Performance Constraints
Integrating performance baselines and performance expectations in the hyperparameter optimization process is furthermore helpful to increase the overall speed and efficiency. Further testing of a specific hyperparameter configuration in further inner-cross-validation folds can be skipped if the given configuration performs worse than a given static or dynamic threshold.
There are three types of contraints implemented in PHOTONAI:
- MinimumPerformanceConstraint: the lower bound is the given threshold (e.g. accuracy of at least 0.8)
- BestPerformanceConstraint: the lower bound (+- margin) is the so far best metric value
- DummyPerformanceConstraint: the lower bound (+-margin) is the dummy performance of the specific metric
The threshold is applied in three strategies:
- any: Computation is skipped if any of the folds is worse than the threshold
- first: Computation is skipped if the first fold performs worse than the threshold
- mean: Computation is skipped if the mean of all folds computed so far is worse than the threshold
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 |
|