RescoringOptions interface
Contains the options for rescoring.
Properties
| default |
Default oversampling factor. Oversampling retrieves a greater set of potential documents to offset the resolution loss due to quantization. This increases the set of results that will be rescored on full-precision vectors. Minimum value is 1, meaning no oversampling (1x). This parameter can only be set when 'enableRescoring' is true. Higher values improve recall at the expense of latency. |
| enable |
If set to true, after the initial search on the compressed vectors, the similarity scores are recalculated using the full-precision vectors. This will improve recall at the expense of latency. |
| rescore |
Controls the storage method for original vectors. This setting is immutable. |
Property Details
defaultOversampling
Default oversampling factor. Oversampling retrieves a greater set of potential documents to offset the resolution loss due to quantization. This increases the set of results that will be rescored on full-precision vectors. Minimum value is 1, meaning no oversampling (1x). This parameter can only be set when 'enableRescoring' is true. Higher values improve recall at the expense of latency.
defaultOversampling?: numberProperty Value
number
enableRescoring
If set to true, after the initial search on the compressed vectors, the similarity scores are recalculated using the full-precision vectors. This will improve recall at the expense of latency.
enableRescoring?: booleanProperty Value
boolean
rescoreStorageMethod
Controls the storage method for original vectors. This setting is immutable.
rescoreStorageMethod?: stringProperty Value
string