@pymc.deterministic

Questions v5
1

Is @pymc.deterministic function header deprecated under V4? If so, how can I create a user-defined-function that takes in variables drawn from a distribution?

2

You can just pass them into functions like regular variables. Picking an example notebook at random, you can take this model

coords = {"observation": data.index.values}

with pm.Model(coords=coords) as binomial_regression_model:
    x = pm.ConstantData("x", data["x"], dims="observation")
    # priors
    beta0 = pm.Normal("beta0", mu=0, sigma=1)
    beta1 = pm.Normal("beta1", mu=0, sigma=1)
    # linear model
    mu = beta0 + beta1 * x
    p = pm.Deterministic("p", pm.math.invlogit(mu), dims="observation")
    # likelihood
    pm.Binomial("y", n=n, p=p, observed=data["y"], dims="observation")

and do this instead

def my_func(beta0, beta1, x):
    return beta0 + beta1 * x

with pm.Model(coords=coords) as binomial_regression_model:
    x = pm.ConstantData("x", data["x"], dims="observation")
    # priors
    beta0 = pm.Normal("beta0", mu=0, sigma=1)
    beta1 = pm.Normal("beta1", mu=0, sigma=1)
    # linear model
    mu = my_func(beta0, beta1, x)
    p = pm.Deterministic("p", pm.math.invlogit(mu), dims="observation")
    # likelihood
    pm.Binomial("y", n=n, p=p, observed=data["y"], dims="observation")
2 Likes