# Copyright 2017-2020 The GPflow Contributors. All Rights Reserved.
#
# 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.
# Eventually, it would be nice to not have to have our own classes for
# probability distributions. The TensorFlow "distributions" framework would
# be a good replacement.
from abc import ABC, abstractmethod
from .base import TensorType
from .experimental.check_shapes import ErrorContext, Shape, check_shapes, register_get_shape
class ProbabilityDistribution(ABC):
"""
This is the base class for a probability distributions,
over which we take the expectations in the expectations framework.
"""
@property
@abstractmethod
def shape(self) -> Shape:
"""
Return the shape of this distribution.
Shape should be some variation of ``[N, D]``, where:
* ``N`` is the number of data points.
* ``D`` is the number of input dimensions.
"""
@register_get_shape(ProbabilityDistribution)
def get_probability_distribution_shape(
shaped: ProbabilityDistribution, context: ErrorContext
) -> Shape:
return shaped.shape
class Gaussian(ProbabilityDistribution):
@check_shapes(
"mu: [N, D]",
"cov: [N, D, D]",
)
def __init__(self, mu: TensorType, cov: TensorType):
self.mu = mu
self.cov = cov
@property
def shape(self) -> Shape:
return self.mu.shape # type: ignore[no-any-return]
class DiagonalGaussian(ProbabilityDistribution):
@check_shapes(
"mu: [N, D]",
"cov: [N, D]",
)
def __init__(self, mu: TensorType, cov: TensorType):
self.mu = mu
self.cov = cov
@property
def shape(self) -> Shape:
return self.mu.shape # type: ignore[no-any-return]
class MarkovGaussian(ProbabilityDistribution):
"""
Gaussian distribution with Markov structure.
Only covariances and covariances between t and t+1 need to be
parameterised. We use the solution proposed by Carl Rasmussen, i.e. to
represent
Var[x_t] = cov[x_t, :, :] * cov[x_t, :, :].T
Cov[x_t, x_{t+1}] = cov[t, :, :] * cov[t+1, :, :]
"""
@check_shapes(
"mu: [N_plus_1, D]",
"cov: [2, N_plus_1, D, D]",
)
def __init__(self, mu: TensorType, cov: TensorType):
self.mu = mu
self.cov = cov
@property
def shape(self) -> Shape:
shape = self.mu.shape
if shape is None:
return shape
N_plus_1, D = shape
N = N_plus_1 - 1
return N, D