rich-payne / dreamer Goto Github PK

Some of the equations are not rendering. Perhaps move them to the vignette, and then reference the vignette in the README.

library(dreamer)
mcmc <- dreamer_mcmc(
  data = NULL,
  model_independent_binary(
    mu_b1 = c(0, 0),
    sigma_b1 = c(5, 5),
    doses = c(0, 1),
    link = "logit"
  )
)
#> Warning in names(model_list)[i] <- paste0("model_", i, "_",
#> class(model_list[[i]])): number of items to replace is not a multiple of
#> replacement length
#> model_1_dreamer_independent_binary
#> start : 2022-10-27 09:26:19.981
#> Compiling model graph
#>    Resolving undeclared variables
#>    Allocating nodes
#> Graph information:
#>    Observed stochastic nodes: 0
#>    Unobserved stochastic nodes: 2
#>    Total graph size: 11
#> 
#> Initializing model
#> finish: 2022-10-27 09:26:20.026
#> total : 0.04 secs

^{Created on 2022-10-27 by the reprex package (v2.0.1)}

Rather than try to name the models, an error should be thrown requiring the user to name the arguments.

May need to change versions, e.g., r-lib/actions#552, https://github.com/omnideconv/SimBu/actions/runs/2207112235/workflow#L45

We just learned about a real-world clinical use case that will probably want to use dreamer and needs to adjust for covariates. Sound feasible to add to the dreamer models?

https://doi.org/10.1080/10543406.2023.2292214

Show off new features of #24.

To make it clear that sigma is a standard deviation.

Error is generated for independent models, but the error is not informative. For binary models, ensure values are 0/1.

library(dreamer)
data <- data.frame(dose = factor(1:2), response = 11:12)
dreamer_mcmc(
  data,
  mod = model_independent(
    mu_b1 = rep(0, 2),
    sigma_b1 = rep(1, 2),
    doses = c(1, 2),
    shape = 1,
    rate = .01
  )
)
#> mod
#> start : 2021-08-26 15:37:02.000
#> Error: Doses specified do not match doses in data

^{Created on 2021-08-26 by the reprex package (v0.3.0)}

This dreamer class may be redundant.

Easiest with usethis::use_github_action("pkgdown").

This could be a simple wrapper called model_hyper_emax. This would be an alternative to "hacking" the prior of b4 to be N(0, .00001^2) or similar in the emax model.

Bug when plotting data with an individual model. Occurs because attr(x, "response_type") is null for individual models.

These are currently stochastic and time-consuming.

e.g. dreamer_mcmc()

Update EMAX parameterization to match standard MCP-Mod form (see: https://cran.r-project.org/web/packages/DoseFinding/DoseFinding.pdf)

This will make the visual comparison better and avoid overlapping lines.

jags_rng, jags_seed

E.g. avoid things like any(grepl("binary", class(x))).