merliseclyde/BAS

Error in BAS:::coef.bas(..., estimator = "MPM"): Error in eval(object$call$weights): object 'object' not found

vandenman opened this issue · 1 comments

vandenman commented

Describe the bug
Calling BAS:::coef.bas(basobj, estimator = "MPM") may error depending on how basobj was created.

To Reproduce
The first example works as intended. The second example changes the seemingly harmless formula = M ~ So + Ed + Po1 + Po2 into formula = form where form is defined earlier form = M ~ So + Ed + Po1 + Po2.

data(UScrime, package = "MASS")
UScrime <- UScrime[, 1:5]

crime.bic <- BAS::bas.lm(
  formula = M ~ So + Ed + Po1 + Po2,
  data = UScrime,
  prior = "JZS",
  initprobs = c(1, 0.5, 0.5, 0.5, 0.5),
  renormalize = TRUE
)
BAS:::coef.bas(crime.bic, estimator = "MPM")
#> 
#>  Marginal Posterior Summaries of Coefficients: 
#> 
#>  Using  MPM 
#> 
#>  Based on the top  1 models 
#>            post mean  post SD    post p(B != 0)
#> Intercept  138.57447    1.40051    1.00000     
#> So          11.53684    3.13294    1.00000     
#> Ed           0.00000    0.00000    0.00000     
#> Po1          0.00000    0.00000    0.00000     
#> Po2         -0.14815    0.05367    1.00000

form <- M ~ So + Ed + Po1 + Po2
crime.bic2 <- BAS::bas.lm(
  formula = form,
  data = UScrime,
  prior = "JZS",
  initprobs = c(1, 0.5, 0.5, 0.5, 0.5),
  renormalize = TRUE
)
BAS:::coef.bas(crime.bic2, estimator = "MPM")
#> Error in eval(object$call$weights): object 'object' not found

Created on 2022-02-02 by the reprex package (v2.0.1)

What goes wrong, is that this code

BAS/R/coefficients.R

Lines 87 to 99 in f92fc3e

object <- bas.lm(
eval(object$call$formula),
data = eval(object$call$data),
weights = eval(object$call$weights),
n.models = 1,
alpha = object$g,
initprobs = object$probne0,
prior = object$prior,
modelprior = object$modelprior,
update = NULL,
bestmodel = models,
prob.local = 0.0
)

is doing things with eval. However, since R is lazy, the evals are not evaluated until we get to

BAS/R/bas_lm.R

Line 514 in f92fc3e

mf <- eval(mf, parent.frame())

where R tries to do eval(expression(stats::model.frame(formula = eval(object$call$formula), data = eval(object$call$data), weights = eval(object$call$weights), drop.unused.levels = TRUE)), parent.frame()) but this fails. I don't really understand why this fails though, as env <- parent.frame(); env$object during debugging does return the objects that R cannot find.

Expected behavior
The same output as in the first version.

Desktop:

  • OS: Ubuntu 20.04
  • R Version 4.1.2
merliseclyde commented

Thank you for the very detailed reproducible example. Issue was with the formula environment as discussed in
StackOverflow https://stackoverflow.com/questions/61164404/call-to-weight-in-lm-within-function-doesnt-evaluate-properly/61164660#61164660?newreg=04a2b71c6da04693a5b172a54a4a43b0
which was key to solving!

Please let me know if there any remaining issues that crop up - m

  • 🎉1