RemixAutoML (version 0.4.2)

RL_Update: RL_Update

Description

RL_Update updates the bandit probabilities for selecting different grids

Usage

RL_Update(
  ExperimentGrid = ExperimentGrid,
  MetricSelection = MetricSelection,
  ModelRun = run,
  NEWGrid = NewGrid,
  TrialVector = Trials,
  SuccessVector = Successes,
  GridIDS = GridIDs,
  BanditArmsCount = BanditArmsN,
  RunsWithoutNewWinner = RunsWithoutNewWinner,
  MaxRunsWithoutNewWinner = MaxRunsWithoutNewWinner,
  MaxNumberModels = MaxNumberModels,
  MaxRunMinutes = MaxRunMinutes,
  TotalRunTime = TotalRunTime,
  BanditProbabilities = BanditProbs
)

Arguments

ExperimentGrid

This is a data.table of grid params and model results

MetricSelection

The chosen metric to evalute models

ModelRun

Model iteration number

NEWGrid

Previous grid passed in

TrialVector

Numeric vector with the total trials for each arm

SuccessVector

Numeric vector with the total successes for each arm

GridIDS

The numeric vector that identifies which grid is which

BanditArmsCount

The number of arms in the bandit

RunsWithoutNewWinner

Counter of the number of models previously built without being a new winner

MaxRunsWithoutNewWinner

Maximum number of models built without a new best model (constraint)

MaxNumberModels

Maximum number of models to build (constraint)

MaxRunMinutes

Run time constraint

TotalRunTime

Cumulative run time in minutes

BanditProbabilities

Inital probabilities from RL_Initialize()

See Also

Other Reinforcement Learning: RL_Initialize(), RL_ML_Update(), RPM_Binomial_Bandit()

Examples

# NOT RUN {
RL_Update_Output <- RL_Update(
  ExperimentGrid = ExperimentGrid,
  MetricSelection = MetricSelection,
  ModelRun = run,
  NEWGrid = NewGrid,
  TrialVector = Trials,
  SuccessVector = Successes,
  GridIDS = GridIDs,
  BanditArmsCount = BanditArmsN,
  RunsWithoutNewWinner = RunsWithoutNewWinner,
  MaxRunsWithoutNewWinner = MaxRunsWithoutNewWinner,
  MaxNumberModels = MaxNumberModels,
  MaxRunMinutes = MaxRunMinutes,
  TotalRunTime = TotalRunTime,
  BanditProbabilities = BanditProbs)
BanditProbs <- RL_Update_Output[["BanditProbs"]]
Trials <- RL_Update_Output[["Trials"]]
Successes <- RL_Update_Output[["Successes"]]
NewGrid <- RL_Update_Output[["NewGrid"]]
# }