Sampling issues with a Wald GLM

I’m attempting to build a wald glm like below but keep getting this error

ValueError: Mass matrix contains zeros on the diagonal.

I cannot seem to know what I’m doing wrong


with pm.Model() as unpooled_glm:

    a = pm.HalfCauchy('a',1, shape = n_vendors)
    b1 = pm.Normal('b1',mu = 0, sd = 1, shape = n_vendors) 
    b2 = pm.Normal('b2', mu = 0, sd = 1, shape = n_vendors)
    
    mu = pm.Deterministic('mu', tt.exp(a[vendor_id] + b1[vendor_id]*d['feet_std'] +  b2[vendor_id]*d['feet_std2']))
    
    
    alpha = pm.Exponential('alpha',1)
    
    time = pm.Wald('duration', mu = mu, lam = lam[vendor_id], alpha = alpha, observed = d.time_mins)
    
    unpooled_glm_trace = pm.sample(2000, tune = 1000, random_seed= seed)

Hi,
This is often related to Bad Initial Energy errors. There is an FAQ about this, which I think will be helpful here :vulcan_salute:

did you ever figure this one out?
I’m hitting similar issues.