NEWS.md

In tpq/rarsim: Response-Adaptive Randomization Simulation Station

rarsim 0.1.4

• Improvements
  • Track total samples allocated with run.benchmark function

rarsim 0.1.3

• Improvements
  • Add better handling for the zero probability error

rarsim 0.1.2

• Track allocation entropy
  • New getEntropy calculates entropy of allocation ratios at each time step
  • Track entropy with run.benchmark function
• Improvements
  • Add handling for sampler.ucb1.normal when all UCB are the same
  • Set heuristic = FALSE by default

rarsim 0.1.1

• New sampler methods
  • Add dynamic.count slot for UCB algorithm to batch allocate patients
  • New sampler.ucb1 uses a batch-based upper confidence bound (UCB) algorithm
  • New sampler.ucb1.normal uses a batch-based upper confidence bound (UCB) algorithm
  • New sampler.epsilon.greedy uses an epsilon-greedy algorithm

rarsim 0.1.0

• Change argument class
  • Have scheduler.start expect a character for sampler
  • Have run.benchmark expect a character for how.stats

rarsim 0.0.9

• Pass arguments to sampler function
  • Add null ... arguments to all sampler functions

rarsim 0.0.8

• Launch unknown variance update
  • Add prior.nu and post.nu slots
  • Add prior nu, alpha, and beta arguments to scheduler.start function
  • Add procedure to guess prior nu, alpha, and beta from prior var
  • Make sampler.auc use t-distribution too
• Pass arguments to sampler function
  • Add sampler.auc.cutoff to allocate patients based on p(sample > cutoff)
  • Add sampler.auc.reference to allocate patients based on p(sample > reference)
  • The scheduler.start will save ... args to sampler.args slot
  • These get passed to the sampler during an update

rarsim 0.0.7

• Prepare unknown variance update
  • Remove prior.prec and post.prec slots for simplicity
  • Add slots to store degrees-of-freedom (set to Inf when heuristic = TRUE)
  • Replace Normal distribution with t distribution
• Update run.benchmark method
  • New fast argument to measure p-value at final time step only

rarsim 0.0.6

• Add sampler methods
  • New sampler.fixed will allocate patients at a fixed ratio
  • New sampler.auc will allocate patients proportional to p(posterior > cutoff)
• Prepare unknown variance update
  • Add prior.alpha and prior.beta slots to anticipate normal-gamma conjugate
  • (prior.var and prior.prec save marginal values for backwards compatibility)
  • Add post.alpha and post.beta slots to anticipate normal-gamma conjugate
  • (post.var and post.prec save marginal values for backwards compatibility)
  • Add heuristic argument to toggle normal-gamma use

rarsim 0.0.5

• Improvements
  • Have scheduler object save name of sampler method
  • New run.trial iterates simulator.draw and scheduler.update steps
  • New run.benchmark estimates proportion of (p < alpha)
  • Name all sampler methods with sampler. prefix

rarsim 0.0.4

• Improvements
  • Consistently name reward lists
  • New stats.anova computes a p-value from the rewards
  • New stats.ttest computes p-values from the rewards
  • New stats.empiric computes empiric p-values

rarsim 0.0.3

• Get ready for statistics modules
  • Add empty @meta slot for future updates
  • New getDivergences function can calculate divergence between two posteriors
  • New getPermutations function can permute rewards in scheduler object

rarsim 0.0.2

• Improvements
  • Add first edition of README for collaborators
• Bug fixes
  • Fix error caused by trying to name simulator rewards

rarsim 0.0.1

• Create S4 scheduler class for scheduling trials
  • This object stores past rewards, updates all posteriors, and recommends new allocations
  • Key function scheduler.start creates a new scheduler object
  • Key function scheduler.update updates a scheduler object with new data
  • Add some getters and basic plots
• Create S4 simulator class for simulating data
  • This object stores a pool of all data and the most recent draw from this pool
  • Key function simulator.start creates a new simulator object
  • Key function simulator.draw draws data from a simulator object
  • Add some getters and setters
• Design notes
  • The scheduler decides on the allocation, which is passed to the simulator
  • The simulator draws patient data from the pool, which is passed back to the scheduler
  • This process is repeated until the analyst is satisfied

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