https://github.com/GPflow/GPflow
Tip revision: b8a05fb755d8b420d55d1b20dcc9559cf83dc152 authored by ST John on 04 January 2020, 00:18:23 UTC
Merge branch 'develop' into st/posterior
Merge branch 'develop' into st/posterior
Tip revision: b8a05fb
vgp.py
# Copyright 2016 James Hensman, Valentine Svensson, alexggmatthews, fujiisoup
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from typing import Optional
import numpy as np
import tensorflow as tf
import gpflow
from ..base import Parameter
from ..conditionals import conditional
from ..config import default_float, default_jitter
from ..kernels import Kernel
from ..kullback_leiblers import gauss_kl
from ..likelihoods import Likelihood
from ..mean_functions import MeanFunction, Zero
from ..models.model import Data, DataPoint, GPModel, MeanAndVariance, GPPosterior
from ..utilities import triangular
class VGP(GPModel):
r"""
This method approximates the Gaussian process posterior using a multivariate Gaussian.
The idea is that the posterior over the function-value vector F is
approximated by a Gaussian, and the KL divergence is minimised between
the approximation and the posterior.
This implementation is equivalent to svgp with X=Z, but is more efficient.
The whitened representation is used to aid optimization.
The posterior approximation is
.. math::
q(\mathbf f) = N(\mathbf f \,|\, \boldsymbol \mu, \boldsymbol \Sigma)
"""
def __init__(self,
kernel: Kernel,
likelihood: Likelihood,
mean_function: Optional[MeanFunction] = None,
num_data: Optional[int] = 10,
num_latent: Optional[int] = 1):
"""
kernel, likelihood, mean_function are appropriate GPflow objects
num_latent is the number of latent GP functions that are needed.
Usually, this is the same as the number of columns of the Y data
matrix.
"""
super().__init__(kernel, likelihood, mean_function)
self.num_latent = num_latent
self.q_mu = Parameter(tf.Variable(shape=(None, num_latent), initial_value=np.zeros((num_data, self.num_latent))))
transform = triangular()
q_sqrt = np.array([np.eye(num_data) for _ in range(self.num_latent)])
q_sqrt = transform.inverse(q_sqrt).numpy()
self.q_sqrt = Parameter(tf.Variable(shape=(num_latent, None), initial_value=q_sqrt), transform=transform)
def elbo(self, data: Data):
"""
Data is a tuple (X, Y) where
X is a data matrix, size [N, D]
Y is a data matrix, size [N, R]
This method computes the variational lower bound on the likelihood,
which is:
E_{q(F)} [ \log p(Y|F) ] - KL[ q(F) || p(F)]
with
q(\mathbf f) = N(\mathbf f \,|\, \boldsymbol \mu, \boldsymbol \Sigma)
"""
x_data, y_data = data
# Get prior KL.
KL = gauss_kl(self.q_mu, self.q_sqrt)
# Get conditionals
num_data = x_data.shape[0]
K = self.kernel(x_data) + tf.eye(num_data, dtype=default_float()) * default_jitter()
L = tf.linalg.cholesky(K)
fmean = tf.linalg.matmul(L, self.q_mu) + self.mean_function(x_data) # [NN, ND] -> ND
q_sqrt_dnn = tf.linalg.band_part(self.q_sqrt, -1, 0) # [D, N, N]
L_tiled = tf.tile(tf.expand_dims(L, 0), tf.stack([self.num_latent, 1, 1]))
LTA = tf.linalg.matmul(L_tiled, q_sqrt_dnn) # [D, N, N]
fvar = tf.reduce_sum(tf.square(LTA), 2)
fvar = tf.transpose(fvar)
# Get variational expectations.
var_exp = self.likelihood.variational_expectations(fmean, fvar, y_data)
return tf.reduce_sum(var_exp) - KL
def objective(self, data: Data):
return -self.elbo(data)
def get_posterior(self, data: Data):
return GPPosterior(mean_function=self.mean_function,
kernel=self.kernel,
likelihood=self.likelihood,
inducing_variable=data[0],
whiten=True,
mean=self.q_mu.numpy(),
variance_sqrt=self.q_sqrt.numpy())
class VGPOpperArchambeau(GPModel):
r"""
This method approximates the Gaussian process posterior using a multivariate Gaussian.
The key reference is:
::
@article{Opper:2009,
title = {The Variational Gaussian Approximation Revisited},
author = {Opper, Manfred and Archambeau, Cedric},
journal = {Neural Comput.},
year = {2009},
pages = {786--792},
}
The idea is that the posterior over the function-value vector F is
approximated by a Gaussian, and the KL divergence is minimised between
the approximation and the posterior. It turns out that the optimal
posterior precision shares off-diagonal elements with the prior, so
only the diagonal elements of the precision need be adjusted.
The posterior approximation is
.. math::
q(\mathbf f) = N(\mathbf f \,|\, \mathbf K \boldsymbol \alpha,
[\mathbf K^{-1} + \textrm{diag}(\boldsymbol \lambda))^2]^{-1})
This approach has only 2ND parameters, rather than the N + N^2 of vgp,
but the optimization is non-convex and in practice may cause difficulty.
"""
def __init__(self,
data: Data,
kernel: Kernel,
likelihood: Likelihood,
mean_function: MeanFunction = None,
num_latent: Optional[int] = None):
"""
X is a data matrix, size [N, D]
Y is a data matrix, size [N, R]
kernel, likelihood, mean_function are appropriate GPflow objects
"""
mean_function = Zero() if mean_function is None else mean_function
super().__init__(kernel, likelihood, mean_function, num_latent)
x_data, y_data = data
self.data = data
self.num_data = x_data.shape[0]
self.num_latent = num_latent or y_data.shape[1]
self.q_alpha = Parameter(np.zeros((self.num_data, self.num_latent)))
self.q_lambda = Parameter(np.ones((self.num_data, self.num_latent)), transform=gpflow.utilities.positive())
def log_likelihood(self):
r"""
q_alpha, q_lambda are variational parameters, size [N, R]
This method computes the variational lower bound on the likelihood,
which is:
E_{q(F)} [ \log p(Y|F) ] - KL[ q(F) || p(F)]
with
q(f) = N(f | K alpha + mean, [K^-1 + diag(square(lambda))]^-1) .
"""
x_data, y_data = self.data
K = self.kernel(x_data)
K_alpha = tf.linalg.matmul(K, self.q_alpha)
f_mean = K_alpha + self.mean_function(x_data)
# compute the variance for each of the outputs
I = tf.tile(tf.eye(self.num_data, dtype=default_float())[None, ...], [self.num_latent, 1, 1])
A = I + tf.transpose(self.q_lambda)[:, None, ...] * tf.transpose(self.q_lambda)[:, :, None, ...] * K
L = tf.linalg.cholesky(A)
Li = tf.linalg.triangular_solve(L, I)
tmp = Li / tf.transpose(self.q_lambda)[:, None, ...]
f_var = 1. / tf.square(self.q_lambda) - tf.transpose(tf.reduce_sum(tf.square(tmp), 1))
# some statistics about A are used in the KL
A_logdet = 2.0 * tf.reduce_sum(tf.math.log(tf.linalg.diag_part(L)))
trAi = tf.reduce_sum(tf.square(Li))
KL = 0.5 * (A_logdet + trAi - self.num_data * self.num_latent + tf.reduce_sum(K_alpha * self.q_alpha))
v_exp = self.likelihood.variational_expectations(f_mean, f_var, y_data)
return tf.reduce_sum(v_exp) - KL
def predict_f(self, predict_at: DataPoint, full_cov: bool = False, full_output_cov: bool = False):
r"""
The posterior variance of F is given by
q(f) = N(f | K alpha + mean, [K^-1 + diag(lambda**2)]^-1)
Here we project this to F*, the values of the GP at Xnew which is given
by
q(F*) = N ( F* | K_{*F} alpha + mean, K_{**} - K_{*f}[K_{ff} +
diag(lambda**-2)]^-1 K_{f*} )
Note: This model cuurently does not allow full output covariances
"""
assert full_output_cov == False
x_data, _y_data = self.data
# compute kernel things
Kx = self.kernel(x_data, predict_at)
K = self.kernel(x_data)
# predictive mean
f_mean = tf.linalg.matmul(Kx, self.q_alpha, transpose_a=True) + self.mean_function(predict_at)
# predictive var
A = K + tf.linalg.diag(tf.transpose(1. / tf.square(self.q_lambda)))
L = tf.linalg.cholesky(A)
Kx_tiled = tf.tile(Kx[None, ...], [self.num_latent, 1, 1])
LiKx = tf.linalg.triangular_solve(L, Kx_tiled)
if full_cov:
f_var = self.kernel(predict_at) - tf.linalg.matmul(LiKx, LiKx, transpose_a=True)
else:
f_var = self.kernel(predict_at, full=False) - tf.reduce_sum(tf.square(LiKx), 1)
return f_mean, tf.transpose(f_var)