorSet interfaces are defined, which abstract quantization so that we can swap different algorithms in/out. While a future PR will include a "real" quantization algorithm, this PR implements the "UnQuantizer", which trivially implements the Quantizer interface by storing the vectors as-is.
Epic: CRDB-42943
Release note: None