def test_prior_mll():
"""
Test that the MLL evaluation works with priors attached to the parameter values.
"""
key = jr.PRNGKey(123)
x = jnp.sort(jr.uniform(key, minval=-5.0, maxval=5.0, shape=(100, 1)),
axis=0)
f = lambda x: jnp.sin(jnp.pi * x) / (jnp.pi * x)
y = f(x) + jr.normal(key, shape=x.shape) * 0.1
posterior = Prior(kernel=RBF()) * Gaussian()
params = initialise(posterior)
config = get_defaults()
constrainer, unconstrainer = build_all_transforms(params.keys(), config)
params = unconstrainer(params)
print(params)
mll = marginal_ll(posterior, transform=constrainer)
priors = {
"lengthscale": tfd.Gamma(1.0, 1.0),
"variance": tfd.Gamma(2.0, 2.0),
"obs_noise": tfd.Gamma(2.0, 2.0),
}
mll_eval = mll(params, x, y)
mll_eval_priors = mll(params, x, y, priors)
assert pytest.approx(mll_eval) == jnp.array(-103.28180663)
assert pytest.approx(mll_eval_priors) == jnp.array(-105.509218857)
def test_non_conjugate():
posterior = Prior(kernel=RBF()) * Bernoulli()
n = 20
x = jnp.linspace(-1.0, 1.0, n).reshape(-1, 1)
y = jnp.sin(x)
params = initialise(posterior, 20)
config = get_defaults()
unconstrainer, constrainer = build_all_transforms(params.keys(), config)
params = unconstrainer(params)
mll = marginal_ll(posterior, transform=constrainer)
assert isinstance(mll, Callable)
neg_mll = marginal_ll(posterior, transform=constrainer, negative=True)
assert neg_mll(params, x, y) == jnp.array(-1.0) * mll(params, x, y)
def test_conjugate():
posterior = Prior(kernel=RBF()) * Gaussian()
x = jnp.linspace(-1.0, 1.0, 20).reshape(-1, 1)
y = jnp.sin(x)
D = Dataset(X=x, y=y)
params = initialise(posterior)
config = get_defaults()
unconstrainer, constrainer = build_all_transforms(params.keys(), config)
params = unconstrainer(params)
mll = marginal_ll(posterior, transform=constrainer)
assert isinstance(mll, Callable)
neg_mll = marginal_ll(posterior, transform=constrainer, negative=True)
assert neg_mll(params, D) == jnp.array(-1.0) * mll(params, D)
def test_conjugate():
key = jr.PRNGKey(123)
kern = to_spectral(RBF(), 10)
posterior = Prior(kernel=kern) * Gaussian()
x = jnp.linspace(-1.0, 1.0, 20).reshape(-1, 1)
y = jnp.sin(x)
params = initialise(key, posterior)
config = get_defaults()
unconstrainer, constrainer = build_all_transforms(params.keys(), config)
params = unconstrainer(params)
mll = marginal_ll(posterior, transform=constrainer)
assert isinstance(mll, Callable)
neg_mll = marginal_ll(posterior, transform=constrainer, negative=True)
assert neg_mll(params, x, y) == jnp.array(-1.0) * mll(params, x, y)
nmll = neg_mll(params, x, y)
assert nmll.shape == ()
def test_build_all_transforms(likelihood):
posterior = Prior(kernel=RBF()) * likelihood()
params = initialise(posterior, 10)
config = get_defaults()
t1, t2 = build_all_transforms(params.keys(), config)
constrainer = build_constrain(params.keys(), config)
constrained = t1(params)
constrained2 = constrainer(params)
assert constrained2.keys() == constrained2.keys()
for u, v in zip(constrained.values(), constrained2.values()):
assert_array_equal(u, v)
assert u.dtype == v.dtype
unconstrained = t2(params)
unconstrainer = build_unconstrain(params.keys(), config)
unconstrained2 = unconstrainer(params)
for u, v in zip(unconstrained.values(), unconstrained2.values()):
assert_array_equal(u, v)
assert u.dtype == v.dtype
def fit(posterior, nits, data, configs):
params = initialise(posterior)
constrainer, unconstrainer = build_all_transforms(params.keys(), configs)
mll = jit(marginal_ll(posterior, transform=constrainer, negative=True))
opt_init, opt_update, get_params = optimizers.adam(step_size=0.05)
opt_state = opt_init(params)
def step(i, opt_state):
p = get_params(opt_state)
v, g = value_and_grad(mll)(p, data)
return opt_update(i, g, opt_state), v
for i in range(nits):
opt_state, mll_estimate = step(i, opt_state)
print(f"{posterior.prior.kernel.name} GP's marginal log-likelihood: {mll_estimate: .2f}")
final_params = constrainer(get_params(opt_state))
return final_params
