With the updated fits I am now seeing some samples with infinite values which was not an issue previously. I assume this is linked to the update to the truncation window or zero handling.
Pulling a random sample of 50 of these (there are 658 samples from all replicates and scenarios) it looks like the problem is coming from the truncation only model which I suppose makes sense.
This is the posterior after filtering out infinite values which does make it clear that this is coming from the tail of a problematic distribution (we are filtering at 2 here for plotting purposes) vs just being a numeric instability etc.)
I think this must be cause by the addition of left truncation. Given that this is a sub-optimal model for the data perhaps we only need to discuss this as the obvious fix is to also include double censoring (unless I introduced a bug in which case of course we need to resolve)?
With the updated fits I am now seeing some samples with infinite values which was not an issue previously. I assume this is linked to the update to the truncation window or zero handling.
Pulling a random sample of 50 of these (there are 658 samples from all replicates and scenarios) it looks like the problem is coming from the truncation only model which I suppose makes sense.
This is the posterior after filtering out infinite values which does make it clear that this is coming from the tail of a problematic distribution (we are filtering at 2 here for plotting purposes) vs just being a numeric instability etc.)
I think this must be cause by the addition of left truncation. Given that this is a sub-optimal model for the data perhaps we only need to discuss this as the obvious fix is to also include double censoring (unless I introduced a bug in which case of course we need to resolve)?