#19 Better Error Message for "Slicer Stuck ..."

open
nobody
None
1
2014-03-22
2014-03-21
Russell Almond
No

I'm in the process of debugging a model which is generating error messages "Slicer Stuck at value with infinite density." I think I've figured out what is wrong, and I've adjusted my model, but the problem is occurring in the adaptation phase, before I'm monitoring, so I don't know what the value of the parameter is when the chain gets stuck. It would be a big help if the error message included the value of the stuck node in addition to which node is stuck, as it might make the issue with the model more transparent.

Discussion

  • Russell Almond
    Russell Almond
    2014-03-22

    Thinking about this some more, I not only need the value of the node, but also the value of the nodes in the Markov blanket. This may be fairly large, for examle in the case of the variance parameter of a hierarchical model with a large number of units. In this case, I probably only need a sample of them (e.g., the first 10 units in the array). Trying to push that into a general rule is a bit tricky though.

     


Anonymous


Cancel   Add attachments