Model Comparison: How to constrain a model to positive mu-differences?

Questions
8

Thanks for the clarification on the data format. So just have it in long format.

I’m sorry, I still don’t understand what type of variable subject_id in subject_mu_all_trial = subject_mu[subject_id] shall be in your example. Is that the single ID of a participant (e.g. single integer)? But then why have subject_mu with shape n_subjects, if I’d have to loop over all subject_id anyway and call with pm.Model() as model: ... within the loop? Or is subject_id an array of all unique subject IDs [0,1,2,3,…]?
Or is it the first column of the data with duplicates [0,0,0,0,0,…,1,1,1,1,1,…,2,2,2,2,2,2,…]? If the shape of subject_mu_all_trials was the number of rows in the long format dataframe (total number of all trials*2) it would be n_users*n_user_trials*n_directions (e.g. 40x90x2=7200)? Wouldn’t that just give me the values of every single trial? I obviously don’t understand what you are trying to tell me. :confused:

Or did you mean to do it in 2 steps:

df = ... # long format dataframe with column 'user' as ID for participants.
n_users = df['user'].nunique()
mus = list()
with pm.Model() as model:
    user_mu = pm.Normal(..., shape=n_users)

for user_id in df['user'].unique():
    with model:
        user_mu_all_trials = user_mu[user_id]
    mus.append(user_mu)  # Otherwise it would be overwritten each iteration.

That doesn’t seem to be what you intended. Thanks for your effort and patience with me!

1 Like