API aggregateΒΆ
The aggregate
strategies are used to combine labels from multiple users into a single label.
This aggregated label can be either a probability distribution (soft label) or a single class label (hard label).
All strategies are located at this direction on GitHub <https://github.com/peerannot/peerannot/tree/main/peerannot/models/aggregation>.
All aggregation-based strategies are available running:
peerannot agginfo
Majority voting Most answered label per task |
|
Naive soft: Frequency distribution |
|
Worker Agreement With Aggregate |
|
Iterative Weighted Majority Vote |
|
Dawid and skene model (1979) |
|
GLAD (Whitehill et. al 2009) |
|
PlantNet aggregation strategy |
|
Two third agreement Accepts the label given with a two third consensus on at least 2 votes and returns -1 otherwise |
|
WDS: Weighted Distribution from Dawid and Skene |