learn_escalation_rule {RLescalation}R Documentation

Build an Optimal Dose Escalation Rule using Reinforcement Learning

Description

Build an Optimal Dose Escalation Rule using Reinforcement Learning

Usage

learn_escalation_rule(
  J,
  target,
  epsilon,
  delta,
  N_total,
  N_cohort,
  seed = NULL,
  rl_config = rl_config_set(),
  rl_scenarios = NULL,
  output_dir = format(Sys.time(), "%Y%m%d_%H%M%S"),
  output_base_dir = "escalation_rules",
  checkpoint_dir = "checkpoints"
)

Arguments

J

A positive integer value. The number of doses.

target

A positive numeric value. The target DLT probability.

epsilon

A positive numeric value. The acceptable range of target DLT probabilities is defined as [target - epsilon, target + epsilon].

delta

A positive numeric value. The unacceptable ranges of target DLT probabilities are defined as [0, target - delta] and [target + delta, 1].

N_total

A positive integer value. The total number of patients.

N_cohort

A positive integer value. The number of patients for each cohort.

seed

An integer value. Random seed for reinforcement learning.

rl_config

A list. Other settings for reinforcement learning. See rl_config_set for details.

rl_scenarios

A list. Scenarios used for reinforcement learning. Default is NULL (use scenarios in the Sect. 2.2 of the original paper). See compute_rl_scenarios for details.

output_dir

A character value. Directory name or path to store the built escalation rule. Default is the current datetime.

output_base_dir

A character value. Parent directory path where the built escalation rule will be stored. Valid only if 'output_dir' does not contain '/'. Default is "escalation_rules".

checkpoint_dir

A character value. Parent directory path to save checkpoints. It enables you to resume learning from that point onwards. Default is "checkpoints".

Value

An EscalationRule object.

Examples

library(RLescalation)

# We obtain an optimal dose escalation rule by executing `learn_escalation_rule()`.
## Not run: 
escalation_rule <- learn_escalation_rule(
  J = 6, target = 0.25, epsilon = 0.04, delta = 0.1,
  N_total = 36, N_cohort = 3, seed = 123,
  rl_config = rl_config_set(iter = 1000)
)
## End(Not run)


[Package RLescalation version 1.0.2 Index]