Hey everyone,

for some reason it seems to occur, that the pointwise logLik matrix contains infinte values.

```
LL <- log_lik(fit_S)
> table(is.finite(LL))
FALSE TRUE
51 4479949
> LL[! is.finite(LL)]
[1] -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf
[17] -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf
[33] -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf -Inf
[49] -Inf -Inf -Inf
```

This makes it impossible to compute waic or loo-CV, i.e.:

```
> waic(fit_S)
Error in validate_ll(x) : All input values must be finite.
>
> w <- waic(fit_S, pointwise = TRUE)
Warning message:
NA (NA%) p_waic estimates greater than 0.4. We recommend trying loo instead.
>
> w
Computed from 8000 by 560 log-likelihood matrix
Estimate SE
elpd_waic NaN NA
p_waic NaN NA
waic NaN NA
NA (NA%) p_waic estimates greater than 0.4. We recommend trying loo instead.
>
> table(is.nan(w$pointwise[,3]))
FALSE TRUE
535 25
>
```

The value of -Inf (or even any other extremely small value) seems to be extremly unlikely, if one has a look to the quantiles of the finte values:

```
> summary(as.numeric(LL))
Min. 1st Qu. Median Mean 3rd Qu. Max.
-Inf -1.315e-07 0.000e+00 -Inf 0.000e+00 0.000e+00
> quantile(as.numeric(LL), c(0.005, 0.01, 0.05, 0.1))
0.5% 1% 5% 10%
-1.328816100 -0.775278871 -0.048196403 -0.002251526
```

Therefore, I don’t know how to handle these values.

Do you have any ideas what could be the cause and how to fix it?

I have tried it using both, the current CRAN as well as github versions of brms and loo.

Some more details on my brmsfit:

```
> summary(fit_S)
Family: categorical
Links: mu1 = logit; mu2 = logit
Formula: y ~ 1
mu1 ~ inside + bexotic * exotic1 + bnative * native1 + bdonation * donation1
mu2 ~ inside + bexotic * exotic2 + bnative * native2 + bdonation * donation2
inside ~ 1 + s(t, id, bs = "fs", xt = "ps", m = c(2, 1), k = 5)
bexotic ~ 1 + s(t, id, bs = "fs", xt = "ps", m = c(2, 1), k = 5)
bnative ~ 1 + s(t, id, bs = "fs", xt = "ps", m = c(2, 1), k = 5)
bdonation ~ 1 + s(t, id, bs = "fs", xt = "ps", m = c(2, 1), k = 5)
Data: data (Number of observations: 560)
Draws: 2 chains, each with iter = 6000; warmup = 2000; thin = 1;
total post-warmup draws = 8000
Smooth Terms:
Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
sds(inside_stid_1) 15.22 12.35 0.62 46.22 1.00 3793 3846
sds(inside_stid_2) 120.14 65.75 9.72 260.15 1.00 1899 2033
sds(bexotic_stid_1) 1.19 0.91 0.06 3.47 1.00 2809 3882
sds(bexotic_stid_2) 27.13 9.01 13.02 47.86 1.00 1752 2695
sds(bnative_stid_1) 5.37 3.12 0.44 12.64 1.01 1044 1566
sds(bnative_stid_2) 17.36 11.03 1.38 43.33 1.00 1031 1848
sds(bdonation_stid_1) 3.68 1.85 0.53 7.86 1.00 1292 1367
sds(bdonation_stid_2) 20.29 7.34 8.90 37.19 1.00 1666 2362
Population-Level Effects:
Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
inside_Intercept 10.29 4.63 0.79 18.98 1.00 4581 5921
bexotic_Intercept 4.16 1.46 1.85 7.44 1.00 1751 2609
bnative_Intercept 9.04 2.68 4.55 14.86 1.00 1719 2453
bdonation_Intercept -4.55 1.45 -7.76 -2.19 1.00 1858 2979
Draws were sampled using sampling(NUTS). For each parameter, Bulk_ESS
and Tail_ESS are effective sample size measures, and Rhat is the potential
scale reduction factor on split chains (at convergence, Rhat = 1).
```