CODEWITHSUNDEEP
pythonpymc3pymc
bayesbeginner
bayesbeginner

Reputation:

Re-sampling the posterior with a new data point in PyMC

I have some data

X = np.array([1.0, 2.0, 3.0, 4.0, 5.0])
y = np.array([0, 1, 0, 1, 1])

I fit a model

with pm.Model() as model:
    X_shared = pm.Data("X_data", X)
    beta_0 = pm.Normal("beta_0", mu=0, sigma=1)
    beta_1 = pm.Normal("beta_1", mu=0, sigma=1)
    mu = beta_0 + beta_1 * X_shared
    p = pm.math.sigmoid(mu)
    y_obs = pm.Bernoulli("y_obs", p=p, observed=y)
    trace = pm.sample(100)

and now I want to re-sample the posterior based on a new data point

with model:
    pm.set_data({"X_data": [1.69]})
    y_new_simulated = pm.sample_posterior_predictive(trace).posterior_predictive.y_obs
    y_new_obs_name = f"y_new_obs_1"  # Create a unique name for the Bernoulli variable
    y_new_obs = pm.Bernoulli(y_new_obs_name, p=pm.math.sigmoid(beta_0 + beta_1 * 1.69), observed=y_new.posterior_predictive.y_obs)
    new_trace = pm.sample(100)

but I get a broadcasting error:

ValueError: Runtime broadcasting not allowed.

My goal is to try to figure out if the new data point X (and pretending we don't observe y and instead must simulate it) adds information by comparing the old trace and the new trace

Upvotes: 0

Views: 47

Answers (0)

Related Questions