acope3
commented
10 months ago I think both versions are broken, but for different reasons.
CRAN version: I ran the tests from the version that should be on CRAN and it passed. We have a test for both with and without $\Phi$ (observed gene expression data) runs but it seg faulted when I applied it to my real data. The difference between the test and real datasets is the latter had 3 measurements while the former only had 1. The code is intended to allow for multiple measures of gene expression, but it is only formally being tested for 1. We can update the test with little issue. @mikegilchrist does your dataset have more than 1 measurement of gene expression and if so, what happens if you reduce it to a single dataset?
More recent version: As we expanded the models used in AnaCoDa, we needed to expand the functionality of the code to allow for observed estimates of gene expression for other models. Originally, this code was ROC-specific. Ideally, these functions would be part of the Model class (the parent class) as they are general to all models. Moving this code to the Model class appears to be the issue. If I keep the code in the ROCModel class, things seem to work fine based on some implementation tests. I think this has to do with the parent Model object being unable to "see" the derived class Parameter object. I Will need to do some more digging to figure out exactly why this is. In the meantime, a simple functional solution is to just have repetitive functions for each of the Model classes (ROC, FONSE, etc.), which was how the code originally was. This kinda flies in the face of the idea of object-oriented programming and polymorphism, but might be an okay temporary solution for our next submission to CRAN.
mikegilchrist
Both the CRAN and more recent versions of the code seg fault when gene expression data is included with the fit.