Can't use Shinystan to look at rstanarm glmer results

[rayleigh]

For a project, I used RStanarm's glmer function in the following way:

stan_fit <- stan_glmer(formula = formula(reg_formula), data = data_matrix,
family = "binomial", iter = 500, sample_file = output_filename)

However, when I tried to upload the files and look at the results in Shinystan, I got the following warnings and error messages:

> naes_one <- read_stan_csv(c("Annenberg_analysis_1_sample_output1", "Annenberg_analysis_1_sample_output2", "Annenberg_analysis_1_sample_output3", "Annenberg_analysis_1_sample_output4"))
There were 12 warnings (use warnings() to see them)
> warnings()
Warning messages:
1: In if (adapt_term_lineno == len) adaptation_info <- "" else { ... :
the condition has length > 1 and only the first element will be used
2: In (adapt_term_lineno + 1):(time_lineno - 1) :
numerical expression has 2 elements: only the first used
3: In 1:(adapt_term_lineno - 1) :
numerical expression has 2 elements: only the first used
4: In if (adapt_term_lineno == len) adaptation_info <- "" else { ... :
the condition has length > 1 and only the first element will be used
5: In (adapt_term_lineno + 1):(time_lineno - 1) :
numerical expression has 2 elements: only the first used
6: In 1:(adapt_term_lineno - 1) :
numerical expression has 2 elements: only the first used
7: In if (adapt_term_lineno == len) adaptation_info <- "" else { ... :
the condition has length > 1 and only the first element will be used
8: In (adapt_term_lineno + 1):(time_lineno - 1) :
numerical expression has 2 elements: only the first used
9: In 1:(adapt_term_lineno - 1) :
numerical expression has 2 elements: only the first used
10: In if (adapt_term_lineno == len) adaptation_info <- "" else { ... :
the condition has length > 1 and only the first element will be used
11: In (adapt_term_lineno + 1):(time_lineno - 1) :
numerical expression has 2 elements: only the first used
12: In 1:(adapt_term_lineno - 1) :
numerical expression has 2 elements: only the first used
> launch_shinystan(naes_one)

Creating shinystan object...
Error in if (.used_vb(x) || !(algo %in% c("NUTS", "HMC"))) warning("Many features are only available for models fit using\n algorithm NUTS or algorithm HMC.", :
missing value where TRUE/FALSE needed

[rayleigh]

I just tried looking at the Stanfit object in Shinystan right after getting glmer results. Shinystan launches fine in that case. So, is this an Rstanarm issue instead?

[jgabry]

Thanks Rayleigh. I'm not sure yet what the problem is, but it might
be either an issue with shinystan + stanfit objects created from csv files
(this at least used to work pre-2.10).

Can you send me the csv files? Or trimmed csv files (if they're big). I
think I should be able to figure it out pretty quickly if I can reproduce
it exactly.

On Tuesday, July 19, 2016, rayleigh <notifications@github.com
javascript:_e(%7B%7D,'cvml','notifications@github.com');> wrote:

I just tried looking at the Stanfit object in Shinystan right after
getting glmer results. Shinystan launches fine in that case. So, is this an
Rstanarm issue instead?

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#128 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AHb4Q1LEVA4wkENELJgjFEAf_A1KynECks5qXS2YgaJpZM4JQFSa
.

[rayleigh]

One csv file is around 20 MB. What can I trim to make it smaller for you?

[jgabry]

I probably only need a small subset of the posterior draws. So you could
discard everything after the first 50 iterations or something like that.

On Tuesday, July 19, 2016, notifications@github.com wrote:

One csv file is around 20 MB. What can I trim to make it smaller for you?

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
#128 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AHb4Qz3mhTO9QN2befvfHXSFIV-sHrbPks5qXZCngaJpZM4JQFSa
.

[mkiang]

Was this ever resolved? I get the same error when trying to read in a sample file generated by RStan.

[martinmodrak]

The problem persists as of Stan 2.17.0, the problem is when reading fits from CSV files. A minimal reproducible example:

stan_code <- "parameters { real a; } model { a ~ normal(0,1); }"
fit <- stan(model_code = stan_code, sample_file = "samples.csv", chain = 1, iter = 100)
fit2 <- read_stan_csv("samples.csv")

#launch_shinystan(fit) #Works OK
launch_shinystan(fit2)

Results in

Error in if (.used_vb(x) || !(algo %in% c("NUTS", "HMC"))) warning("Many features are only available for models fit using\n            algorithm NUTS or algorithm HMC.",  : 
  missing value where TRUE/FALSE needed

[martinmodrak]

So I filed a pull request with a fix. In the long run it might be more reasonable to unify the CSV output of rstan::sampling and CmdStan to avoid those hacks.

[jgabry]

PR already merged and release is coming so I'm closing this one. Thanks @martinmodrak! I agree that in the long term unifying the output is desirable.