def __init__(self,
data: Tuple[tf.Tensor, tf.Tensor],
m: int = 20,
alpha: np.float = 1./np.sqrt(2.),
eps_sq: np.float = 1,
sigma_n_sq: np.float = 1,
sigma_f_sq: np.float = 1):
self.num_data = tf.cast(data[1].shape[0], default_float())
self.data = (tf.cast(tf.squeeze(data[0]), default_float()), tf.cast(data[1], default_float()))
self.const = tf.cast(0.5*data[1].size*np.log(2*np.pi), default_float())
D = data[0].shape[1]
self.flag_1d = D == 1
self.alpha = tf.cast(alpha, default_float())
self.alpha_sq = tf.square(self.alpha)
self.m = tf.cast(m, default_float())
self.this_range = tf.constant(np.asarray(list(product(range(1, m + 1), repeat=D))).squeeze(), dtype=default_float())
self.this_range_1 = self.this_range - 1.
self.this_range_1_2 = self.this_range_1 if self.flag_1d else tf.range(m, dtype=default_float())
self.this_range_1_int = tf.cast(self.this_range_1, tf.int32)
self.tf_range_dnn_out = tf.range(D)
self.this_range_1_ln2 = np.log(2.)*self.this_range_1
self.vander_range = tf.range(m+1, dtype=default_float())
self.eye_k = tf.eye(m**D, dtype=default_float())
self.yTy = tf.reduce_sum(tf.math.square(self.data[1]))
self.coeff_n_tf = tf.constant(np.load(os.path.dirname(os.path.realpath(__file__)) + '/hermite_coeff.npy')[:m, :m], dtype=default_float())
eps_sq = eps_sq*np.ones(D) if D > 1 else eps_sq
self.eps_sq = Parameter(eps_sq, transform=positive(), dtype=default_float())
self.sigma_f_sq = Parameter(sigma_f_sq, transform=positive(), dtype=default_float())
self.sigma_n_sq = Parameter(sigma_n_sq, transform=positive(), dtype=default_float())
def __init__(self, data, Y_var):
super().__init__(active_dims=[0])
self.Y_var = Y_var
self.num_genes = data.m_obs.shape[1]
# l_affine = tfb.AffineScalar(shift=tf.cast(1., tf.float64),
# scale=tf.cast(4-1., tf.float64))
# l_sigmoid = tfb.Sigmoid()
# l_logistic = tfb.Chain([l_affine, l_sigmoid])
self.lengthscale = gpflow.Parameter(1.414, transform=positive())
D_affine = tfb.AffineScalar(shift=tf.cast(0.1, tf.float64),
scale=tf.cast(1.5 - 0.1, tf.float64))
D_sigmoid = tfb.Sigmoid()
D_logistic = tfb.Chain([D_affine, D_sigmoid])
S_affine = tfb.AffineScalar(shift=tf.cast(0.1, tf.float64),
scale=tf.cast(4. - 0.1, tf.float64))
S_sigmoid = tfb.Sigmoid()
S_logistic = tfb.Chain([S_affine, S_sigmoid])
self.D = gpflow.Parameter(np.random.uniform(0.9, 1, self.num_genes),
transform=positive(),
dtype=tf.float64)
# self.D[3].trainable = False
# self.D[3].assign(0.8)
self.S = gpflow.Parameter(np.random.uniform(1, 1, self.num_genes),
transform=positive(),
dtype=tf.float64)
# self.S[3].trainable = False
# self.S[3].assign(1)
self.kervar = gpflow.Parameter(np.float64(1), transform=positive())
self.noise_term = gpflow.Parameter(
0.1353 * tf.ones(self.num_genes, dtype='float64'),
transform=positive())
def test_parameter_assign_validation():
with pytest.raises(tf.errors.InvalidArgumentError):
param = gpflow.Parameter(0.0, transform=positive())
param = gpflow.Parameter(0.1, transform=positive())
param.assign(0.2)
with pytest.raises(tf.errors.InvalidArgumentError):
param.assign(0.0)
def _init_variational_parameters(self, num_inducing, q_mu, q_sqrt, q_diag):
"""
Constructs the mean and cholesky of the covariance of the variational Gaussian posterior.
If a user passes values for `q_mu` and `q_sqrt` the routine checks if they have consistent
and correct shapes. If a user does not specify any values for `q_mu` and `q_sqrt`, the routine
initializes them, their shape depends on `num_inducing` and `q_diag`.
Note: most often the comments refer to the number of observations (=output dimensions) with P,
number of latent GPs with L, and number of inducing points M. Typically P equals L,
but when certain multioutput kernels are used, this can change.
Parameters
----------
:param num_inducing: int
Number of inducing variables, typically refered to as M.
:param q_mu: np.array or None
Mean of the variational Gaussian posterior. If None the function will initialise
the mean with zeros. If not None, the shape of `q_mu` is checked.
:param q_sqrt: np.array or None
Cholesky of the covariance of the variational Gaussian posterior.
If None the function will initialise `q_sqrt` with identity matrix.
If not None, the shape of `q_sqrt` is checked, depending on `q_diag`.
:param q_diag: bool
Used to check if `q_mu` and `q_sqrt` have the correct shape or to
construct them with the correct shape. If `q_diag` is true,
`q_sqrt` is two dimensional and only holds the square root of the
covariance diagonal elements. If False, `q_sqrt` is three dimensional.
"""
q_mu = np.zeros(
(num_inducing, self.num_latent_gps)) if q_mu is None else q_mu
self.q_mu = Parameter(q_mu, dtype=default_float()) # [M, P]
if q_sqrt is None:
if self.q_diag:
ones = np.ones((num_inducing, self.num_latent_gps),
dtype=default_float())
self.q_sqrt = Parameter(ones, transform=positive()) # [M, P]
else:
q_sqrt = [
np.eye(num_inducing, dtype=default_float())
for _ in range(self.num_latent_gps)
]
q_sqrt = np.array(q_sqrt)
self.q_sqrt = Parameter(q_sqrt,
transform=triangular()) # [P, M, M]
else:
if q_diag:
assert q_sqrt.ndim == 2
self.num_latent_gps = q_sqrt.shape[1]
self.q_sqrt = Parameter(q_sqrt,
transform=positive()) # [M, L|P]
else:
assert q_sqrt.ndim == 3
self.num_latent_gps = q_sqrt.shape[0]
num_inducing = q_sqrt.shape[1]
self.q_sqrt = Parameter(q_sqrt,
transform=triangular()) # [L|P, M, M]
def __init__(self,
m=1,
active_dims=[0],
gap_decay=0.1,
match_decay=0.9,
max_subsequence_length=3,
alphabet=[],
maxlen=0):
super().__init__(active_dims=active_dims)
# constrain decay kernel params to between 0 and 1
logistic_gap = tfb.Chain([
tfb.Shift(tf.cast(0, tf.float64))(tfb.Scale(tf.cast(1,
tf.float64))),
tfb.Sigmoid()
])
logisitc_match = tfb.Chain([
tfb.AffineScalar(shift=tf.cast(0, tf.float64),
scale=tf.cast(1, tf.float64)),
tfb.Sigmoid()
])
self.gap_decay = Parameter(gap_decay,
transform=logistic_gap,
name="gap_decay")
self.match_decay = Parameter(match_decay,
transform=logisitc_match,
name="match_decay")
# prepare order coefs params
order_coefs = tf.ones(max_subsequence_length)
self.order_coefs = Parameter(order_coefs,
transform=positive(),
name="order_coefs")
# get split weights
self.m = m
split_weights = tf.ones(2 * self.m - 1)
self.split_weights = Parameter(split_weights,
transform=positive(),
name="order_coefs")
# store additional kernel parameters
self.max_subsequence_length = tf.constant(max_subsequence_length)
self.alphabet = tf.constant(alphabet)
self.alphabet_size = tf.shape(self.alphabet)[0]
self.maxlen = tf.cast(tf.math.ceil(maxlen / self.m), dtype=tf.int32)
self.full_maxlen = tf.constant(maxlen)
# build a lookup table of the alphabet to encode input strings
self.table = tf.lookup.StaticHashTable(
initializer=tf.lookup.KeyValueTensorInitializer(
keys=tf.constant(["PAD"] + alphabet),
values=tf.constant(range(0,
len(alphabet) + 1)),
),
default_value=0)
def __init__(self, variance=1.0, lengthscale=1.0, f_list=None):
"""
:param variance: the (initial) value for the variance parameter
:param lengthscale: the (initial) value for the lengthscale parameter(s),
to induce ARD behaviour this must be initialised as an array the same
length as the the number of active dimensions e.g. [1., 1., 1.]
:param f_list: list with information of the functional inputs
"""
self.variance = Parameter(variance, transform=positive())
self.lengthscale = Parameter(lengthscale, transform=positive())
self.f_list = f_list # list with functional information
def __init__(self, args):
super().__init__(active_dims=[0])
self.var = gpflow.Parameter(10.0, transform=positive())
self.mag = gpflow.Parameter(1.0, transform=positive())
self.args = args
self.re = REMatchKernel(metric="polynomial",
degree=3,
gamma=1,
coef0=0,
alpha=0.5,
threshold=1e-6,
normalize_kernel=True)
def __init__(self,
data: Tuple[tf.Tensor, tf.Tensor],
m: int = 20,
d: int = 1,
alpha: np.float = 1./np.sqrt(2.),
eps_sq: np.float = 1,
sigma_n_sq: np.float = 1,
sigma_f_sq: np.float = 1,
dir_weights: str = None):
if data[1].dtype == np.float64:
K_bd.set_floatx('float64')
else:
set_default_float(np.float32)
self.num_data = tf.cast(data[1].shape[0], default_float())
self.data = (tf.cast(data[0], default_float()), tf.cast(data[1], default_float()))
self.const = tf.cast(0.5*data[1].size*np.log(2*np.pi), default_float())
self.flag_1d = d == 1
self.alpha = tf.cast(alpha, default_float())
self.alpha_sq = tf.square(self.alpha)
self.m = tf.cast(m, default_float())
self.this_range = tf.constant(np.asarray(list(product(range(1, m + 1), repeat=d))).squeeze(), dtype=default_float())
self.this_range_1 = self.this_range - 1.
self.this_range_1_2 = self.this_range_1 if self.flag_1d else tf.range(m, dtype=default_float())
self.this_range_1_int = tf.cast(self.this_range_1, tf.int32)
self.tf_range_dnn_out = tf.range(d)
self.this_range_1_ln2 = np.log(2.)*self.this_range_1
self.vander_range = tf.range(m+1, dtype=default_float())
self.eye_k = tf.eye(m**d, dtype=default_float())
self.yTy = tf.reduce_sum(tf.math.square(self.data[1]))
self.coeff_n_tf = tf.constant(np.load(os.path.dirname(os.path.realpath(__file__)) + '/hermite_coeff.npy')[:m, :m], dtype=default_float())
eps_sq = eps_sq*np.ones(d) if d > 1 else eps_sq
self.eps_sq = Parameter(eps_sq, transform=positive(), dtype=default_float())
self.sigma_f_sq = Parameter(sigma_f_sq, transform=positive(), dtype=default_float())
self.sigma_n_sq = Parameter(sigma_n_sq, transform=positive(), dtype=default_float())
model = models.Sequential()
model.add(layers.Dense(512, activation='tanh', input_dim=data[0].shape[1]))
model.add(layers.Dense(256, activation='tanh'))
model.add(layers.Dense(64, activation='tanh'))
model.add(layers.Dense(d))
if dir_weights is not None:
model.load_weights(dir_weights)
self.neural_net = model
def __init__(self,rank=1,active_dims=[0],gap_decay=0.1, match_decay=0.9,max_subsequence_length=3,
alphabet = [], maxlen=0):
super().__init__(active_dims=active_dims)
# constrain decay kernel params to between 0 and 1
logistic_gap = tfb.Chain([tfb.Shift(tf.cast(0,tf.float64))(tfb.Scale(tf.cast(1,tf.float64))),tfb.Sigmoid()])
logisitc_match = tfb.Chain([tfb.AffineScalar(shift=tf.cast(0,tf.float64),scale=tf.cast(1,tf.float64)),tfb.Sigmoid()])
self.gap_decay= Parameter(gap_decay, transform=logistic_gap ,name="gap_decay")
self.match_decay = Parameter(match_decay, transform=logisitc_match,name="match_decay")
# prepare similarity matrix parameters
self.rank=rank
W = 0.1 * tf.ones((len(alphabet), self.rank))
kappa = tf.ones(len(alphabet))
self.W = Parameter(W,name="W")
self.kappa = Parameter(kappa, transform=positive(),name="kappa")
# store additional kernel parameters
self.max_subsequence_length = tf.constant(max_subsequence_length)
self.alphabet = tf.constant(alphabet)
self.alphabet_size=tf.shape(self.alphabet)[0]
self.maxlen = tf.constant(maxlen)
# build a lookup table of the alphabet to encode input strings
self.table = tf.lookup.StaticHashTable(
initializer=tf.lookup.KeyValueTensorInitializer(
keys=tf.constant(["PAD"]+alphabet),
values=tf.constant(range(0,len(alphabet)+1)),),default_value=0)
def __init__(self,
data: RegressionData,
kernel,
noise_variance: float = 1.0,
parallel=False,
max_parallel=10000):
self.noise_variance = Parameter(noise_variance, transform=positive())
ts, ys = data_input_to_tensor(data)
super().__init__(kernel, None, None, num_latent_gps=ys.shape[-1])
self.data = ts, ys
filter_spec = kernel.get_spec(ts.shape[0])
filter_ys_spec = tf.TensorSpec((ts.shape[0], 1),
config.default_float())
smoother_spec = kernel.get_spec(None)
smoother_ys_spec = tf.TensorSpec((None, 1), config.default_float())
if not parallel:
self._kf = tf.function(
partial(kf, return_loglikelihood=True, return_predicted=False),
input_signature=[filter_spec, filter_ys_spec])
self._kfs = tf.function(
kfs, input_signature=[smoother_spec, smoother_ys_spec])
else:
self._kf = tf.function(
partial(pkf,
return_loglikelihood=True,
max_parallel=ts.shape[0]),
input_signature=[filter_spec, filter_ys_spec])
self._kfs = tf.function(
partial(pkfs, max_parallel=max_parallel),
input_signature=[smoother_spec, smoother_ys_spec])
def grad(dy, variables=None):
# get gradients of unconstrained params
grads = {}
if self.symmetric:
grads['gap_decay:0'] = tf.reduce_sum(
tf.multiply(
dy,
dk_dgap * tf.math.exp(
self.logistic_gap.forward_log_det_jacobian(
self.gap_decay_unconstrained, 0))))
grads['match_decay:0'] = tf.reduce_sum(
tf.multiply(
dy,
dk_dmatch * tf.math.exp(
self.logisitc_match.forward_log_det_jacobian(
self.match_decay_unconstrained, 0))))
dy_tiled = tf.tile(tf.expand_dims(dy, 1),
(1, self.max_subsequence_length))
grads['order_coefs:0'] = tf.reduce_sum(
tf.multiply(
dy_tiled,
dk_dorder_coefs *
tf.math.exp(positive().forward_log_det_jacobian(
self.order_coefs_unconstrained, 0))), 0)
gradient = [grads[v.name] for v in variables]
else:
gradient = [None for v in variables]
return ((None, None), gradient)
def test_positive_lower(env_lower, override_lower):
expected_lower = override_lower or env_lower
with as_context(
Config(positive_bijector="softplus", positive_minimum=env_lower)):
bijector = positive(lower=override_lower)
assert isinstance(bijector, tfp.bijectors.Chain)
assert np.isclose(bijector.bijectors[0].shift, expected_lower)
def grad(dy, variables=None):
# get gradients of unconstrained params
grads = {}
if self.symmetric:
grads['gap_decay:0'] = tf.reduce_sum(
tf.multiply(
dy,
dk_dgap * tf.math.exp(
self.logistic_gap.forward_log_det_jacobian(
self.gap_decay_unconstrained, 0))))
grads['match_decay:0'] = tf.reduce_sum(
tf.multiply(
dy,
dk_dmatch * tf.math.exp(
self.logisitc_match.forward_log_det_jacobian(
self.match_decay_unconstrained, 0))))
dy_tiled = tf.tile(tf.expand_dims(dy, 1),
(1, self.alphabet_size))
grads['kappa:0'] = tf.reduce_sum(
tf.multiply(
dy_tiled,
dk_dkappa *
tf.math.exp(positive().forward_log_det_jacobian(
self.kappa_unconstrained, 0))), 0)
dy_tiled = tf.tile(tf.expand_dims(dy, 1),
(1, self.alphabet_size * self.rank))
grads_temp = tf.reduce_sum(tf.multiply(dy_tiled, dk_dW), 0)
grads['W:0'] = tf.reshape(grads_temp,
(-1, self.alphabet_size, self.rank))
gradient = [grads[v.name] for v in variables]
else:
gradient = [None for v in variables]
return ((None, None), gradient)
def test_positive_calculation_order():
value, lower = -10.0, 10.0
expected = np.exp(value) + lower
with as_context(Config(positive_bijector="exp", positive_minimum=lower)):
result = positive()(value).numpy()
assert np.isclose(result, expected)
assert result >= lower
def __init__(self,
variance,
lengthscales,
name='Kernel',
active_dims=None):
""" Kernel Constructor.
Args:
variance: An (L,L) symmetric, positive definite matrix for the signal variance.
lengthscales: An (L,M) matrix of positive definite lengthscales.
is_lengthscales_trainable: Whether the lengthscales of this kernel are trainable.
name: The name of this kernel.
active_dims: Which of the input dimensions are used. The default None means all of them.
"""
super(AnisotropicStationary, self).__init__(
active_dims=active_dims, name=name
) # Do not call gf.kernels.AnisotropicStationary.__init__()!
self.variance = Variance(value=np.atleast_2d(variance),
name=name + 'Variance')
self._L = self.variance.shape[0]
lengthscales = data_input_to_tensor(lengthscales)
lengthscales_shape = tuple(tf.shape(lengthscales).numpy())
self._M = 1 if lengthscales_shape in ((), (1, ), (1, 1), (
self._L, )) else lengthscales_shape[-1]
lengthscales = tf.reshape(
tf.broadcast_to(lengthscales, (self._L, self._M)),
(self._L, 1, self._M))
self.lengthscales = Parameter(lengthscales,
transform=positive(),
trainable=False,
name=name + 'Lengthscales')
self._validate_ard_active_dims(self.lengthscales[0, 0])
def __init__(self,
variance=1.0,
lengthscale=1.0,
f_list=None,
f_list2=None,
distances=None,
**kwargs):
for kwarg in kwargs:
if kwarg not in {'name', 'active_dims'}:
raise TypeError('Unknown keyword argument:', kwarg)
super().__init__(**kwargs)
self.variance = gpflow.Parameter(variance, transform=positive())
self.lengthscale = gpflow.Parameter(lengthscale, transform=positive())
self.f_list = f_list # list with functional information
self.f_list2 = f_list2 # list with functional information
self.distances = distances # matrix with precomputed distances
def __init__(self,
variance=1.0,
lengthscale=1.0,
alpha=1.0,
active_dims=None):
super().__init__(variance=variance,
lengthscale=lengthscale,
active_dims=active_dims)
self.alpha = Parameter(alpha, transform=positive())
def __init__(self,
variance=1.0,
lengthscale_x=1.0,
lengthscale_f=1.0,
f_list=None,
f_list2=None,
**kwargs):
for kwarg in kwargs:
if kwarg not in {'name', 'active_dims'}:
raise TypeError('Unknown keyword argument:', kwarg)
super().__init__(**kwargs)
#super().__init__(active_dims=[0])
self.variance = gpflow.Parameter(variance, transform=positive())
self.lengthscale_x = gpflow.Parameter(lengthscale_x,
transform=positive())
self.lengthscale_f = gpflow.Parameter(lengthscale_f,
transform=positive())
self.f_list = f_list # list with functional information
self.f_list2 = f_list2 # list with functional information
def __init__(self, **kwargs):
"""
:param kwargs: accepts `name` and `active_dims`, which is a list or
slice of indices which controls which columns of X are used (by
default, all columns are used).
"""
for kwarg in kwargs:
if kwarg not in {"name", "active_dims"}:
raise TypeError("Unknown keyword argument:", kwarg)
super().__init__(**kwargs)
self.variance = gpflow.Parameter(1.0, transform=positive())
def __init__(self, input_dim=1, variance=1.0, lengthscales=1.0,
num_qual=0, dist='manhattan',**kwargs):
"""
:param variance: the (initial) value for the variance parameter
:param lengthscale: the (initial) value for the lengthscale parameter(s),
to induce ARD behaviour this must be initialised as an array the same
length as the the number of active dimensions e.g. [1., 1., 1.]
:param num_qual: the number of qualitative variables to optimiser, these
must be in the end columns of the input array
:param kwargs: accepts `name` and `active_dims`, which is a list of
length input_dim which controls which columns of X are used
"""
for kwarg in kwargs:
if kwarg not in {'name', 'active_dims'}:
raise TypeError('Unknown keyword argument:', kwarg)
super().__init__(**kwargs)
self.variance = gpf.Parameter(variance, transform=positive())
self.lengthscales = gpf.Parameter(lengthscales, transform=positive())
self._validate_ard_active_dims(self.lengthscales)
self.num_qual = num_qual
self.dist = dist
def __init__(self,
data: Tuple[tf.Tensor, tf.Tensor],
m: int = 100,
lengthscales = None,
sigma_n_sq: np.float = 1,
sigma_f_sq: np.float = 1,
randn = None):
self.num_data = tf.cast(data[1].size, default_float())
self.data = (tf.cast(data[0], default_float()), tf.cast(data[1], default_float()))
self.const = tf.cast(0.5*data[1].size*np.log(2*np.pi), default_float())
self.eye_2m = tf.eye(2*m, dtype=default_float())
self.yTy = tf.reduce_sum(tf.math.square(self.data[1]))
self.m_float = tf.cast(m, default_float())
self.randn = tf.random.normal(shape=[m, data[0].shape[1]], dtype=default_float()) if randn is None else tf.cast(randn[:, None], default_float())
lengthscales0 = np.ones(data[0].shape[1]) if lengthscales is None else lengthscales
self.lengthscales = Parameter(lengthscales0, transform=positive(), dtype=default_float())
self.sigma_f_sq = Parameter(sigma_f_sq, transform=positive(), dtype=default_float())
self.sigma_n_sq = Parameter(sigma_n_sq, transform=positive(), dtype=default_float())
def __init__(self,
data: Tuple[tf.Tensor, tf.Tensor],
m: int = 100,
d: int = 4,
lengthscales = None,
sigma_n_sq: np.float = 1,
sigma_f_sq: np.float = 1,
dir_weights: str = None):
if data[1].dtype == np.float64:
K_bd.set_floatx('float64')
else:
set_default_float(np.float32)
self.num_data = tf.cast(data[1].shape[0], default_float())
self.data = (tf.cast(data[0], default_float()), tf.cast(data[1], default_float()))
self.const = tf.cast(0.5*data[1].size*np.log(2*np.pi), default_float())
self.eye_2m = tf.eye(2*m, dtype=default_float())
self.yTy = tf.reduce_sum(tf.math.square(self.data[1]))
self.m_float = tf.cast(m, default_float())
self.randn = tf.random.normal(shape=[m, d], dtype=default_float())
lengthscales0 = np.ones(d) if lengthscales is None else lengthscales
self.lengthscales = Parameter(lengthscales0, transform=positive(), dtype=default_float())
self.sigma_f_sq = Parameter(sigma_f_sq, transform=positive(), dtype=default_float())
self.sigma_n_sq = Parameter(sigma_n_sq, transform=positive(), dtype=default_float())
model = models.Sequential()
model.add(layers.Dense(512, activation='tanh', input_dim=data[0].shape[1]))
model.add(layers.Dense(256, activation='tanh'))
model.add(layers.Dense(64, activation='tanh'))
model.add(layers.Dense(d))
if dir_weights is not None:
model.load_weights(dir_weights)
self.neural_net = model
def __init__(self,
kernels,
lengthscale_f,
f_list=None,
f_list2=None,
distances=None,
W=None,
name=None):
Combination.__init__(self, kernels, name)
self.lengthscale_f = gpflow.Parameter(lengthscale_f,
transform=positive())
self.f_list = f_list # list with functional information
self.f_list2 = f_list2 # list with functional information
self.distances = distances # matrix with precomputed distances
self.W = W
def __init__(
self,
labels: Tuple[int, ...],
variances: Optional[np.ndarray] = None,
active_dims: List[int] = None,
):
super().__init__(active_dims=active_dims)
self.labels = np.array(labels)
self.n = len(labels)
if variances is None:
variances = tf.ones((self.n,))
self.variances = gpf.Parameter(
variances, transform=positive(), dtype=gpf.default_float()
)
def create_models(self, data):
self.models = []
for i in range(self.num_outputs):
kernel = gpflow.kernels.SquaredExponential(
lengthscales=tf.ones([
data[0].shape[1],
], dtype=float_type))
transformed_lengthscales = Parameter(
kernel.lengthscales, transform=positive(lower=1e-3))
kernel.lengthscales = transformed_lengthscales
kernel.lengthscales.prior = tfd.Gamma(f64(1.1), f64(1 / 10.0))
if i == 0:
self.models.append(
FakeGPR((data[0], data[1][:, i:i + 1]), kernel))
else:
self.models.append(
FakeGPR((data[0], data[1][:, i:i + 1]), kernel,
self.models[-1].X))
def __init__(
self,
value,
name: str = 'Variance',
cholesky_diagonal_lower_bound: float = CHOLESKY_DIAGONAL_LOWER_BOUND
):
""" Construct a non-diagonal covariance matrix. Mutable only through it's properties cholesky_diagonal and cholesky_lower_triangle.
Args:
value: A symmetric, positive definite matrix, expressed in tensorflow or numpy.
cholesky_diagonal_lower_bound: Lower bound on the diagonal of the Cholesky decomposition.
"""
super().__init__(name=name)
value = data_input_to_tensor(value)
self._shape = (value.shape[-1], value.shape[-1])
self._broadcast_shape = (value.shape[-1], 1, value.shape[-1], 1)
if value.shape != self._shape:
raise ValueError('Variance must have shape (L,L).')
cholesky = tf.linalg.cholesky(value)
self._cholesky_diagonal = tf.linalg.diag_part(cholesky)
if min(self._cholesky_diagonal) <= cholesky_diagonal_lower_bound:
raise ValueError(
f'The Cholesky diagonal of {name} must be strictly greater than {cholesky_diagonal_lower_bound}.'
)
self._cholesky_diagonal = Parameter(
self._cholesky_diagonal,
transform=positive(lower=cholesky_diagonal_lower_bound),
name=name + '.cholesky_diagonal')
mask = sum([
list(range(i * self._shape[0], i * (self._shape[0] + 1)))
for i in range(1, self._shape[0])
],
start=[])
self._cholesky_lower_triangle = Parameter(
tf.gather(tf.reshape(cholesky, [-1]), mask),
name=name + '.cholesky_lower_triangle')
self._row_lengths = tuple(range(self._shape[0]))
def __init__(self):
super().__init__(active_dims=[0])
self.var = gpflow.Parameter(1.0, transform=positive())
self.mag = gpflow.Parameter(1.0, transform=positive())
def __init__(
self,
inducing_variable: gpflow.inducing_variables.InducingVariables,
kernel: gpflow.kernels.Kernel,
domain: np.ndarray,
q_mu: np.ndarray,
q_S: np.ndarray,
*,
beta0: float = 1e-6,
num_observations: int = 1,
num_events: Optional[int] = None,
):
"""
D = number of dimensions
M = size of inducing variables (number of inducing points)
:param inducing_variable: inducing variables (here only implemented for a gpflow
.inducing_variables.InducingPoints instance, with Z of shape M x D)
:param kernel: the kernel (here only implemented for a gpflow.kernels
.SquaredExponential instance)
:param domain: lower and upper bounds of (hyper-rectangular) domain
(D x 2)
:param q_mu: initial mean vector of the variational distribution q(u)
(length M)
:param q_S: how to initialise the covariance matrix of the variational
distribution q(u) (M x M)
:param beta0: a constant offset, corresponding to initial value of the
prior mean of the GP (but trainable); should be sufficiently large
so that the GP does not go negative...
:param num_observations: number of observations of sets of events
under the distribution
:param num_events: total number of events, defaults to events.shape[0]
(relevant when feeding in minibatches)
"""
super().__init__(kernel, likelihood=None) # custom likelihood
# observation domain (D x 2)
self.domain = domain
if domain.ndim != 2 or domain.shape[1] != 2:
raise ValueError("domain must be of shape D x 2")
self.num_observations = num_observations
self.num_events = num_events
if not (isinstance(kernel, gpflow.kernels.SquaredExponential)
and isinstance(inducing_variable,
gpflow.inducing_variables.InducingPoints)):
raise NotImplementedError(
"This VBPP implementation can only handle real-space "
"inducing points together with the SquaredExponential "
"kernel.")
self.kernel = kernel
self.inducing_variable = inducing_variable
self.beta0 = Parameter(beta0, transform=positive(),
name="beta0") # constant mean offset
# variational approximate Gaussian posterior q(u) = N(u; m, S)
self.q_mu = Parameter(q_mu, name="q_mu") # mean vector (length M)
# covariance:
L = np.linalg.cholesky(
q_S) # S = L L^T, with L lower-triangular (M x M)
self.q_sqrt = Parameter(L, transform=triangular(), name="q_sqrt")
self.psi_jitter = 0.0
def __init__(self,
active_dims=[0],
gap_decay=0.1,
match_decay=0.9,
max_subsequence_length=3,
alphabet=[],
maxlen=0,
batch_size=100):
super().__init__(active_dims=active_dims)
# constrain decay kernel params to between 0 and 1
self.logistic_gap = tfb.Chain([
tfb.Shift(tf.cast(0, tf.float64))(tfb.Scale(tf.cast(1,
tf.float64))),
tfb.Sigmoid()
])
self.logisitc_match = tfb.Chain([
tfb.AffineScalar(shift=tf.cast(0, tf.float64),
scale=tf.cast(1, tf.float64)),
tfb.Sigmoid()
])
self.gap_decay_param = Parameter(gap_decay,
transform=self.logistic_gap,
name="gap_decay")
self.match_decay_param = Parameter(match_decay,
transform=self.logisitc_match,
name="match_decay")
self.order_coefs_param = Parameter(tf.ones(max_subsequence_length,
dtype=tf.float64),
transform=positive(),
name="order_coefs")
# use will use copies of the kernel params to stop building expensive computation graph
# we instead efficientely calculate gradients using dynamic programming
# These params are updated at every call to K and K_diag (to check if parameters have been updated)
self.match_decay = self.match_decay_param.numpy()
self.gap_decay = self.gap_decay_param.numpy()
self.order_coefs = self.order_coefs_param.numpy()
self.match_decay_unconstrained = self.match_decay_param.unconstrained_variable.numpy(
)
self.gap_decay_unconstrained = self.gap_decay_param.unconstrained_variable.numpy(
)
self.order_coefs_unconstrained = self.order_coefs_param.unconstrained_variable.numpy(
)
# store additional kernel parameters
self.max_subsequence_length = tf.constant(max_subsequence_length)
self.alphabet = tf.constant(alphabet)
self.alphabet_size = tf.shape(self.alphabet)[0]
self.maxlen = tf.constant(maxlen)
self.batch_size = tf.constant(batch_size)
# build a lookup table of the alphabet to encode input strings
self.table = tf.lookup.StaticHashTable(
initializer=tf.lookup.KeyValueTensorInitializer(
keys=tf.constant(["PAD"] + alphabet),
values=tf.constant(range(0,
len(alphabet) + 1)),
),
default_value=0)
# initialize helful construction matricies to be lazily computed once needed
self.D = None
self.dD_dgap = None
def test_positive_bijector(env_bijector, override_bijector, expected_class):
with as_context(
Config(positive_bijector=env_bijector, positive_minimum=None)):
bijector = positive(base=override_bijector)
assert isinstance(bijector, expected_class)
