Skip to content

Commit 78ef92a

Browse files
esantorellafacebook-github-bot
esantorella
authored and
facebook-github-bot
committed
Remove indent that existed only to make review of previous diff easier (#1523)
Summary: Pull Request resolved: #1523 Remove indent that existed only to make review of previous diff easier Reviewed By: Balandat Differential Revision: D41509531 fbshipit-source-id: 918b73576a5d6b1729c9ea044e86745f48eb4850
1 parent 6137f18 commit 78ef92a

File tree

1 file changed

+101
-107
lines changed

1 file changed

+101
-107
lines changed

‎test/models/test_model_list_gp_regression.py‎

Lines changed: 101 additions & 107 deletions
Original file line numberDiff line numberDiff line change
@@ -75,120 +75,114 @@ class TestModelListGP(BotorchTestCase):
7575
def _base_test_ModelListGP(
7676
self, fixed_noise: bool, dtype, use_octf: bool
7777
) -> ModelListGP:
78-
# this is to make review easier -- will be removed in the next
79-
# commit in the stack and never landed
80-
unneccessary_condition_for_indentation_remove_me = True
81-
if unneccessary_condition_for_indentation_remove_me:
82-
tkwargs = {"device": self.device, "dtype": dtype}
83-
model = _get_model(fixed_noise=fixed_noise, use_octf=use_octf, **tkwargs)
84-
self.assertIsInstance(model, ModelListGP)
85-
self.assertIsInstance(model.likelihood, LikelihoodList)
86-
for m in model.models:
87-
self.assertIsInstance(m.mean_module, ConstantMean)
88-
self.assertIsInstance(m.covar_module, ScaleKernel)
89-
matern_kernel = m.covar_module.base_kernel
90-
self.assertIsInstance(matern_kernel, MaternKernel)
91-
self.assertIsInstance(matern_kernel.lengthscale_prior, GammaPrior)
92-
if use_octf:
93-
self.assertIsInstance(m.outcome_transform, Standardize)
94-
95-
# test constructing likelihood wrapper
96-
mll = SumMarginalLogLikelihood(model.likelihood, model)
97-
for mll_ in mll.mlls:
98-
self.assertIsInstance(mll_, ExactMarginalLogLikelihood)
99-
100-
# test model fitting (sequential)
101-
with warnings.catch_warnings():
102-
warnings.filterwarnings("ignore", category=OptimizationWarning)
103-
mll = fit_gpytorch_mll(
104-
mll, optimizer_kwargs={"options": {"maxiter": 1}}, max_attempts=1
105-
)
106-
with warnings.catch_warnings():
107-
warnings.filterwarnings("ignore", category=OptimizationWarning)
108-
# test model fitting (joint)
109-
mll = fit_gpytorch_mll(
110-
mll,
111-
optimizer_kwargs={"options": {"maxiter": 1}},
112-
max_attempts=1,
113-
sequential=False,
114-
)
115-
116-
# test subset outputs
117-
subset_model = model.subset_output([1])
118-
self.assertIsInstance(subset_model, ModelListGP)
119-
self.assertEqual(len(subset_model.models), 1)
120-
sd_subset = subset_model.models[0].state_dict()
121-
sd = model.models[1].state_dict()
122-
self.assertTrue(set(sd_subset.keys()) == set(sd.keys()))
123-
self.assertTrue(all(torch.equal(v, sd[k]) for k, v in sd_subset.items()))
124-
125-
# test posterior
126-
test_x = torch.tensor([[0.25], [0.75]], **tkwargs)
127-
posterior = model.posterior(test_x)
128-
self.assertIsInstance(posterior, GPyTorchPosterior)
129-
self.assertIsInstance(posterior.distribution, MultitaskMultivariateNormal)
78+
tkwargs = {"device": self.device, "dtype": dtype}
79+
model = _get_model(fixed_noise=fixed_noise, use_octf=use_octf, **tkwargs)
80+
self.assertIsInstance(model, ModelListGP)
81+
self.assertIsInstance(model.likelihood, LikelihoodList)
82+
for m in model.models:
83+
self.assertIsInstance(m.mean_module, ConstantMean)
84+
self.assertIsInstance(m.covar_module, ScaleKernel)
85+
matern_kernel = m.covar_module.base_kernel
86+
self.assertIsInstance(matern_kernel, MaternKernel)
87+
self.assertIsInstance(matern_kernel.lengthscale_prior, GammaPrior)
13088
if use_octf:
131-
# ensure un-transformation is applied
132-
submodel = model.models[0]
133-
p0 = submodel.posterior(test_x)
134-
tmp_tf = submodel.outcome_transform
135-
del submodel.outcome_transform
136-
p0_tf = submodel.posterior(test_x)
137-
submodel.outcome_transform = tmp_tf
138-
expected_var = tmp_tf.untransform_posterior(p0_tf).variance
139-
self.assertTrue(torch.allclose(p0.variance, expected_var))
140-
141-
# test output_indices
142-
posterior = model.posterior(
143-
test_x, output_indices=[0], observation_noise=True
89+
self.assertIsInstance(m.outcome_transform, Standardize)
90+
91+
# test constructing likelihood wrapper
92+
mll = SumMarginalLogLikelihood(model.likelihood, model)
93+
for mll_ in mll.mlls:
94+
self.assertIsInstance(mll_, ExactMarginalLogLikelihood)
95+
96+
# test model fitting (sequential)
97+
with warnings.catch_warnings():
98+
warnings.filterwarnings("ignore", category=OptimizationWarning)
99+
mll = fit_gpytorch_mll(
100+
mll, optimizer_kwargs={"options": {"maxiter": 1}}, max_attempts=1
101+
)
102+
with warnings.catch_warnings():
103+
warnings.filterwarnings("ignore", category=OptimizationWarning)
104+
# test model fitting (joint)
105+
mll = fit_gpytorch_mll(
106+
mll,
107+
optimizer_kwargs={"options": {"maxiter": 1}},
108+
max_attempts=1,
109+
sequential=False,
144110
)
145-
self.assertIsInstance(posterior, GPyTorchPosterior)
146-
self.assertIsInstance(posterior.distribution, MultivariateNormal)
147111

148-
# test condition_on_observations
149-
f_x = [torch.rand(2, 1, **tkwargs) for _ in range(2)]
150-
f_y = torch.rand(2, 2, **tkwargs)
151-
if fixed_noise:
152-
noise = 0.1 + 0.1 * torch.rand_like(f_y)
153-
cond_kwargs = {"noise": noise}
154-
else:
155-
cond_kwargs = {}
156-
cm = model.condition_on_observations(f_x, f_y, **cond_kwargs)
157-
self.assertIsInstance(cm, ModelListGP)
158-
159-
# test condition_on_observations batched
160-
f_x = [torch.rand(3, 2, 1, **tkwargs) for _ in range(2)]
161-
f_y = torch.rand(3, 2, 2, **tkwargs)
162-
cm = model.condition_on_observations(f_x, f_y, **cond_kwargs)
163-
self.assertIsInstance(cm, ModelListGP)
164-
165-
# test condition_on_observations batched (fast fantasies)
166-
f_x = [torch.rand(2, 1, **tkwargs) for _ in range(2)]
167-
f_y = torch.rand(3, 2, 2, **tkwargs)
168-
cm = model.condition_on_observations(f_x, f_y, **cond_kwargs)
169-
self.assertIsInstance(cm, ModelListGP)
112+
# test subset outputs
113+
subset_model = model.subset_output([1])
114+
self.assertIsInstance(subset_model, ModelListGP)
115+
self.assertEqual(len(subset_model.models), 1)
116+
sd_subset = subset_model.models[0].state_dict()
117+
sd = model.models[1].state_dict()
118+
self.assertTrue(set(sd_subset.keys()) == set(sd.keys()))
119+
self.assertTrue(all(torch.equal(v, sd[k]) for k, v in sd_subset.items()))
170120

171-
# test condition_on_observations (incorrect input shape error)
172-
with self.assertRaises(BotorchTensorDimensionError):
173-
model.condition_on_observations(
174-
f_x, torch.rand(3, 2, 3, **tkwargs), **cond_kwargs
175-
)
121+
# test posterior
122+
test_x = torch.tensor([[0.25], [0.75]], **tkwargs)
123+
posterior = model.posterior(test_x)
124+
self.assertIsInstance(posterior, GPyTorchPosterior)
125+
self.assertIsInstance(posterior.distribution, MultitaskMultivariateNormal)
126+
if use_octf:
127+
# ensure un-transformation is applied
128+
submodel = model.models[0]
129+
p0 = submodel.posterior(test_x)
130+
tmp_tf = submodel.outcome_transform
131+
del submodel.outcome_transform
132+
p0_tf = submodel.posterior(test_x)
133+
submodel.outcome_transform = tmp_tf
134+
expected_var = tmp_tf.untransform_posterior(p0_tf).variance
135+
self.assertTrue(torch.allclose(p0.variance, expected_var))
136+
137+
# test output_indices
138+
posterior = model.posterior(test_x, output_indices=[0], observation_noise=True)
139+
self.assertIsInstance(posterior, GPyTorchPosterior)
140+
self.assertIsInstance(posterior.distribution, MultivariateNormal)
176141

177-
# test X having wrong size
178-
with self.assertRaises(AssertionError):
179-
model.condition_on_observations(f_x[:1], f_y)
142+
# test condition_on_observations
143+
f_x = [torch.rand(2, 1, **tkwargs) for _ in range(2)]
144+
f_y = torch.rand(2, 2, **tkwargs)
145+
if fixed_noise:
146+
noise = 0.1 + 0.1 * torch.rand_like(f_y)
147+
cond_kwargs = {"noise": noise}
148+
else:
149+
cond_kwargs = {}
150+
cm = model.condition_on_observations(f_x, f_y, **cond_kwargs)
151+
self.assertIsInstance(cm, ModelListGP)
152+
153+
# test condition_on_observations batched
154+
f_x = [torch.rand(3, 2, 1, **tkwargs) for _ in range(2)]
155+
f_y = torch.rand(3, 2, 2, **tkwargs)
156+
cm = model.condition_on_observations(f_x, f_y, **cond_kwargs)
157+
self.assertIsInstance(cm, ModelListGP)
158+
159+
# test condition_on_observations batched (fast fantasies)
160+
f_x = [torch.rand(2, 1, **tkwargs) for _ in range(2)]
161+
f_y = torch.rand(3, 2, 2, **tkwargs)
162+
cm = model.condition_on_observations(f_x, f_y, **cond_kwargs)
163+
self.assertIsInstance(cm, ModelListGP)
164+
165+
# test condition_on_observations (incorrect input shape error)
166+
with self.assertRaises(BotorchTensorDimensionError):
167+
model.condition_on_observations(
168+
f_x, torch.rand(3, 2, 3, **tkwargs), **cond_kwargs
169+
)
180170

181-
# test posterior transform
182-
X = torch.rand(3, 1, **tkwargs)
183-
weights = torch.tensor([1, 2], **tkwargs)
184-
post_tf = ScalarizedPosteriorTransform(weights=weights)
185-
posterior_tf = model.posterior(X, posterior_transform=post_tf)
186-
self.assertTrue(
187-
torch.allclose(
188-
posterior_tf.mean,
189-
model.posterior(X).mean @ weights.unsqueeze(-1),
190-
)
171+
# test X having wrong size
172+
with self.assertRaises(AssertionError):
173+
model.condition_on_observations(f_x[:1], f_y)
174+
175+
# test posterior transform
176+
X = torch.rand(3, 1, **tkwargs)
177+
weights = torch.tensor([1, 2], **tkwargs)
178+
post_tf = ScalarizedPosteriorTransform(weights=weights)
179+
posterior_tf = model.posterior(X, posterior_transform=post_tf)
180+
self.assertTrue(
181+
torch.allclose(
182+
posterior_tf.mean,
183+
model.posterior(X).mean @ weights.unsqueeze(-1),
191184
)
185+
)
192186

193187
return model
194188

0 commit comments

Comments
 (0)