def test_inference_deepGP():
gp1 = GPRegression(
X, None,
RBF(input_dim=3,
variance=torch.tensor(3.),
lengthscale=torch.tensor(2.)))
Z, _ = gp1.model()
gp2 = VariationalSparseGP(Z, y2D, Matern32(input_dim=3), Z.clone(),
Gaussian(torch.tensor(1e-6)))
class DeepGP(torch.nn.Module):
def __init__(self, gp1, gp2):
super(DeepGP, self).__init__()
self.gp1 = gp1
self.gp2 = gp2
def model(self):
Z, _ = self.gp1.model()
self.gp2.set_data(Z, y2D)
self.gp2.model()
def guide(self):
self.gp1.guide()
self.gp2.guide()
deepgp = DeepGP(gp1, gp2)
train(deepgp, num_steps=1)
def test_mean_function_VGP_whiten():
X, y, Xnew, ynew, kernel, mean_fn = _pre_test_mean_function()
likelihood = Gaussian()
gpmodule = VariationalGP(X, y, kernel, likelihood, mean_function=mean_fn,
whiten=True)
optimizer = torch.optim.Adam(gpmodule.parameters(), lr=0.1)
train(gpmodule, optimizer)
_post_test_mean_function(gpmodule, Xnew, ynew)
def test_mean_function_VSGP():
X, y, Xnew, ynew, kernel, mean_fn = _pre_test_mean_function()
Xu = X[::20].clone()
likelihood = Gaussian()
gpmodule = VariationalSparseGP(X, y, kernel, Xu, likelihood, mean_function=mean_fn)
optimizer = torch.optim.Adam(gpmodule.parameters(), lr=0.02)
train(gpmodule, optimizer)
_post_test_mean_function(gpmodule, Xnew, ynew)
def test_inference_with_whiten(model_class, X, y, kernel, likelihood):
# regression models don't use whiten
if model_class is GPRegression or model_class is SparseGPRegression:
return
elif model_class is VariationalGP:
gp = model_class(X, y, kernel, likelihood, whiten=True)
else: # model_class is SparseVariationalGP
gp = model_class(X, y, kernel, X.clone(), likelihood, whiten=True)
train(gp, num_steps=1)
def test_inference_with_empty_latent_shape(model_class, X, y, kernel, likelihood):
# regression models don't use latent_shape (default=torch.Size([]))
if model_class is GPRegression or model_class is SparseGPRegression:
return
elif model_class is VariationalGP:
gp = model_class(X, y, kernel, likelihood, latent_shape=torch.Size([]))
else: # model_class is SparseVariationalGP
gp = model_class(X, y, kernel, X.clone(), likelihood, latent_shape=torch.Size([]))
train(gp, num_steps=1)
def test_mean_function_SGPR_FITC():
X, y, Xnew, ynew, kernel, mean_fn = _pre_test_mean_function()
Xu = X[::20].clone()
gpmodule = SparseGPRegression(X,
y,
kernel,
Xu,
mean_function=mean_fn,
approx="FITC")
train(gpmodule)
_post_test_mean_function(gpmodule, Xnew, ynew)
def test_gplvm(model_class, X, y, kernel, likelihood):
if model_class is SparseGPRegression or model_class is VariationalSparseGP:
gp = model_class(X, y, kernel, X.clone(), likelihood)
else:
gp = model_class(X, y, kernel, likelihood)
gplvm = GPLVM(gp)
# test inference
train(gplvm, num_steps=1)
# test forward
gplvm(Xnew=X)
def test_inference_with_empty_latent_shape(model_class, X, y, kernel, likelihood):
if isinstance(likelihood, MultiClass):
latent_shape = torch.Size([likelihood.num_classes])
else:
latent_shape = torch.Size([])
if model_class is VariationalSparseGP:
gp = model_class(X, y, kernel, X, likelihood, latent_shape=latent_shape)
else:
gp = model_class(X, y, kernel, likelihood, latent_shape=latent_shape)
train(gp, num_steps=1)
def test_inference(model_class, X, y, kernel, likelihood):
if isinstance(likelihood, MultiClass):
latent_shape = y.shape[:-1] + (likelihood.num_classes,)
else:
latent_shape = y.shape[:-1]
if model_class is VariationalSparseGP:
gp = model_class(X, y, kernel, X, likelihood, latent_shape=latent_shape)
else:
gp = model_class(X, y, kernel, likelihood, latent_shape=latent_shape)
train(gp, num_steps=1)
def test_inference_whiten_vsgp():
N = 1000
X = dist.Uniform(torch.zeros(N), torch.ones(N)*5).sample()
y = 0.5 * torch.sin(3*X) + dist.Normal(torch.zeros(N), torch.ones(N)*0.5).sample()
kernel = RBF(input_dim=1)
Xu = torch.arange(0., 5.5, 0.5)
vsgp = VariationalSparseGP(X, y, kernel, Xu, Gaussian(), whiten=True)
train(vsgp)
Xnew = torch.arange(0., 5.05, 0.05)
loc, var = vsgp(Xnew, full_cov=False)
target = 0.5 * torch.sin(3*Xnew)
assert_equal((loc - target).abs().mean().item(), 0, prec=0.07)
def test_inference_sgpr():
N = 1000
X = dist.Uniform(torch.zeros(N), torch.ones(N)*5).sample()
y = 0.5 * torch.sin(3*X) + dist.Normal(torch.zeros(N), torch.ones(N)*0.5).sample()
kernel = RBF(input_dim=1)
Xu = torch.arange(0., 5.5, 0.5)
sgpr = SparseGPRegression(X, y, kernel, Xu)
train(sgpr)
Xnew = torch.arange(0., 5.05, 0.05)
loc, var = sgpr(Xnew, full_cov=False)
target = 0.5 * torch.sin(3*Xnew)
assert_equal((loc - target).abs().mean().item(), 0, prec=0.07)
def test_inference(model_class, X, y, kernel, likelihood):
# skip variational GP models because variance/lengthscale highly
# depend on variational parameters
if model_class is VariationalGP or model_class is VariationalSparseGP:
return
elif model_class is GPRegression:
gp = model_class(X, y, RBF(input_dim=3), likelihood)
else: # model_class is SparseGPRegression
gp = model_class(X, y, RBF(input_dim=3), X, likelihood)
# fix inducing points because variance/lengthscale highly depend on it
gp.Xu.requires_grad_(False)
generator = dist.MultivariateNormal(torch.zeros(X.shape[0]), kernel(X))
target_y = generator(sample_shape=torch.Size([1000])).detach()
gp.set_data(X, target_y)
train(gp)
y_cov = gp.kernel(X)
target_y_cov = kernel(X)
assert_equal(y_cov, target_y_cov, prec=0.1)
def test_mean_function_VGP():
X, y, Xnew, ynew, kernel, mean_fn = _pre_test_mean_function()
likelihood = Gaussian()
gpmodule = VariationalGP(X, y, kernel, likelihood, mean_function=mean_fn)
train(gpmodule)
_post_test_mean_function(gpmodule, Xnew, ynew)
def test_mean_function_GPR():
X, y, Xnew, ynew, kernel, mean_fn = _pre_test_mean_function()
gpmodule = GPRegression(X, y, kernel, mean_function=mean_fn)
train(gpmodule)
_post_test_mean_function(gpmodule, Xnew, ynew)