Beta Version 1.8.0
Added
-
Added “QuickSetups” section.
-
Added “LinkLayers” section.
-
Added “RegularizationFunctions” section.
-
Added HingeLoss, FastHingeLoss, EpsilonInsensitiveLoss and FastEpsilonInsensitiveLoss under the “CostFunctions” section.
-
Added TemporalDifferenceActorCritic, DeepNStepQLearning, DeepNStepStateActionRewardStateAction and DeepNStepExpectedStateActionRewardStateAction under the “ReinforcementLearningModels” section.
Fixes
-
Fixed some bug where the first derivative tensor calculations are wrong for the subtraction and division operations under the “AutomaticDifferentiationTensor” section.
-
Fixed some bugs where the first derivative tensor do not collapse to the input tensors’ dimension size array under the “AutomaticDifferentiationTensor” section.
-
Fixed a bug where the minimum constructor uses math.max instead of math.min in its internal calculation under the “AutomaticDifferentiationTensor” section.