How to extend an existing model

Questions
4

That’s a really interesting question! I like using a “functional” pattern, though you could do this with graph rewriting in Theano if you have a really specific need. Here’s an example of my approach:

def hierarchical_normal(name, model):
    with model:
        mean = pm.Normal(f'{name}_mean', 0, 1)
        sd = pm.HalfNormal(f'{name}_sd', 1)
    return model

def make_priors(model):
    model = hierarchical_normal('a', model)
    model = hierarchical_normal('b', model)
    return model

def model_version_one():
    model = pm.Model()
    model = make_priors(model)
    with model:
        c = pm.Normal('c', model['a'] + model['b'], observed=data)
    return model
1 Like