package policyeval
- Alphabetic
- Public
- All
Type Members
- case class BanditEstimator(lower: Double, upper: Double) extends Product with Serializable
-
class
CressieRead extends Aggregator[CressieReadInput, CressieReadBuffer, Double] with Serializable with SynapseMLLogging
Cressie-Read off-policy evaluation metric.
Cressie-Read off-policy evaluation metric.
Background http://www.machinedlearnings.com/2020/12/distributionally-robust-contextual.html
- final case class CressieReadBuffer(wMin: Float = 0, wMax: Float = 0, n: KahanSum = 0, sumw: KahanSum = 0, sumwsq: KahanSum = 0, sumwr: KahanSum = 0, sumwrsqr: KahanSum = 0, sumr: KahanSum = 0) extends Product with Serializable
- final case class CressieReadInput(probLog: Float, reward: Float, probPred: Float, count: Float, wMin: Float, wMax: Float) extends Product with Serializable
-
class
CressieReadInterval extends Aggregator[CressieReadIntervalInput, CressieReadIntervalBuffer, BanditEstimator] with Serializable with SynapseMLLogging
Cressie-Read with intervals off-policy evaluation metric.
Cressie-Read with intervals off-policy evaluation metric.
Background http://www.machinedlearnings.com/2020/12/distributionally-robust-contextual.html
- final case class CressieReadIntervalBuffer(wMin: Float = 0, wMax: Float = 0, rewardMin: Float = 0, rewardMax: Float = 0, n: KahanSum = 0, sumw: KahanSum = 0, sumwsq: KahanSum = 0, sumwr: KahanSum = 0, sumwsqr: KahanSum = 0, sumwsqrsq: KahanSum = 0) extends Product with Serializable
- final case class CressieReadIntervalInput(probLog: Float, reward: Float, probPred: Float, count: Float, wMin: Float, wMax: Float, rewardMin: Float, rewardMax: Float) extends Product with Serializable
-
class
Ips extends Aggregator[IpsInput, IpsBuffer, Float] with Serializable with SynapseMLLogging
Simplest off-policy evaluation metric: IPS (Inverse Propensity Score)
Simplest off-policy evaluation metric: IPS (Inverse Propensity Score)
See https://courses.cs.washington.edu/courses/cse599m/19sp/notes/off_policy.pdf
- final case class IpsBuffer(exampleCount: Float, weightedReward: Float) extends Product with Serializable
- final case class IpsInput(probabilityLogged: Float, reward: Float, probabilityPredicted: Float, count: Float) extends Product with Serializable
-
class
Snips extends Aggregator[SnipsInput, SnipsBuffer, Float] with Serializable with SynapseMLLogging
SNIPS off policy estimator: the Self-Normalized Estimator for Counterfactual Learning
SNIPS off policy estimator: the Self-Normalized Estimator for Counterfactual Learning
See https://papers.nips.cc/paper/2015/file/39027dfad5138c9ca0c474d71db915c3-Paper.pdf
- final case class SnipsBuffer(weightedExampleCount: Double, weightedReward: Double) extends Product with Serializable
- final case class SnipsInput(probabilityLogged: Float, reward: Float, probabilityPredicted: Float, count: Float) extends Product with Serializable
Value Members
- object PolicyEvalUDAFUtil