a note on DIC in dockerHDDM

[XiaoyuZeng]

Hi all.
As a newcomer to dockerHDDM, I found it harder to directly and easily extract the DIC of one model.
In the tutorial ipynb linked in your paper, I indeed found code related to extracting DIC from several models:

tmp_dic = []
indx_name = []
for m_key, model in models.items():
    m_tmp = kabuki.utils.concat_models(model)
    tmp_dic.append(m_tmp.dic)
    indx_name.append(m_key)
    
comp_dic = pd.DataFrame(tmp_dic, index=indx_name, columns=['dic'])
comp_dic = comp_dic.sort_values(by=['dic'])
comp_dic = comp_dic.reset_index()
comp_dic.rename(columns={'index':'rank'}, inplace=True)

However, models here refers to the output of the HDDMarviz function that simultaneously fits several models. If someone separately fits several models, they can be confused about the models variable here, just like me : )

m_tmp = kabuki.utils.concat_models(ms3res)
m_tmp.dic

Surely this can be easily achieved by combining data using kabuki and then extracting DIC, but it would be better if a user could easily extract DIC from output data. Or maybe just explicitly clarify how to extract DIC when separately fitting models?

Best,
Xiaoyu

[panwanke]

In new version of docker HDDM, once you have sampled a model m.sample(500, chains = 4), you could directly get the dict by running m.dic, which concat traces in the background.