Creating an example of BayesSurvive(MRF.G = FALSE)

[wleoncio]

For #11, I am trying to create an example that triggers func_MCMC_graph(). I've slightly adapted the example for BayesSurvive() as follows:

library("BayesSurvive")
set.seed(123)

# Load the example dataset
data("simData", package = "BayesSurvive")

dataset <- list(
  "X" = simData[[1]]$X,
  "t" = simData[[1]]$time,
  "di" = simData[[1]]$status
)

# Initial value: null model without covariates
initial <- list("gamma.ini" = rep(0, ncol(dataset$X)))
# Hyperparameters
hyperparPooled <- list(
  "c0"     = 2, # prior of baseline hazard
  "tau"    = 0.0375, # sd (spike) for coefficient prior
  "cb"     = 20, # sd (spike) for coefficient prior
  "pi.ga"  = 0.02, # prior variable selection probability for standard Cox models
  "a"      = -4, # hyperparameter in MRF prior
  "b"      = 0.1, # hyperparameter in MRF prior
  "G"      = simData$G # hyperparameter in MRF prior
)

# run Bayesian Cox with graph-structured priors
fit <- BayesSurvive(
  survObj = dataset, hyperpar = hyperparPooled, MRF.G = FALSE,
  initial = initial, nIter = 100
)

This creates an "operator invalid for atomic vectors" error on https://github.com/ocbe-uio/BayesSurvive/blob/7ee9d98aaaa5db2556a95c0bf2b53168f8f973bb/R/BayesSurvive.R#L143C1-L143C60 because survObj is, in this case, not a nested list.

This should be straightforward to fix (e.g. by using survObj$X in this particular case), but perhaps this is not a bug and I am just using the wrong dataset. What would you suggest?

[zhizuio]

There is one example in README similar to your case, i.e. "Run a 'Pooled' Bayesian Cox model with graphical learning". There, I specified survObj as a list and added a few more hyperparameters in hyperpar.

I know this is a bit confusing, since I wanted to include quite many methods together via the combinations of 2 arguments i.e. model.type and MRF.G . Note that model.type has only two valid options (model.type="Sub-struct" seems redundant, maybe not? I need to check). Therefore, there should at least 4 combinations for 4 methods, which are the 4 examples in README.

MRF.G = FALSE enables graphical learning for a Markov random field prior (no specified sort of true graph). This requires survObj to be a nested list suitable for one homogeneous group (a list of only one component with model.type= 'Pooled') AND multiple groups (a list of multiple components with model.type= 'CoxBVSSL') of individuals.

MRF.G = TRUE disables graphical learning for a Markov random field prior, instead specifying sort of true graph. The simplest case will be for homogeneous individuals (i.e. no groups of individuals), and then model.type= 'Pooled' and survObj is NOT a nested list. If individuals are divided into multiple groups (of course survObj should be a nested list), we also need to specify a list of multiple true graphs for the hyperparameter hyperpar$G and specify model.type= 'CoxBVSSL'.

Probably we need a better idea to re-structure the combinations of arguments for many methods. As I mentioned, we also wanted to include another method implemented in the R package psbcSpeedUp:-)

[zhizuio]

I have just checked and thought a bit carefully. model.type="Sub-struct" is not redundant! func_MCMC_graph() has two big for-loops. If func_MCMC_graph() is triggered by MRF.G = FALSE, the 1st for-loop inside will be run. If the 2nd for-loop in func_MCMC_graph() is not triggered, it will be model.type="Sub-struct"; otherwise it will be model.type="CoxBVSSL".

Sorry for the confusing structures!

[wleoncio]

Thanks for the clarification! Looks to me like there's no need to modify the code for this, but I wanna get those examples from the README and turn them into unit tests to keep me on track while I work on #11. Sounds good?

[zhizuio]

Good! 13. juni 2024 kl. 06:17 skrev Waldir Leoncio ***@***.***>:  Thanks for the clarification! Looks to me like there's no need to modify the code for this, but I wanna get those examples from the README and turn them into unit tests to keep me on track while I work on #11<#11>. Sounds good? — Reply to this email directly, view it on GitHub<#13 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AJOLJRZQW5PK5BOJJOK4WRTZHEMMJAVCNFSM6AAAAABJGDOKIOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCNRUGMZTGNJYGA>. You are receiving this because you commented.Message ID: ***@***.***>

[wleoncio]

85e59f1 says it solves the issue, but that's because I messed up and thought I needed MRF.G = TRUE, but it actually needs to be FALSE for func_MCMC_graph() to be called. So I'll need to work a little bit more on that README.md file before I can issue a PR for this one and then continue working on #11.