merged initial DDM distribution #20
Conversation
Codecov ReportPatch coverage has no change and project coverage change:
Additional details and impacted files@@ Coverage Diff @@
## master #20 +/- ##
===========================================
- Coverage 75.70% 54.51% -21.20%
===========================================
Files 10 10
Lines 391 543 +152
===========================================
Hits 296 296
- Misses 95 247 +152
☔ View full report in Codecov by Sentry. |
|
@kiante-fernandez Thank you for incorporating the code into SequentialSamplingModels.jl. I'm very much in favor of the negative-RT parameterisation, but could you elaborate on how this helped in resolving issues with Turing.jl compatibility? |
|
Thanks Kiante for the PR, and thanks t-alfers for allowing us to integrate your code. I was struggling to view the PR locally, but I am going to merge anyways. For future reference, do you know of a solution? I will add some tests and release a new version sometime today. |
I have merged the DDM implementation from the https://github.com/t-alfers/WienerDiffusionModel.jl repo (thanks @t-alfers). I chose to use the response time implementation that utilizes the negative-RT solution. The negative RTs indicate a lower-bound response. This approach is more straightforward and, importantly, significantly reduces the probability of coding the model incorrectly. It follows the original JAGS implementation (see Vandekerckhove 2014) where the lower-bound hits as negative reaction times were used as a cost-effective way of encoding what essentially is a bivariate distribution (strictly positive RT and binary response). Refer to this old thread for discussion. It may also help resolve some issues we were experiencing with Turing.jl compatibility.