gtsam 4.1.1
gtsam
gtsam::ISAM2Params Struct Reference

Public Member Functions

 ISAM2Params (OptimizationParams _optimizationParams=ISAM2GaussNewtonParams(), RelinearizationThreshold _relinearizeThreshold=0.1, int _relinearizeSkip=10, bool _enableRelinearization=true, bool _evaluateNonlinearError=false, Factorization _factorization=ISAM2Params::CHOLESKY, bool _cacheLinearizedFactors=true, const KeyFormatter &_keyFormatter=DefaultKeyFormatter, bool _enableDetailedResults=false)
 Specify parameters as constructor arguments See the documentation of member variables above. More...
 
void print (const std::string &str="") const
 print iSAM2 parameters
 
Getters and Setters for all properties
OptimizationParams getOptimizationParams () const
 
RelinearizationThreshold getRelinearizeThreshold () const
 
int getRelinearizeSkip () const
 
bool isEnableRelinearization () const
 
bool isEvaluateNonlinearError () const
 
std::string getFactorization () const
 
bool isCacheLinearizedFactors () const
 
KeyFormatter getKeyFormatter () const
 
bool isEnableDetailedResults () const
 
bool isEnablePartialRelinearizationCheck () const
 
void setOptimizationParams (OptimizationParams optimizationParams)
 
void setRelinearizeThreshold (RelinearizationThreshold relinearizeThreshold)
 
void setRelinearizeSkip (int relinearizeSkip)
 
void setEnableRelinearization (bool enableRelinearization)
 
void setEvaluateNonlinearError (bool evaluateNonlinearError)
 
void setFactorization (const std::string &factorization)
 
void setCacheLinearizedFactors (bool cacheLinearizedFactors)
 
void setKeyFormatter (KeyFormatter keyFormatter)
 
void setEnableDetailedResults (bool enableDetailedResults)
 
void setEnablePartialRelinearizationCheck (bool enablePartialRelinearizationCheck)
 
GaussianFactorGraph::Eliminate getEliminationFunction () const
 

Static Public Member Functions

Some utilities
static Factorization factorizationTranslator (const std::string &str)
 
static std::string factorizationTranslator (const Factorization &value)
 

Public Attributes

OptimizationParams optimizationParams
 Optimization parameters, this both selects the nonlinear optimization method and specifies its parameters, either ISAM2GaussNewtonParams or ISAM2DoglegParams. More...
 
RelinearizationThreshold relinearizeThreshold
 Only relinearize variables whose linear delta magnitude is greater than this threshold (default: 0.1). More...
 
int relinearizeSkip
 Only relinearize any variables every relinearizeSkip calls to ISAM2::update (default: 10)
 
bool enableRelinearization
 Controls whether ISAM2 will ever relinearize any variables (default: true)
 
bool evaluateNonlinearError
 Whether to evaluate the nonlinear error before and after the update, to return in ISAM2Result from update()
 
Factorization factorization
 Specifies whether to use QR or CHOESKY numerical factorization (default: CHOLESKY). More...
 
bool cacheLinearizedFactors
 Whether to cache linear factors (default: true). More...
 
KeyFormatter keyFormatter
 A KeyFormatter for when keys are printed during debugging (default: DefaultKeyFormatter)
 
bool enableDetailedResults
 Whether to compute and return ISAM2Result::detailedResults, this can increase running time (default: false)
 
bool enablePartialRelinearizationCheck
 Check variables for relinearization in tree-order, stopping the check once a variable does not need to be relinearized (default: false). More...
 
bool findUnusedFactorSlots
 When you will be removing many factors, e.g. More...
 

Public Types

enum  Factorization { CHOLESKY , QR }
 
typedef boost::variant< ISAM2GaussNewtonParams, ISAM2DoglegParamsOptimizationParams
 Either ISAM2GaussNewtonParams or ISAM2DoglegParams.
 
typedef boost::variant< double, FastMap< char, Vector > > RelinearizationThreshold
 Either a constant relinearization threshold or a per-variable-type set of thresholds.
 

Constructor & Destructor Documentation

◆ ISAM2Params()

gtsam::ISAM2Params::ISAM2Params ( OptimizationParams  _optimizationParams = ISAM2GaussNewtonParams(),
RelinearizationThreshold  _relinearizeThreshold = 0.1,
int  _relinearizeSkip = 10,
bool  _enableRelinearization = true,
bool  _evaluateNonlinearError = false,
Factorization  _factorization = ISAM2Params::CHOLESKY,
bool  _cacheLinearizedFactors = true,
const KeyFormatter _keyFormatter = DefaultKeyFormatter,
bool  _enableDetailedResults = false 
)
inline

Specify parameters as constructor arguments See the documentation of member variables above.

Parameters
_keyFormattersee ISAM2::Params::keyFormatter,

Member Data Documentation

◆ cacheLinearizedFactors

bool gtsam::ISAM2Params::cacheLinearizedFactors

Whether to cache linear factors (default: true).

This can improve performance if linearization is expensive, but can hurt performance if linearization is very cleap due to computation to look up additional keys.

◆ enablePartialRelinearizationCheck

bool gtsam::ISAM2Params::enablePartialRelinearizationCheck

Check variables for relinearization in tree-order, stopping the check once a variable does not need to be relinearized (default: false).

This can improve speed by only checking a small part of the top of the tree. However, variables below the check cut-off can accumulate significant deltas without triggering relinearization. This is particularly useful in exploration scenarios where real-time performance is desired over correctness. Use with caution.

◆ factorization

Factorization gtsam::ISAM2Params::factorization

Specifies whether to use QR or CHOESKY numerical factorization (default: CHOLESKY).

Cholesky is faster but potentially numerically unstable for poorly-conditioned problems, which can occur when uncertainty is very low in some variables (or dimensions of variables) and very high in others. QR is slower but more numerically stable in poorly-conditioned problems. We suggest using the default of Cholesky unless gtsam sometimes throws IndefiniteLinearSystemException when your problem's Hessian is actually positive definite. For positive definite problems, numerical error accumulation can cause the problem to become numerically negative or indefinite as solving proceeds, especially when using Cholesky.

◆ findUnusedFactorSlots

bool gtsam::ISAM2Params::findUnusedFactorSlots

When you will be removing many factors, e.g.

when using ISAM2 as a fixed-lag smoother, enable this option to add factors in the first available factor slots, to avoid accumulating nullptr factor slots, at the cost of having to search for slots every time a factor is added.

◆ optimizationParams

OptimizationParams gtsam::ISAM2Params::optimizationParams

Optimization parameters, this both selects the nonlinear optimization method and specifies its parameters, either ISAM2GaussNewtonParams or ISAM2DoglegParams.

In the former, Gauss-Newton optimization will be used with the specified parameters, and in the latter Powell's dog-leg algorithm will be used with the specified parameters.

◆ relinearizeThreshold

RelinearizationThreshold gtsam::ISAM2Params::relinearizeThreshold

Only relinearize variables whose linear delta magnitude is greater than this threshold (default: 0.1).

If this is a FastMap<char,Vector> instead of a double, then the threshold is specified for each dimension of each variable type. This parameter then maps from a character indicating the variable type to a Vector of thresholds for each dimension of that variable. For example, if Pose keys are of type TypedSymbol<'x',Pose3>, and landmark keys are of type TypedSymbol<'l',Point3>, then appropriate entries would be added with:

FastMap<char,Vector> thresholds;
thresholds['x'] = (Vector(6) << 0.1, 0.1, 0.1, 0.5, 0.5, 0.5).finished();
// 0.1 rad rotation threshold, 0.5 m translation threshold thresholds['l'] =
Vector3(1.0, 1.0, 1.0); // 1.0 m landmark position threshold
params.relinearizeThreshold = thresholds;

The documentation for this struct was generated from the following files: