Error in tmp$r_vals[lr - 1] * tmp$r_vals[lr] :

[hstojic]

Hello,

after fitting a hierarchical model, results seem relatively ok (few divergences but n_eff adn Rhat seem fine), but I get the following error:

Error in tmp$r_vals[lr - 1] * tmp$r_vals[lr] : 
  non-numeric argument to binary operator
In addition: Warning message:
logml could not be estimated within maxiter, rerunning with adjusted starting value. 
Estimate might be more variable than usual. 

This was a test run with 2000 iterations. I know that for reliable model evidence estimation I need a much longer run, so perhaps this error is related to a small sample size? Nontheless, I though its worthwhile reporting this error.

Hrvoje

[singmann]

Hey Hrvoje,

Are you sure you are using the latest CRAN version, 0.7-2?
We have fixed a similar problem some time ago (albeit with a different error message). If you are indeed using the latest version, I fear we need a reproducible example to see what the actual value of tmp$r_vals[lr - 1] and tmp$r_vals[lr] (or probably tmp$r_vals) are to be able to fix it.

Henrik

[hstojic]

Yes, I'm using 0.7-2. Ok, I'll look into producing a reproducible example.

[hstojic]

A quick update.

At the end this error started appearing in 80% of my models. In previous package version it wasn't there so as a temporary workaround, I reverted back to 0.6-2 and there everything is fine, I don't get these errors.

I will submit this paper in about 10 days and then I'll put up the data and the model.

[singmann]

Thanks for this message. Unfortunately, it does sound like we added the bug in our latest fix. That would not be too good.

[RemiDav]

I ran into the same issue and I do not know what is causing it.
I have two stan_glmer regression differing only by the value of one dummy variable (the models are identical, only the data changes).
One regression output works well in bridge_sampler, the other doesn't and result in the error.
I tried both 0.7-2 and 0.6-0.

Packages loaded:

other attached packages:
[1] bridgesampling_0.6-0 rstan_2.19.2         ggplot2_3.2.1        StanHeaders_2.19.0   rstanarm_2.19.2     
[6] Rcpp_1.0.2 

[RemiDav]

I changed the seed of the regression that caused me trouble and it fixed the issue.

[Mathieu-Chevalier]

Hello,
I have the exact same issue as above:
Error in tmp$r_vals[lr - 1] * tmp$r_vals[lr] : non-numeric argument to binary operator In addition: Warning message: logml could not be estimated within maxiter, rerunning with adjusted starting value. Estimate might be more variable than usual.
This error appears with version 0.7.2 of the package. With the version 0.6.0, everything is fine.
Anyway to work around this error message?
Thanks in advance.
Best,
Mathieu

[singmann]

Should be solved now. We will update on CRAN asap.