Package ai.djl.training.optimizer
Class RmsProp.Builder
java.lang.Object
ai.djl.training.optimizer.Optimizer.OptimizerBuilder<RmsProp.Builder>
ai.djl.training.optimizer.RmsProp.Builder
- Enclosing class:
- RmsProp
The Builder to construct an
RmsProp
object.-
Method Summary
Modifier and TypeMethodDescriptionbuild()
Builds aRmsProp
block.optCentered
(boolean centered) Sets which version of RMSProp to use.optEpsilon
(float epsilon) Sets \(epsilon\) - a small quantity for numerical stability.optLearningRateTracker
(ParameterTracker learningRateTracker) Sets theParameterTracker
for this optimizer.optMomentum
(float momentum) Sets the momentum factor.optRho
(float rho) Sets the decay factor for the moving average over the past squared gradient.protected RmsProp.Builder
self()
Methods inherited from class ai.djl.training.optimizer.Optimizer.OptimizerBuilder
optBeginNumUpdate, optClipGrad, optWeightDecays, setRescaleGrad
-
Method Details
-
self
- Specified by:
self
in classOptimizer.OptimizerBuilder<RmsProp.Builder>
-
optLearningRateTracker
Sets theParameterTracker
for this optimizer.- Parameters:
learningRateTracker
- theParameterTracker
to be set- Returns:
- this
Builder
-
optRho
Sets the decay factor for the moving average over the past squared gradient.- Parameters:
rho
- the decay factor for the moving average over past squared gradient- Returns:
- this
Builder
-
optMomentum
Sets the momentum factor. This is only used if centered is set to true.- Parameters:
momentum
- the momentum factor- Returns:
- this
Builder
-
optEpsilon
Sets \(epsilon\) - a small quantity for numerical stability.- Parameters:
epsilon
- a small quantity for numerical stability- Returns:
- this
Builder
-
optCentered
Sets which version of RMSProp to use.True: Grave's version False: Tieleman and Hinton's version
- Parameters:
centered
- the RMSProp version- Returns:
- this
Builder
-
build
Builds aRmsProp
block.- Returns:
- the
RmsProp
block
-