def test_kernel_learning_NN(self):
for learning_mode in ['learn_kernel', 'both']:
gpr_model_vanilla = GPRegressionLearned(
self.x_train,
self.y_train_sin,
learning_mode='learn_kernel',
num_iter_fit=1,
mean_module='zero',
covar_module='NN')
gpr_model_vanilla.fit()
gpr_model_learn_kernel = GPRegressionLearned(
self.x_train,
self.y_train_sin,
learning_mode=learning_mode,
num_iter_fit=500,
mean_module='constant',
covar_module='NN',
kernel_nn_layers=(16, 16),
mean_nn_layers=(16, 16))
gpr_model_learn_kernel.fit(valid_x=self.x_train,
valid_t=self.y_train_sin)
ll_vanilla, rmse_vanilla, _ = gpr_model_vanilla.eval(
self.x_train, self.y_train_sin)
ll_kernel, rmse_kernel, _ = gpr_model_learn_kernel.eval(
self.x_train, self.y_train_sin)
print('learning_mode', learning_mode)
print(ll_kernel, ll_vanilla)
print(rmse_kernel, rmse_vanilla)
self.assertGreater(ll_kernel, ll_vanilla)
self.assertLess(rmse_kernel, rmse_vanilla)
def test_mean_learning(self):
for mean_module in ['NN']:
gpr_model_vanilla = GPRegressionLearned(self.x_train,
self.y_train_sin,
learning_mode='vanilla',
num_iter_fit=20,
mean_module='constant',
covar_module='SE')
gpr_model_vanilla.fit()
gpr_model_learn_mean = GPRegressionLearned(
self.x_train,
self.y_train_sin,
learning_mode='learn_mean',
num_iter_fit=100,
mean_module=mean_module,
covar_module='SE',
mean_nn_layers=(16, 16))
gpr_model_learn_mean.fit()
ll_vanilla, rmse_vanilla, _ = gpr_model_vanilla.eval(
self.x_train, self.y_train_two)
ll_mean, rmse_mean, _ = gpr_model_learn_mean.eval(
self.x_train, self.y_train_sin)
print(ll_mean, ll_vanilla)
print(rmse_mean, rmse_vanilla)
self.assertGreater(ll_mean, ll_vanilla)
self.assertLess(rmse_mean, rmse_vanilla)
def fit_eval(x_context, y_context, x_test, y_test, params):
from meta_learn.GPR_mll import GPRegressionLearned
torch.set_num_threads(1)
model = GPRegressionLearned(x_context,
y_context,
**params,
random_seed=seed)
model.fit(verbose=False)
return model.eval(x_test, y_test)
def fit_eval_gpr(x_context, t_context, x_test, t_test):
gpr = GPRegressionLearned(x_context,
t_context,
learning_mode='both',
mean_nn_layers=(64, 64),
covar_module='SE',
mean_module='NN',
weight_decay=0.0,
num_iter_fit=num_iter_fit)
gpr.fit(valid_x=x_test, valid_t=t_test)
return gpr.eval(x_test, t_test)[0]
def test_random_seed_consistency(self):
gpr_model_1 = GPRegressionLearned(self.x_train,
self.y_train_two,
learning_mode='both',
num_iter_fit=5,
mean_module='NN',
covar_module='NN',
random_seed=22)
gpr_model_2 = GPRegressionLearned(self.x_train,
self.y_train_two,
learning_mode='both',
num_iter_fit=5,
mean_module='NN',
covar_module='NN',
random_seed=22)
gpr_model_1.fit()
t_predict_1 = gpr_model_1.predict(self.x_test)
gpr_model_2.fit()
t_predict_2 = gpr_model_2.predict(self.x_test)
self.assertTrue(np.array_equal(t_predict_1, t_predict_2))
def test_kernel_learning_COS(self):
for learning_mode in ['learn_kernel', 'both']:
gpr_model_vanilla = GPRegressionLearned(
self.x_train,
self.y_train_sin,
learning_mode='vanilla',
num_iter_fit=1,
mean_module='constant',
covar_module=CosineKernel())
gpr_model_vanilla.fit()
gpr_model_learn_kernel = GPRegressionLearned(
self.x_train,
self.y_train_sin,
learning_mode='learn_kernel',
num_iter_fit=500,
mean_module='constant',
covar_module=CosineKernel())
print(gpr_model_learn_kernel.model.covar_module.lengthscale)
gpr_model_learn_kernel.fit(valid_x=self.x_train,
valid_t=self.y_train_sin)
print(gpr_model_learn_kernel.model.covar_module.lengthscale)
ll_vanilla, rmse_vanilla, _ = gpr_model_vanilla.eval(
self.x_train, self.y_train_sin)
ll_kernel, rmse_kernel, _ = gpr_model_learn_kernel.eval(
self.x_train, self.y_train_sin)
print('learning_mode', learning_mode)
print(ll_kernel, ll_vanilla)
print(rmse_kernel, rmse_vanilla)
self.assertGreater(ll_kernel, ll_vanilla)
self.assertLess(rmse_kernel, rmse_vanilla)
def test_serializable(self):
torch.manual_seed(40)
np.random.seed(22)
import itertools
# check that more datasets improve performance
for mean_module, covar_module in itertools.product(['constant', 'NN'],
['SE', 'NN']):
gpr_model = GPRegressionLearned(self.x_train,
self.y_train_two,
learning_mode='both',
num_iter_fit=10,
mean_module=mean_module,
covar_module='NN',
random_seed=22)
gpr_model.fit()
pred_1 = gpr_model.predict(self.x_train)
gpr_model2 = GPRegressionLearned(self.x_train,
self.y_train_two,
learning_mode='both',
num_iter_fit=1,
mean_module=mean_module,
covar_module='NN',
random_seed=345)
gpr_model2.fit()
pred_2 = gpr_model2.predict(self.x_train)
file = ('/tmp/test_torch_serialization.pkl')
torch.save(gpr_model.state_dict(), file)
gpr_model2.load_state_dict(torch.load(file))
pred_3 = gpr_model2.predict(self.x_train)
assert not np.array_equal(pred_1, pred_2)
assert np.array_equal(pred_1, pred_3)