def test_model_fit():
indim = 32
X = np.random.randn(50, indim)
y = np.random.randn(50)
t = dklGPR(indim, precision="single")
assert_equal(len(t.train_loss), 0)
t.fit(X, y, 2)
assert_equal(len(t.train_loss), 2)
def test_model_fit_ensemble(shared_emb):
indim = 32
X = np.random.randn(50, indim)
y = np.random.randn(50)
t = dklGPR(indim, precision="single", shared_embedding_space=shared_emb)
assert_equal(len(t.train_loss), 0)
t.fit_ensemble(X, y, 2, n_models=3)
assert_equal(len(t.train_loss), 2)
def test_model_decode_vector(embedim):
indim = 32
X = np.random.randn(50, indim)
y = np.random.randn(3, 50)
t = dklGPR(indim, embedim, precision="single")
t.fit(X, y)
z = np.random.randn(2, embedim)
decoded = t.decode(z)
assert_equal(decoded[0].shape, (2, 3))
def test_model_embed(embedim):
indim = 32
X = np.random.randn(50, indim)
y = np.random.randn(50)
X_test = np.random.randn(50, indim)
t = dklGPR(indim, embedim, precision="single")
t.fit(X, y)
y_embedded = t.embed(X_test)
assert_equal(y_embedded.shape[0], X_test.shape[0])
assert_equal(y_embedded.shape[1], embedim)
assert_(isinstance(y_embedded, np.ndarray))
def test_thompson_scalarize():
indim = 32
X = np.random.randn(50, indim)
y = np.random.randn(3, 50)
X_test = np.random.randn(50, indim)
m = dklGPR(indim, precision="single")
m.fit(X, y)
sample, xnext = m.thompson(X_test, scalarize_func=lambda x: x.mean(0))
assert_(isinstance(xnext, np.ndarray))
assert_(isinstance(sample, np.ndarray))
assert_equal(sample.shape, (1, 50))
def test_thompson(reg_dim):
indim = 32
X = np.random.randn(50, indim)
y = np.random.randn(reg_dim, 50)
X_test = np.random.randn(50, indim)
m = dklGPR(indim, precision="single")
m.fit(X, y)
sample, xnext = m.thompson(X_test)
assert_(isinstance(xnext, np.ndarray))
assert_(isinstance(sample, np.ndarray))
assert_equal(sample.shape, (reg_dim, 50))
def test_multi_model_predict():
indim = 32
X = np.random.randn(50, indim)
y = np.random.randn(2, 50)
X_test = np.random.randn(50, indim)
t = dklGPR(indim, shared_embedding_space=False, precision="single")
t.fit(X, y)
y_pred = t.predict(X_test)
assert_equal(y_pred[0].shape, y_pred[1].shape)
assert_equal(y_pred[0].shape[1], X.shape[0])
assert_(isinstance(y_pred[0], np.ndarray))
assert_(isinstance(y_pred[1], np.ndarray))
def test_model_predict():
indim = 32
X = np.random.randn(50, indim)
y = np.random.randn(50)
X_test = np.random.randn(50, indim)
t = dklGPR(indim, precision="single")
t.fit(X, y)
y_pred = t.predict(X_test)
assert_equal(len(y_pred[0]), len(y_pred[1]))
assert_equal(y_pred[0].shape[0], X.shape[0])
assert_(isinstance(y_pred[0], np.ndarray))
assert_(isinstance(y_pred[1], np.ndarray))
def test_sample_from_multi_model_posterior():
indim = 32
num_samples = 100
X = np.random.randn(50, indim)
y = np.random.randn(2, 50)
X_test = np.random.randn(50, indim)
t = dklGPR(indim, shared_embedding_space=False, precision="single")
t.fit(X, y)
samples = t.sample_from_posterior(X_test, num_samples)
assert_equal(samples.shape[0], num_samples)
assert_equal(samples.shape[1], 2)
assert_equal(samples.shape[2], len(X_test))
assert_(isinstance(samples, np.ndarray))
def test_sample_from_posterior(reg_dim):
indim = 32
num_samples = 100
X = np.random.randn(50, indim)
y = np.random.randn(reg_dim, 50)
X_test = np.random.randn(50, indim)
t = dklGPR(indim, precision="single")
t.fit(X, y)
samples = t.sample_from_posterior(X_test, num_samples)
assert_equal(samples.shape[0], num_samples)
assert_equal(samples.shape[1], reg_dim)
assert_equal(samples.shape[2], len(X))
assert_(isinstance(samples, np.ndarray))
def test_ensemble_predict(shared_emb, ydim):
indim = 32
n_models = 3
X = np.random.randn(50, indim)
y = np.random.randn(*ydim)
X_test = np.random.randn(50, indim)
t = dklGPR(indim, shared_embedding_space=shared_emb, precision="single")
t.fit_ensemble(X, y, 1, n_models=n_models)
y_pred = t.predict(X_test)
assert_equal(y_pred[0].shape, y_pred[1].shape)
assert_equal(y_pred[0].shape[1], X.shape[0])
assert_equal(y_pred[0].shape[0], n_models)
assert_(isinstance(y_pred[0], np.ndarray))
assert_(isinstance(y_pred[1], np.ndarray))
def test_multi_model_embed(embedim):
indim = 32
X = np.random.randn(50, indim)
y = np.random.randn(3, 50)
X_test = np.random.randn(50, indim)
t = dklGPR(indim,
embedim,
shared_embedding_space=False,
precision="single")
t.fit(X, y)
y_embedded = t.embed(X_test)
print(y_embedded.shape)
assert_equal(y_embedded.shape[0], 3)
assert_equal(y_embedded.shape[1], X_test.shape[0])
assert_equal(y_embedded.shape[2], embedim)
assert_(isinstance(y_embedded, np.ndarray))