def test_construction(self):
with self.test_context():
gpflow.ParamList([])
gpflow.ParamList([gpflow.Param(1)])
gpflow.ParamList([1.0, np.array([1, 2]), gpflow.Param(1.0)])
with self.assertRaises(ValueError):
gpflow.ParamList([gpflow.Param(1), 'stringsnotallowed'])
with self.assertRaises(ValueError):
# tuples not valid in constuctor:
gpflow.ParamList((gpflow.Param(1), ))
with self.assertRaises(ValueError):
# param objects not valid in constructor (must be in list)
gpflow.ParamList(gpflow.Param(1))
with gpflow.defer_build():
p = gpflow.ParamList([0.0])
p[0] = gpflow.Param(1.0)
with self.assertRaises(ValueError):
p[0] = 1.0
with self.assertRaises(ValueError):
p[0] = "test"
p = gpflow.ParamList([])
p.append(gpflow.Param(1.0))
p.append(gpflow.Param(2.0))
p.append(2.0)
self.assertEqual(len(p), 3)
with self.assertRaises(ValueError):
p.append("test")
def test_setitem(self):
with self.test_context():
p1 = gpflow.Param(1.2)
p2 = gpflow.Param(np.array([3.4, 5.6], settings.float_type))
param_list = gpflow.ParamList([p1, p2],
name='param_list',
autobuild=False)
self.assertEqual(p1.read_value(), param_list[0].read_value())
self.assertTrue(
np.all(param_list[1].read_value() == p2.read_value()))
param_list[0] = gpflow.Param(2.0)
self.assertEqual(p1.read_value(), 1.2)
self.assertEqual(p1.root, p1)
self.assertEqual(param_list[0].read_value(), 2.0)
arr = np.array([1.1, 2.2], settings.float_type)
param_list[1] = gpflow.Param(arr)
self.assertEqual(p2.root, p2)
self.assertTrue(np.all(param_list[1].read_value() == arr))
param_list.compile()
with self.assertRaises(GPflowError):
param_list[0] = gpflow.Param(12)
def test_naming(self):
with self.test_context():
p1 = gpflow.Param(1.2)
p2 = gpflow.Param(np.array([3.4, 5.6], settings.float_type))
l = gpflow.ParamList([p1, p2])
assert p1.pathname == l.name + '/0'
assert p2.pathname == l.name + '/1'
def test_naming(self):
with self.test_context():
p1 = gpflow.Param(1.2)
p2 = gpflow.Param(np.array([3.4, 5.6], settings.np_float))
gpflow.ParamList([p1, p2])
self.assertEqual(p1.name, 'item0')
self.assertEqual(p2.name, 'item1')
def __init__(self, X, Y, ms, a, b, kerns):
for kern in kerns:
assert isinstance(
kern, (gpflow.kernels.Matern12, gpflow.kernels.Matern32,
gpflow.kernels.Matern52))
likelihood = gpflow.likelihoods.Gaussian()
mean_function = gpflow.mean_functions.Zero()
gpflow.models.GPModel.__init__(self, X, Y, None, likelihood,
mean_function)
self.kerns = gpflow.ParamList(kerns)
self.num_data = X.shape[0]
self.num_latent = Y.shape[1]
self.a = a
self.b = b
self.ms = ms
# count the inducing variables:
self.Ms = []
for kern in kerns:
Ncos_d = self.ms.size
Nsin_d = self.ms.size - 1
self.Ms.append(Ncos_d + Nsin_d)
assert np.all(X > a)
assert np.all(X < b)
# pre compute static quantities
Kuf = [
make_Kuf_np(X[:, i:i + 1], a, b, self.ms)
for i, (a, b) in enumerate(zip(self.a, self.b))
]
self.Kuf = make_kvs_np(Kuf)
self.KufY = np.dot(self.Kuf, Y)
self.KufKfu = np.dot(self.Kuf, self.Kuf.T)
self.tr_YTY = np.sum(np.square(Y))
def test_len(self):
with self.test_context():
p1 = gpflow.Param(1.2)
p2 = gpflow.Param(np.array([3.4, 5.6], settings.np_float))
l = gpflow.ParamList([p1])
l.append(p2)
self.assertTrue(len(l) == 2)
def test_with_parameterized(self):
with self.test_context():
pzd = gpflow.params.Parameterized()
p = gpflow.Param(1.2)
pzd.p = p
param_list = gpflow.ParamList([pzd])
param_list[0].p = 5.
self.assertEqual(param_list[0].p.read_value(), 5)
def test_append(self):
with self.test_context():
p1 = gpflow.Param(1.2)
p2 = gpflow.Param(np.array([3.4, 5.6], settings.np_float))
param_list = gpflow.ParamList([p1])
param_list.append(p2)
self.assertTrue(p2 in param_list.params)
with self.assertRaises(ValueError):
param_list.append('foo')
def __init__(self, X, Y, ms, a, b, kerns, likelihood):
"""
Here we assume the interval is [a,b]
"""
assert a.size == b.size == len(kerns) == X.shape[1]
for kern in kerns:
assert isinstance(
kern, (gpflow.kernels.Matern12, gpflow.kernels.Matern32,
gpflow.kernels.Matern52))
mf = gpflow.mean_functions.Zero()
gpflow.models.GPModel.__init__(self,
X,
Y,
kern=None,
likelihood=likelihood,
mean_function=mf)
self.num_latent = 1 # multiple columns not supported in this version
self.a = a
self.b = b
self.ms = ms
self.input_dim = X.shape[1]
# initialize variational parameters
Ms = []
for kern in kerns:
Ncos_d = self.ms.size
Nsin_d = self.ms.size - 1
if isinstance(kern, gpflow.kernels.Matern12):
Ncos_d += 1
elif isinstance(kern, gpflow.kernels.Matern32):
Ncos_d += 1
Nsin_d += 1
else:
raise NotImplementedError
Ms.append(Ncos_d + Nsin_d)
self.kerns = gpflow.ParamList(kerns)
self.q_mu = gpflow.Param(np.zeros((np.sum(Ms), 1)))
pos = gpflow.transforms.positive
self.q_sqrt = gpflow.ParamList(
[gpflow.Param(np.ones(M), pos) for M in Ms])
def __init__(self, X, Y, ms, a, b, kerns, likelihood):
"""
Here we assume the interval is [a,b]
We do *not* assume that the variance of q(u) has a kronecker structure.
This can get very computationally heavy very quickly, use with caution!.
"""
assert a.size == b.size == len(kerns) == X.shape[1]
for kern in kerns:
assert isinstance(
kern, (gpflow.kernels.Matern12, gpflow.kernels.Matern32,
gpflow.kernels.Matern52))
mf = gpflow.mean_functions.Zero()
gpflow.models.GPModel.__init__(self,
X,
Y,
kern=None,
likelihood=likelihood,
mean_function=mf)
self.num_latent = 1 # multiple columns not supported in this version
self.a = a
self.b = b
self.ms = ms
# initialize variational parameters
Ms = []
for kern in kerns:
Ncos_d = self.ms.size
Nsin_d = self.ms.size - 1
Ms.append(Ncos_d + Nsin_d)
self.Ms = Ms
self.kerns = gpflow.ParamList(kerns)
self.q_mu = gpflow.Param(np.zeros((np.prod(Ms), 1)))
# The covariance matrix gets very big very quickly
self.q_sqrt = gpflow.Param(np.eye(np.prod(Ms)))
# pre-compute the Kuf matrices
self._Kuf = [
tf.constant(make_Kuf_np(X[:, i:i + 1], ai, bi, self.ms))
for i, (ai, bi) in enumerate(zip(self.a, self.b))
]
def test_construction(self):
with self.test_context():
gpflow.ParamList([])
gpflow.ParamList([gpflow.Param(1)])
gpflow.ParamList([1.0, np.array([1, 2]), gpflow.Param(1.0)])
with self.assertRaises(ValueError):
gpflow.ParamList([gpflow.Param(1), 'stringsnotallowed'])
with self.assertRaises(ValueError):
# tuples not valid in constuctor:
gpflow.ParamList((gpflow.Param(1), ))
with self.assertRaises(ValueError):
# param objects not valid in constructor (must be in list)
gpflow.ParamList(gpflow.Param(1))
def test_append(self):
with self.test_context():
p1 = gpflow.Param(1.2)
p4 = gpflow.Param(np.array([3.4, 5.6], settings.float_type))
with gpflow.defer_build():
p2 = gpflow.Param(1.2)
param_list = gpflow.ParamList([p1])
param_list.append(p2)
p3 = gpflow.Param(1.2)
param_list.append(p3)
param_list.compile()
with self.assertRaises(gpflow.GPflowError):
param_list.append(p4)
self.assertTrue(p1 in param_list.params)
self.assertTrue(p2 in param_list.params)
self.assertTrue(p3 in param_list.params)
self.assertFalse(p4 in param_list.params)
with self.assertRaises(ValueError):
param_list.append('foo')
def __init__(self, X, Y, ms, a, b, kern_list):
assert X.shape[1] == len(kern_list)
assert a.size == len(kern_list)
assert b.size == len(kern_list)
for kern in kern_list:
assert isinstance(
kern, (gpflow.kernels.Matern12, gpflow.kernels.Matern32,
gpflow.kernels.Matern52))
likelihood = gpflow.likelihoods.Gaussian()
mean_function = gpflow.mean_functions.Zero()
gpflow.models.GPModel.__init__(self, X, Y, None, likelihood,
mean_function)
self.num_data = X.shape[0]
self.num_latent = Y.shape[1]
self.a = a
self.b = b
self.ms = ms
self.kerns = gpflow.ParamList(kern_list)
# pre compute static quantities: chunk data to save memory
self.tr_YTY = gpflow.DataHolder(np.sum(np.square(Y)))
Mtotal = (2 * self.ms.size - 1) * X.shape[1]
self.KufY = np.zeros((Mtotal, 1))
self.KufKfu = np.zeros((Mtotal, Mtotal))
for i in range(0, (X.shape[0]), 10000):
Xchunk = X[i:i + 10000]
Ychunk = Y[i:i + 10000]
Kuf_chunk = np.empty((0, Xchunk.shape[0]))
KufY_chunk = np.empty((0, Ychunk.shape[1]))
for i, (ai, bi) in enumerate(zip(self.a, self.b)):
assert np.all(Xchunk[:, i] > ai)
assert np.all(Xchunk[:, i] < bi)
Kuf = make_Kuf_np(Xchunk[:, i:i + 1], ai, bi, self.ms)
KufY_chunk = np.vstack((KufY_chunk, np.dot(Kuf, Ychunk)))
Kuf_chunk = np.vstack((Kuf_chunk, Kuf))
self.KufKfu += np.dot(Kuf_chunk, Kuf_chunk.T)
self.KufY += KufY_chunk
self.KufY = gpflow.DataHolder(self.KufY)
self.KufKfu = gpflow.DataHolder(self.KufKfu)
def test_len(self):
with self.test_context():
p1 = gpflow.Param(1.2)
p2 = gpflow.Param(np.array([3.4, 5.6], settings.float_type))
l = gpflow.ParamList([p1, p2])
self.assertTrue(len(l) == 2)
def __init__(self,
latent_dim,
Y,
transitions,
T_latent=None,
inputs=None,
emissions=None,
px1_mu=None,
px1_cov=None,
Xmu=None,
Xchol=None,
n_samples=100,
batch_size=None,
seed=None,
name=None):
super().__init__(latent_dim,
Y[0],
transitions,
T_latent=None,
inputs=None,
emissions=emissions,
px1_mu=px1_mu,
px1_cov=None,
Xmu=None,
Xchol=None,
n_samples=n_samples,
seed=seed,
name=name)
self.T = [Y_s.shape[0] for Y_s in Y]
self.T_latent = T_latent or self.T
self.n_seq = len(self.T)
self.T_tf = tf.constant(self.T, dtype=gp.settings.int_type)
self.T_latent_tf = tf.constant(self.T_latent,
dtype=gp.settings.int_type)
self.sum_T = float(sum(self.T))
self.sum_T_latent = float(sum(self.T_latent))
self.batch_size = batch_size
self.Y = gp.ParamList(Y, trainable=False)
self.inputs = None if inputs is None else gp.ParamList(inputs,
trainable=False)
_Xmu = [np.zeros((T_s, self.latent_dim))
for T_s in self.T_latent] if Xmu is None else Xmu
self.X = gp.ParamList(_Xmu)
_Xchol = [np.eye(T_s * self.latent_dim)
for T_s in self.T_latent] if Xchol is None else Xchol
xc_tr = lambda xc: None if xc.ndim == 1 else gtf.LowerTriangular(
xc.shape[-1],
num_matrices=1 if xc.ndim == 2 else xc.shape[0],
squeeze=xc.ndim == 2)
self.Xchol = gp.ParamList(
[gp.Param(xc, transform=xc_tr(xc)) for xc in _Xchol])
self.multi_diag_px1_cov = False
if isinstance(px1_cov, list): # different prior for each sequence
_x1_cov = np.stack(px1_cov)
_x1_cov = np.sqrt(
_x1_cov) if _x1_cov.ndim == 2 else np.linalg.cholesky(_x1_cov)
_transform = None if _x1_cov.ndim == 2 else gtf.LowerTriangular(
self.latent_dim, num_matrices=self.n_seq)
self.multi_diag_px1_cov = _x1_cov.ndim == 2
elif isinstance(px1_cov, np.ndarray): # same prior for each sequence
assert px1_cov.ndim < 3
_x1_cov = np.sqrt(
px1_cov) if px1_cov.ndim == 1 else np.linalg.cholesky(px1_cov)
_transform = None if px1_cov.ndim == 1 else gtf.LowerTriangular(
self.latent_dim, squeeze=True)
else:
_x1_cov = np.eye(self.latent_dim)
_transform = gtf.LowerTriangular(self.latent_dim, squeeze=True)
self.px1_cov_chol = gp.Param(_x1_cov,
trainable=False,
transform=_transform)
def __init__(self):
gpflow.models.Model.__init__(self)
self.param_list = gpflow.ParamList(
[gpflow.Param(1.), gpflow.Param(12.)])
def __init__(self,
X,
Y,
ms,
a,
b,
kerns,
likelihood,
use_two_krons=False,
use_extra_ranks=0):
"""
Here we assume the interval is [a,b]
"""
assert a.size == b.size == len(kerns) == X.shape[1]
for kern in kerns:
assert isinstance(
kern, (gpflow.kernels.Matern12, gpflow.kernels.Matern32,
gpflow.kernels.Matern52))
mf = gpflow.mean_functions.Zero()
gpflow.models.GPModel.__init__(self,
X,
Y,
kern=None,
likelihood=likelihood,
mean_function=mf)
self.num_latent = 1 # multiple columns not supported in this version
self.a = a
self.b = b
self.ms = ms
self.kerns = gpflow.ParamList(kerns)
# initialize variational parameters
self.Ms = []
for kern in kerns:
Ncos_d = self.ms.size
Nsin_d = self.ms.size - 1
self.Ms.append(Ncos_d + Nsin_d)
self.q_mu = gpflow.Param(np.zeros((np.prod(self.Ms), 1)))
# The covariance matrix
self.q_sqrt_kron = gpflow.ParamList(
[gpflow.Param(np.eye(M)) for M in self.Ms])
self.use_two_krons = use_two_krons
self.use_extra_ranks = use_extra_ranks
assert not (use_extra_ranks and use_two_krons
), "can only use one extra covariance structure at a time!"
if use_two_krons:
# same as above, but with different init to break symmetry
self.q_sqrt_kron_2 = gpflow.ParamList(
[gpflow.Param(np.eye(M) + 0.01) for M in self.Ms])
elif use_extra_ranks:
self.q_sqrt_W = gpflow.Param(
np.zeros((np.prod(self.Ms), use_extra_ranks)))
# pre-compute Kuf
self._Kuf = [
make_Kuf_np(X[:, i:i + 1], ai, bi, self.ms)
for i, (ai, bi) in enumerate(zip(self.a, self.b))
]