Best practices for MAP estimation

Questions
4

Solved it myself, switched to a beta likelihood function and it’s all working nicely. Thanks for your guidance @junpenglao!!

data = np.array([0.05, 0.1, 0.1, 0.15, 0.2, 0.1])

with pm.Model():
    const = pm.Normal("const", mu=0, sd=10)
    mu = sigmoid(const)
    sd = mu * (1 - mu)  # bernoulli SD per pymc3 docs
    pred = pm.Beta("pred", mu=mu, sd=sd, observed=data)

    map_estimate, optresult = pm.find_MAP(
        progressbar=False, method="L-BFGS-B", return_raw=True
    )

    print(optresult)

    map_estimate = float(map_estimate["const"])
1 Like