Metropolis proposal rejected because location parameter is infinite

Hi! I’m new to Stan, and I’m trying to build a fairly simple model of people’s performances on crosswords. The data is a set of records with a crossword time, which crossword it was, and which user solved the crossword. The model estimates log-seconds by adding a constant, a per-user effect, and a per-crossword effect.

The current Stan code can be found here. (It contains some extra effects but they are minor.)

When I train my model, during the warm-up phase only, I get several messages like these:

Informational Message: The current Metropolis proposal is about to be rejected because of the following issue:
Exception: normal_lpdf: Location parameter[1] is inf, but must be finite!  (in 'unknown file name' at line 57)

If this warning occurs sporadically, such as for highly constrained variable types like covariance matrices, then the sampler is fine,
but if this warning occurs often then your model may be either severely ill-conditioned or misspecified.

Each run sees about ten of these messages, and they don’t appear past the warm-up phase, so my understanding is that they are harmless. I have found that the early warmup samples take much longer than later samples, and I wonder if these error messages are pointing at some underlying problem. Is there some way to modify my model to avoid these warnings and the underlying bad parameter samples?

I wouldn’t worry about them that much if they occur during the warmup phase and the diagnostics are good at the end. But for a hierarchical model such as this, you typically want to start with or at least try a non-centered parameterization

    predictions = mu
                + skill_effect[uids] * skill_dev
                + nth_effect[nth]
                + date_effect[dates] * date_dev
                + sat_effect * to_vector(is_sat);

model {
    // Priors on standardized parameters
    skill_effect ~ std_normal();
    date_effect ~ std_normal();

Ok, thanks, I’ll try that.