Persistent convergence issues with choiceRT_ddm model

[Pritha Sen]

Hi all,

I’m currently working with the choiceRT_ddm model and am encountering persistent convergence issues, specifically, one problematic chain seems to fail to mix properly.

I've tried adjusting priors to narrower priors for all the four parameters (0 to 0.5), increasing the number of iterations and warmups, changing adapt_delta to 0.99, max_treedepth to 15, inits to random, but nothing helps so far.

I've used other models in the hBayesDM package before, including choiceRT_ddm on different datasets, but never have I encountered this issue where only one chain seems to be extremely divergent. I would really appreciate any suggestions or guidance.

I'm attaching the data as well as some of the trace plots for reference.

Thanks in advance for your help — I'm looking forward to any advice you might have!

Best,

Pritha

[Jeongyeon Shin]

Hi Pritha,

Sorry for the delayed response.

I think the issue might be related to the initial values. After looking closely at the trace plots, I noticed that one chain appears to be stuck at unrealistic values and shows almost no variation throughout the sampling.

To check, I ran the model on your data with 100 iterations and 50 warm-up samples, keeping the other settings at their defaults. The model seemed to converge well, and all chains started from similar values.

To address this, you can try setting your own initial values. Please refer to the “2) Fit candidate models” section in the following guide:
https://ccs-lab.github.io/hBayesDM/articles/getting_started.html

Let me know if that doesn’t help or if you have any questions!

Best regards,
Jeongyeon

2025년 4월 25일 금요일 오전 8시 53분 17초 UTC+9에 pritha...@gmail.com님이 작성: