gtsam 4.2.0
gtsam
|
A GaussianConditional functions as the node in a Bayes network.
This is the base class for all conditional distributions/densities, which are implemented as specialized factors.
It has a set of parents y,z, etc. and implements a probability density on x. The negative log-density is given by \( \frac{1}{2} |Rx - (d - Sy - Tz - ...)|^2 \)
This class does not store any data other than its keys. Derived classes store data such as matrices and probability tables.
The evaluate
method is used to evaluate the factor, and together with logProbability
is the main methods that need to be implemented in derived classes. These two methods relate to the error
method in the factor by: probability(x) = k exp(-error(x)) where k is a normalization constant making \int probability(x) == 1.0, and logProbability(x) = K - error(x) i.e., K = log(K). The normalization constant K is assumed to not depend on any argument, only (possibly) on the conditional parameters. This class provides a default logNormalizationConstant() == 0.0.
There are four broad classes of conditionals that derive from Conditional:
Constructors | |
GaussianConditional () | |
default constructor needed for serialization | |
GaussianConditional (Key key, const Vector &d, const Matrix &R, const SharedDiagonal &sigmas=SharedDiagonal()) | |
constructor with no parents |Rx-d| | |
GaussianConditional (Key key, const Vector &d, const Matrix &R, Key parent1, const Matrix &S, const SharedDiagonal &sigmas=SharedDiagonal()) | |
constructor with only one parent |Rx+Sy-d| | |
GaussianConditional (Key key, const Vector &d, const Matrix &R, Key parent1, const Matrix &S, Key parent2, const Matrix &T, const SharedDiagonal &sigmas=SharedDiagonal()) | |
constructor with two parents |Rx+Sy+Tz-d| | |
template<typename TERMS > | |
GaussianConditional (const TERMS &terms, size_t nrFrontals, const Vector &d, const SharedDiagonal &sigmas=SharedDiagonal()) | |
Constructor with arbitrary number of frontals and parents. | |
template<typename KEYS > | |
GaussianConditional (const KEYS &keys, size_t nrFrontals, const VerticalBlockMatrix &augmentedMatrix, const SharedDiagonal &sigmas=SharedDiagonal()) | |
Constructor with arbitrary number keys, and where the augmented matrix is given all together instead of in block terms. | |
static GaussianConditional | FromMeanAndStddev (Key key, const Vector &mu, double sigma) |
Construct from mean mu and standard deviation sigma . | |
static GaussianConditional | FromMeanAndStddev (Key key, const Matrix &A, Key parent, const Vector &b, double sigma) |
Construct from conditional mean A1 p1 + b and standard deviation. | |
static GaussianConditional | FromMeanAndStddev (Key key, const Matrix &A1, Key parent1, const Matrix &A2, Key parent2, const Vector &b, double sigma) |
Construct from conditional mean A1 p1 + A2 p2 + b and standard deviation sigma . | |
template<typename... Args> | |
static shared_ptr | sharedMeanAndStddev (Args &&... args) |
Create shared pointer by forwarding arguments to fromMeanAndStddev. | |
template<typename ITERATOR > | |
static shared_ptr | Combine (ITERATOR firstConditional, ITERATOR lastConditional) |
Combine several GaussianConditional into a single dense GC. | |
Public Member Functions | |
Testable | |
void | print (const std::string &="GaussianConditional", const KeyFormatter &formatter=DefaultKeyFormatter) const override |
print | |
bool | equals (const GaussianFactor &cg, double tol=1e-9) const override |
equals function | |
Standard Interface | |
double | logNormalizationConstant () const override |
normalization constant = 1.0 / sqrt((2*pi)^n*det(Sigma)) log = - 0.5 * n*log(2*pi) - 0.5 * log det(Sigma) | |
double | logProbability (const VectorValues &x) const |
Calculate log-probability log(evaluate(x)) for given values x : -error(x) - 0.5 * n*log(2*pi) - 0.5 * log det(Sigma) where x is the vector of values, and Sigma is the covariance matrix. | |
double | evaluate (const VectorValues &x) const |
Calculate probability density for given values x : exp(logProbability(x)) == exp(-GaussianFactor::error(x)) / sqrt((2*pi)^n*det(Sigma)) where x is the vector of values, and Sigma is the covariance matrix. | |
double | operator() (const VectorValues &x) const |
Evaluate probability density, sugar. | |
VectorValues | solve (const VectorValues &parents) const |
Solves a conditional Gaussian and writes the solution into the entries of x for each frontal variable of the conditional. | |
VectorValues | solveOtherRHS (const VectorValues &parents, const VectorValues &rhs) const |
void | solveTransposeInPlace (VectorValues &gy) const |
Performs transpose backsubstition in place on values. | |
JacobianFactor::shared_ptr | likelihood (const VectorValues &frontalValues) const |
Convert to a likelihood factor by providing value before bar. | |
JacobianFactor::shared_ptr | likelihood (const Vector &frontal) const |
Single variable version of likelihood. | |
VectorValues | sample (std::mt19937_64 *rng) const |
Sample from conditional, zero parent version Example: std::mt19937_64 rng(42); auto sample = gbn.sample(&rng);. | |
VectorValues | sample (const VectorValues &parentsValues, std::mt19937_64 *rng) const |
Sample from conditional, given missing variables Example: std::mt19937_64 rng(42); VectorValues given = ...; auto sample = gbn.sample(given, &rng);. | |
VectorValues | sample () const |
Sample, use default rng. | |
VectorValues | sample (const VectorValues &parentsValues) const |
Sample with given values, use default rng. | |
Linear algebra. | |
constABlock | R () const |
Return a view of the upper-triangular R block of the conditional. | |
constABlock | S () const |
Get a view of the parent blocks. | |
constABlock | S (const_iterator it) const |
Get a view of the S matrix for the variable pointed to by the given key iterator. | |
const constBVector | d () const |
Get a view of the r.h.s. | |
double | determinant () const |
Compute the determinant of the R matrix. | |
double | logDeterminant () const |
Compute the log determinant of the R matrix. | |
HybridValues methods. | |
double | logProbability (const HybridValues &x) const override |
Calculate log-probability log(evaluate(x)) for HybridValues x . | |
double | evaluate (const HybridValues &x) const override |
Calculate probability for HybridValues x . | |
double | error (const VectorValues &c) const override |
![]() | |
JacobianFactor (const GaussianFactor &gf) | |
Convert from other GaussianFactor. | |
JacobianFactor (const JacobianFactor &jf) | |
Copy constructor. | |
JacobianFactor (const HessianFactor &hf) | |
Conversion from HessianFactor (does Cholesky to obtain Jacobian matrix) | |
JacobianFactor () | |
default constructor for I/O | |
JacobianFactor (const Vector &b_in) | |
Construct Null factor. | |
JacobianFactor (Key i1, const Matrix &A1, const Vector &b, const SharedDiagonal &model=SharedDiagonal()) | |
Construct unary factor. | |
JacobianFactor (Key i1, const Matrix &A1, Key i2, const Matrix &A2, const Vector &b, const SharedDiagonal &model=SharedDiagonal()) | |
Construct binary factor. | |
JacobianFactor (Key i1, const Matrix &A1, Key i2, const Matrix &A2, Key i3, const Matrix &A3, const Vector &b, const SharedDiagonal &model=SharedDiagonal()) | |
Construct ternary factor. | |
template<typename TERMS > | |
JacobianFactor (const TERMS &terms, const Vector &b, const SharedDiagonal &model=SharedDiagonal()) | |
Construct an n-ary factor. | |
template<typename KEYS > | |
JacobianFactor (const KEYS &keys, const VerticalBlockMatrix &augmentedMatrix, const SharedDiagonal &sigmas=SharedDiagonal()) | |
Constructor with arbitrary number keys, and where the augmented matrix is given all together instead of in block terms. | |
JacobianFactor (const GaussianFactorGraph &graph) | |
Build a dense joint factor from all the factors in a factor graph. | |
JacobianFactor (const GaussianFactorGraph &graph, const VariableSlots &p_variableSlots) | |
Build a dense joint factor from all the factors in a factor graph. | |
JacobianFactor (const GaussianFactorGraph &graph, const Ordering &ordering) | |
Build a dense joint factor from all the factors in a factor graph. | |
JacobianFactor (const GaussianFactorGraph &graph, const Ordering &ordering, const VariableSlots &p_variableSlots) | |
Build a dense joint factor from all the factors in a factor graph. | |
~JacobianFactor () override | |
Virtual destructor. | |
GaussianFactor::shared_ptr | clone () const override |
Clone this JacobianFactor. | |
void | print (const std::string &s="", const KeyFormatter &formatter=DefaultKeyFormatter) const override |
print | |
bool | equals (const GaussianFactor &lf, double tol=1e-9) const override |
Equals for testable. | |
Vector | unweighted_error (const VectorValues &c) const |
Vector | error_vector (const VectorValues &c) const |
(A*x-b) | |
double | error (const VectorValues &c) const override |
Matrix | augmentedInformation () const override |
Return the augmented information matrix represented by this GaussianFactor. | |
Matrix | information () const override |
Return the non-augmented information matrix represented by this GaussianFactor. | |
void | hessianDiagonalAdd (VectorValues &d) const override |
Add the current diagonal to a VectorValues instance. | |
void | hessianDiagonal (double *d) const override |
Raw memory access version of hessianDiagonal. | |
std::map< Key, Matrix > | hessianBlockDiagonal () const override |
Return the block diagonal of the Hessian for this factor. | |
std::pair< Matrix, Vector > | jacobian () const override |
Returns (dense) A,b pair associated with factor, bakes in the weights. | |
std::pair< Matrix, Vector > | jacobianUnweighted () const |
Returns (dense) A,b pair associated with factor, does not bake in weights. | |
Matrix | augmentedJacobian () const override |
Return (dense) matrix associated with factor. | |
Matrix | augmentedJacobianUnweighted () const |
Return (dense) matrix associated with factor. | |
const VerticalBlockMatrix & | matrixObject () const |
Return the full augmented Jacobian matrix of this factor as a VerticalBlockMatrix object. | |
VerticalBlockMatrix & | matrixObject () |
Mutable access to the full augmented Jacobian matrix of this factor as a VerticalBlockMatrix object. | |
GaussianFactor::shared_ptr | negate () const override |
Construct the corresponding anti-factor to negate information stored stored in this factor. | |
bool | isConstrained () const |
is noise model constrained ? | |
DenseIndex | getDim (const_iterator variable) const override |
Return the dimension of the variable pointed to by the given key iterator todo: Remove this in favor of keeping track of dimensions with variables? | |
size_t | rows () const |
return the number of rows in the corresponding linear system | |
size_t | cols () const |
return the number of columns in the corresponding linear system | |
const SharedDiagonal & | get_model () const |
get a copy of model | |
SharedDiagonal & | get_model () |
get a copy of model (non-const version) | |
const constBVector | getb () const |
Get a view of the r.h.s. | |
constABlock | getA (const_iterator variable) const |
Get a view of the A matrix for the variable pointed to by the given key iterator. | |
constABlock | getA () const |
Get a view of the A matrix, not weighted by noise. | |
BVector | getb () |
Get a view of the r.h.s. | |
ABlock | getA (iterator variable) |
Get a view of the A matrix for the variable pointed to by the given key iterator (non-const version) | |
ABlock | getA () |
Get a view of the A matrix. | |
void | updateHessian (const KeyVector &keys, SymmetricBlockMatrix *info) const override |
Update an information matrix by adding the information corresponding to this factor (used internally during elimination). | |
Vector | operator* (const VectorValues &x) const |
Return A*x. | |
void | transposeMultiplyAdd (double alpha, const Vector &e, VectorValues &x) const |
x += alpha * A'*e. | |
void | multiplyHessianAdd (double alpha, const VectorValues &x, VectorValues &y) const override |
y += alpha * A'*A*x | |
void | multiplyHessianAdd (double alpha, const double *x, double *y, const std::vector< size_t > &accumulatedDims) const |
Raw memory access version of multiplyHessianAdd y += alpha * A'*A*x Requires the vector accumulatedDims to tell the dimension of each variable: e.g.: x0 has dim 3, x2 has dim 6, x3 has dim 2, then accumulatedDims is [0 3 9 11 13] NOTE: size of accumulatedDims is size of keys + 1!! TODO(frank): we should probably kill this if no longer needed. | |
VectorValues | gradientAtZero () const override |
A'*b for Jacobian. | |
void | gradientAtZero (double *d) const override |
A'*b for Jacobian (raw memory version) | |
Vector | gradient (Key key, const VectorValues &x) const override |
Compute the gradient wrt a key at any values. | |
JacobianFactor | whiten () const |
Return a whitened version of the factor, i.e. | |
std::pair< boost::shared_ptr< GaussianConditional >, shared_ptr > | eliminate (const Ordering &keys) |
Eliminate the requested variables. | |
void | setModel (bool anyConstrained, const Vector &sigmas) |
set noiseModel correctly | |
boost::shared_ptr< GaussianConditional > | splitConditional (size_t nrFrontals) |
splits a pre-factorized factor into a conditional, and changes the current factor to be the remaining component. | |
virtual double | error (const VectorValues &c) const |
double | error (const HybridValues &c) const override |
All factor types need to implement an error function. | |
VectorValues | hessianDiagonal () const |
Using the base method. | |
virtual void | hessianDiagonal (double *d) const=0 |
Using the base method. | |
![]() | |
GaussianFactor () | |
Default constructor creates empty factor. | |
template<typename CONTAINER > | |
GaussianFactor (const CONTAINER &keys) | |
Construct from container of keys. | |
virtual | ~GaussianFactor () |
Destructor. | |
VectorValues | hessianDiagonal () const |
Return the diagonal of the Hessian for this factor. | |
![]() | |
virtual | ~Factor ()=default |
Default destructor. | |
bool | empty () const |
Whether the factor is empty (involves zero variables). | |
Key | front () const |
First key. | |
Key | back () const |
Last key. | |
const_iterator | find (Key key) const |
find | |
const KeyVector & | keys () const |
Access the factor's involved variable keys. | |
const_iterator | begin () const |
Iterator at beginning of involved variable keys. | |
const_iterator | end () const |
Iterator at end of involved variable keys. | |
size_t | size () const |
virtual void | printKeys (const std::string &s="Factor", const KeyFormatter &formatter=DefaultKeyFormatter) const |
print only keys | |
bool | equals (const This &other, double tol=1e-9) const |
check equality | |
KeyVector & | keys () |
iterator | begin () |
Iterator at beginning of involved variable keys. | |
iterator | end () |
Iterator at end of involved variable keys. | |
![]() | |
void | print (const std::string &s="Conditional", const KeyFormatter &formatter=DefaultKeyFormatter) const |
print with optional formatter | |
bool | equals (const This &c, double tol=1e-9) const |
check equality | |
size_t | nrFrontals () const |
return the number of frontals | |
size_t | nrParents () const |
return the number of parents | |
Key | firstFrontalKey () const |
Convenience function to get the first frontal key. | |
Frontals | frontals () const |
return a view of the frontal keys | |
Parents | parents () const |
return a view of the parent keys | |
double | operator() (const HybridValues &x) const |
Evaluate probability density, sugar. | |
double | normalizationConstant () const |
Non-virtual, exponentiate logNormalizationConstant. | |
size_t & | nrFrontals () |
Mutable version of nrFrontals. | |
FACTOR::const_iterator | beginFrontals () const |
Iterator pointing to first frontal key. | |
FACTOR::iterator | beginFrontals () |
Mutable iterator pointing to first frontal key. | |
FACTOR::const_iterator | endFrontals () const |
Iterator pointing past the last frontal key. | |
FACTOR::iterator | endFrontals () |
Mutable iterator pointing past the last frontal key. | |
FACTOR::const_iterator | beginParents () const |
Iterator pointing to the first parent key. | |
FACTOR::iterator | beginParents () |
Mutable iterator pointing to the first parent key. | |
FACTOR::const_iterator | endParents () const |
Iterator pointing past the last parent key. | |
FACTOR::iterator | endParents () |
Mutable iterator pointing past the last parent key. | |
Public Types | |
typedef GaussianConditional | This |
Typedef to this class. | |
typedef boost::shared_ptr< This > | shared_ptr |
shared_ptr to this class | |
typedef JacobianFactor | BaseFactor |
Typedef to our factor base class. | |
typedef Conditional< BaseFactor, This > | BaseConditional |
Typedef to our conditional base class. | |
![]() | |
typedef JacobianFactor | This |
Typedef to this class. | |
typedef GaussianFactor | Base |
Typedef to base class. | |
typedef boost::shared_ptr< This > | shared_ptr |
shared_ptr to this class | |
typedef VerticalBlockMatrix::Block | ABlock |
typedef VerticalBlockMatrix::constBlock | constABlock |
typedef ABlock::ColXpr | BVector |
typedef constABlock::ConstColXpr | constBVector |
![]() | |
typedef GaussianFactor | This |
This class. | |
typedef boost::shared_ptr< This > | shared_ptr |
shared_ptr to this class | |
typedef Factor | Base |
Our base class. | |
![]() | |
typedef KeyVector::iterator | iterator |
Iterator over keys. | |
typedef KeyVector::const_iterator | const_iterator |
Const iterator over keys. | |
![]() | |
typedef boost::iterator_range< typename FACTOR::const_iterator > | Frontals |
View of the frontal keys (call frontals()) | |
typedef boost::iterator_range< typename FACTOR::const_iterator > | Parents |
View of the separator keys (call parents()) | |
Friends | |
class | boost::serialization::access |
Serialization function. | |
Additional Inherited Members | |
![]() | |
template<typename CONTAINER > | |
static DenseIndex | Slot (const CONTAINER &keys, Key key) |
![]() | |
static bool | CheckInvariants (const GaussianConditional &conditional, const VALUES &x) |
Check invariants of this conditional, given the values x . | |
![]() | |
template<typename TERMS > | |
void | fillTerms (const TERMS &terms, const Vector &b, const SharedDiagonal &noiseModel) |
Internal function to fill blocks and set dimensions. | |
![]() | |
Factor () | |
Default constructor for I/O. | |
template<typename CONTAINER > | |
Factor (const CONTAINER &keys) | |
Construct factor from container of keys. | |
template<typename ITERATOR > | |
Factor (ITERATOR first, ITERATOR last) | |
Construct factor from iterator keys. | |
![]() | |
Conditional () | |
Empty Constructor to make serialization possible. | |
Conditional (size_t nrFrontals) | |
Constructor. | |
![]() | |
template<typename CONTAINER > | |
static Factor | FromKeys (const CONTAINER &keys) |
Construct factor from container of keys. | |
template<typename ITERATOR > | |
static Factor | FromIterators (ITERATOR first, ITERATOR last) |
Construct factor from iterator keys. | |
![]() | |
VerticalBlockMatrix | Ab_ |
noiseModel::Diagonal::shared_ptr | model_ |
![]() | |
KeyVector | keys_ |
The keys involved in this factor. | |
![]() | |
size_t | nrFrontals_ |
The first nrFrontal variables are frontal and the rest are parents. | |
GaussianConditional::GaussianConditional | ( | const TERMS & | terms, |
size_t | nrFrontals, | ||
const Vector & | d, | ||
const SharedDiagonal & | sigmas = SharedDiagonal() |
||
) |
Constructor with arbitrary number of frontals and parents.
TERMS | A container whose value type is std::pair<Key, Matrix>, specifying the collection of keys and matrices making up the conditional. |
GaussianConditional::GaussianConditional | ( | const KEYS & | keys, |
size_t | nrFrontals, | ||
const VerticalBlockMatrix & | augmentedMatrix, | ||
const SharedDiagonal & | sigmas = SharedDiagonal() |
||
) |
Constructor with arbitrary number keys, and where the augmented matrix is given all together instead of in block terms.
Note that only the active view of the provided augmented matrix is used, and that the matrix data is copied into a newly-allocated matrix in the constructed factor.
|
static |
Combine several GaussianConditional into a single dense GC.
The conditionals enumerated by first
and last
must be in increasing order, meaning that the parents of any conditional may not include a conditional coming before it.
firstConditional | Iterator to the first conditional to combine, must dereference to a shared_ptr<GaussianConditional>. |
lastConditional | Iterator to after the last conditional to combine, must dereference to a shared_ptr<GaussianConditional>. |
|
inline |
Get a view of the r.h.s.
vector d
|
inline |
Compute the determinant of the R matrix.
The determinant is computed in log form using logDeterminant for numerical stability and then exponentiated.
Note, the covariance matrix \( \Sigma = (R^T R)^{-1} \), and hence \( \det(\Sigma) = 1 / \det(R^T R) = 1 / determinant()^ 2 \).
|
overridevirtual |
equals function
Implements gtsam::GaussianFactor.
|
overridevirtual |
Reimplemented from gtsam::GaussianFactor.
|
overridevirtual |
Calculate probability for HybridValues x
.
Simply dispatches to VectorValues version.
Reimplemented from gtsam::Conditional< JacobianFactor, GaussianConditional >.
double GaussianConditional::logDeterminant | ( | ) | const |
Compute the log determinant of the R matrix.
For numerical stability, the determinant is computed in log form, so it is a summation rather than a multiplication.
Note, the covariance matrix \( \Sigma = (R^T R)^{-1} \), and hence \( \log \det(\Sigma) = - \log \det(R^T R) = - 2 logDeterminant() \).
|
overridevirtual |
normalization constant = 1.0 / sqrt((2*pi)^n*det(Sigma)) log = - 0.5 * n*log(2*pi) - 0.5 * log det(Sigma)
Reimplemented from gtsam::Conditional< JacobianFactor, GaussianConditional >.
|
overridevirtual |
Calculate log-probability log(evaluate(x)) for HybridValues x
.
Simply dispatches to VectorValues version.
Reimplemented from gtsam::Conditional< JacobianFactor, GaussianConditional >.
double GaussianConditional::logProbability | ( | const VectorValues & | x | ) | const |
Calculate log-probability log(evaluate(x)) for given values x
: -error(x) - 0.5 * n*log(2*pi) - 0.5 * log det(Sigma) where x is the vector of values, and Sigma is the covariance matrix.
This differs from error as it is log, not negative log, and it includes the normalization constant.
|
overridevirtual |
VectorValues GaussianConditional::solve | ( | const VectorValues & | parents | ) | const |
Solves a conditional Gaussian and writes the solution into the entries of x
for each frontal variable of the conditional.
The parents are assumed to have already been solved in and their values are read from x
. This function works for multiple frontal variables.
Given the Gaussian conditional with log likelihood \( |R x_f - (d - S x_s)|^2 \), where \( f \) are the frontal variables and \( s \) are the separator variables of this conditional, this solve function computes \( x_f = R^{-1} (d - S x_s) \) using back-substitution.
parents | VectorValues containing solved parents \( x_s \). |