PMFDistanceMetric
Enums · AiDotNet.MetaLearning.Options
The metric PMF’s ProtoNet classifier scores a query against the class prototypes with.
For Beginners
Each class is summarized by the mean feature vector of its support examples (its prototype). The metric decides how a query is compared with each prototype: by the angle between them (cosine) or by their straight-line distance (squared Euclidean).
Fields
| Field | Summary |
|---|---|
PMFDistanceMetric Cosine |
Cosine similarity scaled by a learnable inverse temperature, cos(f(x), c_k) / tau: the paper’s choice (Hu et al. |
PMFDistanceMetric SquaredEuclidean |
Negative squared Euclidean distance, -||f(x) - c_k||^2, as in the original ProtoNet (Snell et al. |
int value__ |