Input and Strategy Combinations
Selection input and selection strategies can be combined according to the following tables.
Depending on the input, we create numerical and categorical vector or scalar representations which can be controlled by a strategy. As such, not every combination is possible.
Input + Strategy
| Input | Representation | Strategy |
|---|---|---|
| Embeddings | vector (numeric) | Diversity , Similarity, Typicality |
| Metadata | scalar | Weights, Threshold, Balance |
| Prediction + Category Count | scalar (numeric) | Weights, Threshold |
| Prediction + Class Distribution | vector (distribution / categorical) | Balance |
| Random | scalar (numeric) | Weights, Threshold |
| Scores | scalar (numeric) | Weights, Threshold |
Strategy + Input
| Strategy | Representation | Input |
|---|---|---|
| Balance | scalar (categorical) or vector (distribution / categorical) | Metadata, Prediction + Class Distribution |
| Diversity | vector (numeric) | Embeddings |
| Similarity | vector (numeric) | Embeddings |
| Threshold | scalar (numeric) | Scores, Prediction + Category Count, Random, Metadata |
| Weights | scalar (numeric) | Scores, Prediction + Category Count, Random, Metadata |
| Typicality | vector(numeric) | Embeddings |
Updated 4 months ago