scheduler: S4 Scheduler Class Object
In tpq/rarsim: Response-Adaptive Randomization Simulation Station

Description Usage Arguments Slots Methods Getters Setters Plotters Samplers

S4 Scheduler Class Object

## S4 method for signature 'scheduler'
show(object)

getAllocation(object)

getHistory(object)

getEntropy(scheduler)

scheduler.start(
  prior.mean,
  prior.var,
  N.burn.in,
  sampler = "sampler.thompson",
  heuristic = FALSE,
  prior.nu,
  prior.alpha,
  prior.beta,
  ...
)

scheduler.update(scheduler, data.ingest, N.allocate)

plotPrior(scheduler)

plotPosterior(scheduler)

plotHistory(scheduler)

plotAllocation(scheduler)

sampler.thompson(scheduler, ...)

sampler.fixed(scheduler, ...)

sampler.auc.cutoff(scheduler, cutoff = 0, ...)

sampler.auc.reference(scheduler, reference = NULL, ...)

sampler.ucb1(scheduler, c = 2, batch = TRUE, ...)

sampler.ucb1.normal(scheduler, c = 16, batch = TRUE, ...)

sampler.epsilon.greedy(scheduler, epsilon = 0.1, ...)

`object, scheduler`	A `scheduler` object.
`prior.mean`	A vector of prior means. One mean for each arm
`prior.var`	A vector of prior variances. One variance for each arm.
`N.burn.in`	An integer. The number patients to allocate each arm during the initial 'burn-in' phase. Ideally, around 20-30 per arm.
`sampler`	A string. The sampling method used to allocate patients.
`heuristic`	A logical. Toggles whether to assume precision is known, but instead estimate it from the data. When taking this approach, the conjugate prior distribution is normal.
`prior.nu, prior.alpha, prior.beta`	The prior statistics for each arm. These apply when the conjugate prior distribution is normal-gamma (i.e., when `heuristic = TRUE`).
`...`	Arguments passed to `sampler` function. For the `sampler` itself, a null argument.
`data.ingest`	A list of rewards. Must match the structure of `scheduler@ingest`.
`N.allocate`	An integer. The total number of patients to allocate next.
`cutoff`	The cutoff used for `sampler.auc.cutoff`. Allocation is proportional to area under the posterior greater than this cutoff.
`reference`	The reference used for `sampler.auc.reference`. Allocation is proportional to area under the posterior greater than the posterior mean of this reference.
`c`	The exploration hyper-parameter used by several UCB algorithms.
`batch`	Toggles whether the UCB algorithm should allocate patients to groups randomly based on the rank of the UCB score.
`epsilon`	The epsilon used for `sampler.epsilon.greedy`. Allocation is to a random group `epsilon` percent of the time; otherwise, it is to the group with the largest posterior mean.

prior.df: The degrees-of-freedom for the t-distribution that describes the uncertainty about the mean. Set to Inf when heuristic = TRUE.
prior.mean: The prior statistics for each arm.
prior.var: The uncertainty about the mean. When the conjugate prior distribution is normal-gamma, this slot contains the variance of the marginal distribution of the mean.
prior.nu,prior.alpha,prior.beta: The prior statistics for each arm. These apply when the conjugate prior distribution is normal-gamma.
N.burn.in: An integer. The number of patients in each arm at first time step.
K.arms: An integer. The number of experimental groups (also called 'arms').
step: An integer. The number of time steps so far. Each time step is a kind of 'mini-trial'. This initializes at 0.
rewards: A list of all rewards observed to date.
meta: A list of tables for all rewards and meta-data available. This is optional, but may be preferred for complex simulations.
online.count,online.sum,online.mean,online.var,online.prec: The empiric statistics observed to date.
heuristic: A logical. Toggles whether to assume precision is known, but instead estimate it from the data. When taking this approach, the conjugate prior distribution is normal.
post.df: The degrees-of-freedom for the t-distribution that describes the uncertainty about the mean. Set to Inf when heuristic = TRUE.
post.mean: The posterior statistics for each arm.
post.var: The uncertainty about the mean. When the conjugate prior distribution is normal-gamma, this slot contains the variance of the marginal distribution of the mean.
post.nu,post.alpha,post.beta: The posterior statistics for each arm. These apply when the conjugate prior distribution is normal-gamma.
sampler.args: A list. Arguments for the sampling method.
sampler: A string. The sampling method used to allocate patients.
dynamic.count: Like @online.count, but used by the UCB algorithm.
allocation: A vector of groups to which to allocate new patients.
ingest: A list describing the structure of the data expected by the next scheduler.update call. This slot is only used to verify the incoming data.
allocation: The most recent patient allocations. This slot implies the structure of the @ingest slot.
history.post: A table of all posterior statistics and the allocation ratios at each time step in the experiment. At @step = 0, the table records the prior statistics and burn-in conditions.
history: A list of lists of all rewards. Unlike @rewards, the rewards here are organized by time step.

show: Method to show scheduler object.

getAllocation: Method to retrieve allocations from a scheduler object. This function returns a vector of groups to which to allocate new patients.

getHistory: Method to retrieve history from a scheduler object. This function returns a table of all posterior statistics and the allocation ratios at each time step in the experiment.

getEntropy: Method to retrieve allocation entropy from a scheduler object. This function returns the entropy of the allocation ratios at each time step.

scheduler.start: Method to initiate a scheduler object. This function returns an updated scheduler object.

scheduler.start: Method to update a scheduler object with new data. This function returns an updated scheduler object.

plotPrior: Method to plot prior distributions.

plotPosterior: Method to plot the most recent posterior distributions.

plotHistory: Method to plot all posterior distributions.

plotAllocation: Method to plot the allocation ratios for each time step.

sampler.thompson: Method to allocate patients by Thompson sampling. This function returns an integer corresponding to the group to which the patient is randomly allocated.

sampler.fixed: Method to allocate patients at a fixed ratio. This function returns an integer corresponding to the group to which the patient is randomly allocated.

sampler.auc.cutoff: Method to allocate patients proportional to the probability that the posterior expected reward is greater than a cutoff. This function returns an integer corresponding to the group to which the patient is randomly allocated.

sampler.auc.reference: Method to allocate patients proportional to the probability that the posterior expected reward is greater than a reference. This function returns an integer corresponding to the group to which the patient is randomly allocated.

sampler.ucb1: Method to allocate patients based on a UCB algorithm, assuming that the rewards fall within the range of 0 and 1. This function returns an integer corresponding to the group to which the patient is randomly allocated.

sampler.ucb1.normal: Method to allocate patients based on a UCB algorithm, assuming that the rewards are normally distributed. This function returns an integer corresponding to the group to which the patient is randomly allocated.

sampler.epsilon.greedy: Method to allocate patients based on an epsilon-greedy algorithm. By default, epsilon = 0.1. This function returns an integer corresponding to the group to which the patient is randomly allocated.

tpq/rarsim documentation built on April 4, 2020, 3:49 a.m.