package value
- Alphabetic
- Public
- All
Type Members
-
case class
ConstantStep(value: Double, time: Time) extends Ordered[ConstantStep] with Product with Serializable
Exponential recency-weighted average.
Exponential recency-weighted average. This is similar to a weighted average, but instead of weighting by the count, it uses a constant weighting factor.
TODO consider changing Numeric to ToDouble?
TODO there is some interesting thing going on here, where you're summing up constantly weighted factors... and then in the importance weighting you're doing the same thing, you just send in a different weight every time.
The time is a way of skipping big swathes and injecting zeros in, but it's really the same.
You can maybe also think about this like you're just keeping track of some weighted numerator, but, same thing, injecting more sum of the weights into the denominator, but nothing, zeros, into the top.
That would be a nice thing to unify together.
- class ConstantStepGroup extends Group[ConstantStep]
-
sealed
trait
DecayState[A] extends Product with Serializable
This represents a value that's weighted as you move away from it.
This represents a value that's weighted as you move away from it. This is useful because we can KEEP GOING, and continue to weight it.
-
case class
WeightedAverage(weightSum: Weight, value: Double) extends Product with Serializable
This is of course extremely similar to the averaged value implementation in Algebird...
This is of course extremely similar to the averaged value implementation in Algebird... it just keeps track of a numerator AND denominator
Value Members
- object ConstantStep extends Serializable
- object DecayState extends Serializable
- object WeightedAverage extends Serializable
-
object
WeightedAverageGroup extends Group[WeightedAverage] with CommutativeGroup[WeightedAverage]
Group implementation for WeightedAverage.
edit this text on github
ScalaRL
This is the API documentation for the ScalaRL functional reinforcement learning library.
Further documentation for ScalaRL can be found at the documentation site.
Check out the ScalaRL package list for all the goods.