Aggregate

Library of implemented Aggregate functions in AUCMEDI.

An Aggregate function can be passed to an ensemble and merges multiple class predictions into a single prediction.

Assembled predictions encoded in a NumPy matrix with shape (N_models, N_classes).
Example: [[0.5, 0.4, 0.1],
          [0.4, 0.3, 0.3],
          [0.5, 0.2, 0.3]]
-> shape (3, 3)

Merged prediction encoded in a NumPy matrix with shape (1, N_classes).
Example: [[0.4, 0.3, 0.3]]
-> shape (1, 3)

Example

# Recommended: Apply an Ensemble like Augmenting (test-time augmentation) with Majority Vote
preds = predict_augmenting(model, test_datagen, n_cycles=5, aggregate="majority_vote")

# Manual: Apply an Ensemble like Augmenting (test-time augmentation) with Majority Vote
from aucmedi.ensemble.aggregate import MajorityVote
my_agg = MajorityVote()
preds = predict_augmenting(model, test_datagen, n_cycles=5, aggregate=my_agg)

Aggregate functions are based on the abstract base class Aggregate_Base, which allows simple integration of custom aggregate methods for Ensemble.

aggregate_dict = {'mean': AveragingMean, 'median': AveragingMedian, 'majority_vote': MajorityVote, 'softmax': Softmax, 'global_argmax': GlobalArgmax} module-attribute

Dictionary of implemented Aggregate functions.