BestDose : posterior probability issue

[R-garreau]

I'm oppening this issue for now.
I'm translating at LLM pace all the BestDose code, and it seems like posterior value are still 100% or 0%
I'll ding in deeper once everything is translated to see if it is a bestdose problem or a pmcore.

Good news is that there is no more in between layer : it either PMcore or BestDose (see img)

[R-garreau]

i Think it works in BestDose now.

The problem seems to arise from Burke algorithm that discard the weight after running.
The work around is back in Bestdose maybe we should leave it there for now ?

[mhovd]

Based on the plots, you seem to have at least 3 active support points.

Are you testing with a simulated patient whose parameter values are exactly equal to a support point? What kind of error function are you using?

[R-garreau]

@mhovd no i did test with several real patients. But we could use known data if you want

[mhovd]

Then they seem to be well represented by your parameter distribution. I don't see any immediate issues, but I am working on a way to both diagnose and infer from these "degenerate" posterior distributions.

[R-garreau]

At the moment there is no obvious apparent issues.
The application is almost ready for beta testing (i have to work on the report display, and minor issue)
I will try to see if they can test it in Lyon to have some feedback