ebf2c8bb8e5f14d371d5052fafd3a0361fffae0b,doc/source/notebooks/advanced/varying_noise.pct.py,,,#,204

Before Change

set_trainable(model.likelihood, False)

opt = gpflow.optimizers.Scipy()
opt.minimize(objective_closure,
             model.trainable_variables,
             options=dict(maxiter=ci_niter(1000)))

// %% [markdown]
// We"ve now fitted the VGP model to the data, but without optimizing over the hyperparameters. Plotting the data, we see that the fit is not terrible, but hasn"t made use of our knowledge of the varying noise. 

After Change

def objective_closure():
    return - model.log_marginal_likelihood()

for _ in range(ci_niter(1000)):
    natgrad.minimize(objective_closure, [(model.q_mu, model.q_sqrt)])

// %% [markdown]
// We"ve now fitted the VGP model to the data, but without optimizing over the hyperparameters. Plotting the data, we see that the fit is not terrible, but hasn"t made use of our knowledge of the varying noise. 

// %%
// let"s do some plotting!
xx = np.linspace(-5, 5, 200)[:, None]

mu, var = model.predict_f(xx)

Instances