Predict function doesn't work in developmental version about brms

[paul-buerkner]

Thanks for reporting! It is hard to tell why this happens without more information.

How large is your new data? Does this happen for every (subset of) new data. You may try only one or two rows of new data and see if the error still occurs.

Can you also show me the traceback of the error (write traceback() after the error ocurred)?

EDIT: Can you also show me the model as passed to brm that produces the error? In addition, does this happen for all of your models or only for some?

[alireza202]

The new data is pretty small, ~ 100. For now I have reverted to the old version, but I'll check again, and let you know.

This is the model:

data {
int<lower=1> N; # number of observations
real Y[N]; # response variable
int<lower=1> K; # number of fixed effects
matrix[N,K] X; # FE design matrix

data for random effects of subcat

int<lower=1> J_1[N]; # RE levels
int<lower=1> N_1; # number of levels
int<lower=1> K_1; # number of REs
real Z_1[N]; # RE design matrix
}
transformed data {
}
parameters {
vector[K] b; # fixed effects
vector[N_1] pre_1; # unscaled REs
real<lower=0> sd_1; # RE standard deviation
real<lower=0> sigma; # residual SD
}
transformed parameters {
vector[N] eta; # linear predictor
vector[N_1] r_1; # REs

compute linear predictor

eta <- X_b;
r_1 <- sd_1 * (pre_1); # scale REs

if available add REs to linear predictor

for (n in 1:N) {
eta[n] <- eta[n] + Z_1[n]_r_1[J_1[n]];
}
}
model {

prior specifications

sigma ~ cauchy(0,5);
sd_1 ~ cauchy(0,5);
pre_1 ~ normal(0,1);

likelihood contribution

Y ~ normal(eta,sigma);
}

[paul-buerkner]

Thanks for the model code, but I rather ment the function call brm(formula, data, etc.) so that I can replicate your model with dummy data and see if the same error occurs (sorry that my formulation was misleading).

[alireza202]

Sure, here it is:

brm(formula = y ~ x1 + x2 + x3 + (1 | subcat),
data = df,
family = "gaussian",
n.chains = 4,
n.cluster = 4)

[alireza202]

I solved it, but I'm not sure how. It had something to do with my factor levels. When I refactored them again, it worked. Factors in R are much like wildcards... you never know what's gonna happen with them...

[paul-buerkner]

I am glad to hear that, although it would have been interesting to understand what was really causing this issue. ;-)