Loading Datasets
Datasets need to be loaded onto the APU.
Loaded datasets will be displayed in the Loaded datasets section at the top of the Datasets window.
You can load more than one dataset to the APU, but at least one must be loaded to enable selecting a query.
Dataset Load
To load an available dataset to the APU
In the Datasets window, click on the + icon on the left to expand the dataset that you wish to load.
While it’s recommended to use the dataset’s default settings, you’ll can modify these settings for the current search before loading the dataset.
Configuration settings vary according to the type of dataset.
Flat uint8 datasets
| Parameter | Description |
|---|---|
Typical NQueries | Specify the number of queries expected to be used in the search. · The default value is 10. |
Max NQueries | Specify the maximum number of queries that the search can run. · The default value is 50 queries. |
Algorithm | Specify which algorithm to use for the binary search: Hamming or Tanimoto. |
Flat float32 datasets
| Parameter | Description |
|---|---|
Typical NQueries | Specify the number of queries expected to be used in the search. · The default value is 10. |
Max NQueries | Specify the maximum number of queries that the search can run. · The default value is 50 queries. |
N bits | Read-only value indicating the number of nBits in the binary dataset vectors. · Displayed only if you trained multiple times on the same dataset using different nBit values. |
Hamming-k | Specify the required number of matches to find when running the initial hamming-k search. · If you run the search accuracy during the training , you will see the Hamming k results of it in this field. |
Normalize | Specify whether the vector types are normalized. · Check to indicate normalized vectors. · Clear to if the dataset vectors are not normalized. |
IVF float32 datasets
| Parameter | Description |
|---|---|
Typical NQueries | Specify the number of queries expected to be used in the search. · The default value is 10. |
Max NQueries | Specify the maximum number of queries that the search can run. · The default value is 50 queries. |
Number of clusters | Read-only field indicating the number of centroids. If you’ve trained the dataset using multiple cluster values, you can select one of the centroids values. |
Centroids hamming-k | Specify how many matches to find when running the initial centroids hamming-k search. · The greater the value, the more accurate the search but the search time will be increase. |
Centroids rerank | Search preparation process to improve search accuracy by narrowing down the centroids KNN. |
Hamming K | Specify the required number of matches to find when running the initial records with a hamming-k search. |
Top-K | Specify how many matches to find in the record rerank search. If no value is specified, then the search results will be the Hamming K number of records. |
Normalize | Specify whether the vector types are normalized. · Check to indicate normalized vectors. · Clear to if the dataset vectors are not normalized. |
HNSW float32 datasets
- Click the Load button.
Dataset focus
When you load multiple datasets into the search, FVS searches the dataset that is currently in focus. The focused dataset is the one that will be loaded into the APU.
By default, when loading multiple datasets to a search, the first dataset you load will have focus. If only one dataset is loaded, it is automatically selected as the focus dataset.
The focused dataset will always be highlighted in green.
To assign focus to a different dataset:
In the Loaded datasets section, select a different dataset by clicking on the radio button on the left.
