Initialization fails when using LogNormal

[ricardoV94]

My guess is that Stan is not initializing at the mean but at zero. You can try to set the initvals manually:

@rusty

with pm.Model() as model:
    w_param_0=pm.Lognormal('w_param_0',54.28338623046875,39.43290710449219, initval=1)
    w=pm.Normal('w',w_param_0,10.0, shape=(10,), initval=np.zeros(10))
    b=pm.Normal('b',1.0,10.0)
    obs1=pm.Normal('obs1',aesara.tensor.dot(data['x'],w)+b,1.0, observed=data['y'])