Error sampling from prior for mixture model

Interfaces brms
1

Question in brief: I get an error when I set ‘sample_prior’ to ‘yes’ or TRUE when trying to sample from a model based on this formula and these priors:

tmp.bf ← bf(RT ~ 1,
theta2 ~ cond * group,
family = mixture(brmsfamily(‘gamma’, link = ‘log’, link_shape = ‘log’),
brmsfamily(‘gamma’, link = ‘log’, link_shape = ‘log’),
order = TRUE))

tmp.priors ← set_prior(‘logistic(-3, 0.35)’, class = ‘Intercept’, dpar = ‘theta2’) +
set_prior(‘normal(-0.25, 0.5)’, class = ‘Intercept’, dpar = ‘mu1’) +
set_prior(‘normal(0.8, 0.5)’, class = ‘Intercept’, dpar = ‘mu2’) +
set_prior(‘gamma(9,3.5)’, class = ‘shape1’) +
set_prior(‘gamma(2.25, 1.25)’, class = ‘shape2’) +
set_prior(‘logistic(0, 0.1)’, class = ‘b’, dpar = ‘theta2’)

bmt ← brm(tmp.bf, data = mydata, sample_prior = ‘yes’, prior = tmp.priors)

The error is this:

SYNTAX ERROR, MESSAGE(S) FROM PARSER:
Variable definition base type mismatch, variable declared as base type vector variable definition has
base type realVariable definition dimensions mismatch, definition specifies 1, declaration specifies 0
error in ‘model53682ca521fe_file536814f21ac0’ at line 87, column 56

85:   real prior_Intercept_mu2 = normal_rng(0.8,0.5);
86:   real prior_shape2 = gamma_rng(2.25,1.25);
87:   vector[Kc_theta2] prior_b_theta2 = logistic_rng(0,0.1);
                                                           ^
88:   real prior_Intercept_theta2 = logistic_rng(-3,0.35);

Error in stanc(model_code = paste(program, collapse = “\n”), model_name = model_cppname, :
failed to parse Stan model ‘file536814f21ac0’ due to the above error.

SYNTAX ERROR, MESSAGE(S) FROM PARSER:
Variable definition base type mismatch, variable declared as base type vector variable definition has
base type realVariable definition dimensions mismatch, definition specifies 1, declaration specifies 0
error in ‘model5368692d6f86_file536814f21ac0’ at line 87, column 56

85:   real prior_Intercept_mu2 = normal_rng(0.8,0.5);
86:   real prior_shape2 = gamma_rng(2.25,1.25);
87:   vector[Kc_theta2] prior_b_theta2 = logistic_rng(0,0.1);
                                                           ^
88:   real prior_Intercept_theta2 = logistic_rng(-3,0.35);

Error in stanc(model_code = paste(program, collapse = “\n”), model_name = model_cppname, :
failed to parse Stan model ‘file536814f21ac0’ due to the above error.

I’ve tried different kinds of priors for the b_theta2 effects but I always get essentially the same error. Without a prior on those effects, i.e. :

test.priors ← set_prior(‘logistic(-3, 0.35)’, class = ‘Intercept’, dpar = ‘theta2’) +
set_prior(‘normal(-0.25, 0.5)’, class = ‘Intercept’, dpar = ‘mu1’) +
set_prior(‘normal(0.8, 0.5)’, class = ‘Intercept’, dpar = ‘mu2’) +
set_prior(‘gamma(9,3.5)’, class = ‘shape1’) +
set_prior(‘gamma(2.25, 1.25)’, class = ‘shape2’)

the model runs without problems. The model also runs with the logistic prior, but with sample_prior = "no". Any ideas why this occurs? What am I doing wrong?

Of course I’d be happy to provide more background in case that helps (but unfortunately I can’t provide the data without further ado).

Update: actually I just realized that model also runs with sample_prior = "only", which to me suggests it’s not just me miss-specifying the prior. Is this perhaps a bug then?

Please also provide the following information in addition to your question:

  • Operating System: Windows 10 Enterprise, Version 1909, OS build 18363.657
  • brms Version: 2.10.0
2

I think it is a bug. Let me check whether I can do something about it.

3

I think you need to update brms to at least 2.12.0. There it works for me.

1 Like
4

Yes, thank you, it works in 2.12.0.

I thought I had updated to the latest version but didn’t realize that the update had failed because I had two vignette pdfs open in acrobat reader…