ASIRFilter< StateVar, MeasVar > Class Template Reference

ASIR: Auxiliary Particle Filter. More...

#include <asirfilter.h>

Inheritance diagram for ASIRFilter< StateVar, MeasVar >:

ParticleFilter< StateVar, MeasVar > Filter< StateVar,MeasVar >

List of all members.

Public Member Functions

 ASIRFilter (MCPdf< StateVar > *prior, int resampleperiod=0, double resamplethreshold=0, int resamplescheme=DEFAULT_RS)
 Constructor.
virtual ~ASIRFilter ()
 Destructor.
virtual void ProposalSet (ConditionalPdf< StateVar, StateVar > *const cpdf)
 Set the proposal density.
ConditionalPdf< StateVar,
StateVar > * 
ProposalGet ()
 Get a pointer to the proposal density.
virtual MCPdf< StateVar > * PostGet ()
 Get Posterior density.
virtual void Reset (Pdf< StateVar > *prior)
 Reset Filter.
virtual bool Update (SystemModel< StateVar > *const sysmodel, const StateVar &u, MeasurementModel< MeasVar, StateVar > *const measmodel, const MeasVar &z, const StateVar &s)
 Full Update (system with inputs/sensing params).
virtual bool Update (SystemModel< StateVar > *const sysmodel, MeasurementModel< MeasVar, StateVar > *const measmodel, const MeasVar &z, const StateVar &s)
 Full Update (system without inputs, with sensing params).
virtual bool Update (SystemModel< StateVar > *const sysmodel, MeasurementModel< MeasVar, StateVar > *const measmodel, const MeasVar &z)
 Full Update (system without inputs/sensing params).
virtual bool Update (SystemModel< StateVar > *const sysmodel, const StateVar &u, MeasurementModel< MeasVar, StateVar > *const measmodel, const MeasVar &z)
 Full Update (system with inputs, without sensing params).
virtual bool Update (SystemModel< StateVar > *const sysmodel, const StateVar &u)
 System Update (system with inputs).
virtual bool Update (SystemModel< StateVar > *const sysmodel)
 System Update (system without inputs).
virtual bool Update (MeasurementModel< MeasVar, StateVar > *const measmodel, const MeasVar &z, const StateVar &s)
 Measurement Update (system with "sensing params").
virtual bool Update (MeasurementModel< MeasVar, StateVar > *const measmodel, const MeasVar &z)
 Measurement Update (system without "sensing params").
int TimeStepGet () const
 Get current time.

Protected Member Functions

virtual void UpdateInternal (SystemModel< StateVar > *const sysmodel, const StateVar &u, MeasurementModel< MeasVar, StateVar > *const measmodel, const MeasVar &z, const StateVar &s)
 Actual implementation of updateinternal.
virtual bool ProposalStepInternal (SystemModel< StateVar > *const sysmodel, const StateVar &u, MeasurementModel< MeasVar, StateVar > *const measmodel, const MeasVar &z, const StateVar &s)
 Proposal step.
virtual bool UpdateWeightsInternal (SystemModel< StateVar > *const sysmodel, const StateVar &u, MeasurementModel< MeasVar, StateVar > *const measmodel, const MeasVar &z, const StateVar &s)
 Update Weights.
virtual bool DynamicResampleStep ()
 Resample if necessary.
virtual bool StaticResampleStep ()
 Resample if wanted.
virtual bool Resample ()
 Actual Resampling happens here;.

Protected Attributes

ConditionalPdf< StateVar,
StateVar > * 
_proposal
 Pointer to the Proposal Density.
WeightedSample< StateVar > _sample
 While updating use sample<StateVar>.
vector< WeightedSample
< StateVar > > 
_old_samples
 While updating store list of old samples.
vector< WeightedSample
< StateVar > > 
_new_samples
 While updating store list of new samples.
vector< Sample< StateVar > > _new_samples_unweighted
 While resampling.
vector< WeightedSample
< StateVar > >::iterator 
_os_it
 Iterator for old list of samples.
vector< WeightedSample
< StateVar > >::iterator 
_ns_it
 Iterator for new list of samples.
int _resamplePeriod
 Number of timestep between resampling from the Posterior Pdf.
double _resampleThreshold
 Threshold used when dynamic resampling.
int _resampleScheme
 Which resample algorithm (see top of particle.h for defines).
bool _dynamicResampling
 Dynamic resampling or fixed period resampling?
bool _proposal_depends_on_meas
 Proposal depends on last measurement?
bool _created_post
 created own post
Pdf< StateVar > * _prior
 prior Pdf
Pdf< StateVar > * _post
 Pointer to the Posterior Pdf.
int _timestep
 Represents the current timestep of the filter.


Detailed Description

template<typename StateVar, typename MeasVar>
class BFL::ASIRFilter< StateVar, MeasVar >

ASIR: Auxiliary Particle Filter.

This is a possible implementation of a particle filter, which in some/many cases will yield better results, since measures are taken to make the proposal density more similar to the posterior. See

      @Article{	  pitt_auxiliary,
      author	= {Pitt, M. and Shephard, N.},
      title	= {Filtering via simulation: auxiliary particle filter},
      journal	= {Journal of the American Statistical Association},
      year	= {1999},
      note	= {forthcoming}
      }
      

for more details.

Note that this particular implementation:

Particular issue with of proposalstep in case of ASIR filter... The current implementation uses the approximation (see p. 11 of the Pitt and Shephard paper, we use their notation here ---

\[ \alpha \]

denoting the state and

\[ y \]

denoting the measurement)

\[ f(y_{t+1} | \alpha_{t+1}) \approx f(y_{t+1} | \mu_{t+1}) \]

where

\[ \mu \]

is the mean value of the system pdf.

Note that the ASIR needs the measurementmodel for its proposalstep, to obtain a better proposal. Note also that normally, the ASIR performs better that the standard SIR filter in case of outliers, but worse for "normal data" (due to the extra resampling stage).

Definition at line 79 of file asirfilter.h.


Constructor & Destructor Documentation

ASIRFilter ( MCPdf< StateVar > *  prior,
int  resampleperiod = 0,
double  resamplethreshold = 0,
int  resamplescheme = DEFAULT_RS 
)

Constructor.

Precondition:
you created the necessary models and the prior
Parameters:
prior pointer to the Monte Carlo Pdf prior density
resampleperiod fixed resampling period (if desired)
resamplethreshold threshold used when dynamic resampling
resamplescheme resampling scheme, see header file for different defines and their meaning


Member Function Documentation

virtual bool DynamicResampleStep (  )  [protected, virtual, inherited]

Resample if necessary.

Bug:
let the user implement her/his own resamplescheme

virtual MCPdf<StateVar >* PostGet (  )  [virtual, inherited]

Get Posterior density.

Get the current Posterior density

Returns:
a pointer to the current posterior

Reimplemented from Filter< StateVar,MeasVar >.

ConditionalPdf<StateVar ,StateVar >* ProposalGet (  )  [inherited]

Get a pointer to the proposal density.

Returns:
a pointer to the proposal density

virtual void ProposalSet ( ConditionalPdf< StateVar , StateVar > *const  cpdf  )  [virtual, inherited]

Set the proposal density.

Parameters:
cpdf the new proposal density. The order of the conditional arguments is fixed and should be: x (state), u (input), z (measurement), s (sensor param). Off course all of them are optional

virtual bool ProposalStepInternal ( SystemModel< StateVar > *const  sysmodel,
const StateVar &  u,
MeasurementModel< MeasVar , StateVar > *const  measmodel,
const MeasVar &  z,
const StateVar &  s 
) [protected, virtual, inherited]

Proposal step.

Implementation of proposal step

Parameters:
sysmodel pointer to the used system model
u input param for proposal density
measmodel pointer to the used measurementmodel
z measurement param for proposal density
s sensor param for proposal density
Bug:
Make sampling method variable. See implementation.

virtual bool StaticResampleStep (  )  [protected, virtual, inherited]

Resample if wanted.

Bug:
let the user implement her/his own resamplescheme

int TimeStepGet (  )  const [inherited]

Get current time.

Get the current time of the filter

Returns:
the current timestep

virtual bool Update ( MeasurementModel< MeasVar , StateVar > *const  measmodel,
const MeasVar &  z 
) [virtual, inherited]

Measurement Update (system without "sensing params").

Parameters:
measmodel pointer to the measurement model to use for update
z measurement

Reimplemented in Optimalimportancefilter< StateVar, MeasVar >.

virtual bool Update ( MeasurementModel< MeasVar , StateVar > *const  measmodel,
const MeasVar &  z,
const StateVar &  s 
) [virtual, inherited]

Measurement Update (system with "sensing params").

Parameters:
measmodel pointer to the measurement model to use for update
z measurement
s "sensing parameter"

Reimplemented in Optimalimportancefilter< StateVar, MeasVar >.

virtual bool Update ( SystemModel< StateVar > *const  sysmodel  )  [virtual, inherited]

System Update (system without inputs).

Parameters:
sysmodel pointer to the system model to use for update

Reimplemented in Optimalimportancefilter< StateVar, MeasVar >.

virtual bool Update ( SystemModel< StateVar > *const  sysmodel,
const StateVar &  u 
) [virtual, inherited]

System Update (system with inputs).

Parameters:
sysmodel pointer to the system model to use for update
u input to the system

Reimplemented in Optimalimportancefilter< StateVar, MeasVar >.

virtual bool Update ( SystemModel< StateVar > *const  sysmodel,
const StateVar &  u,
MeasurementModel< MeasVar , StateVar > *const  measmodel,
const MeasVar &  z 
) [virtual, inherited]

Full Update (system with inputs, without sensing params).

Parameters:
sysmodel pointer to the system model to use for update
u input to the system
measmodel pointer to the measurement model to use for update
z measurement

Reimplemented in Optimalimportancefilter< StateVar, MeasVar >.

virtual bool Update ( SystemModel< StateVar > *const  sysmodel,
MeasurementModel< MeasVar , StateVar > *const  measmodel,
const MeasVar &  z 
) [virtual, inherited]

Full Update (system without inputs/sensing params).

Parameters:
sysmodel pointer to the system model to use for update
measmodel pointer to the measurement model to use for update
z measurement

Reimplemented in Optimalimportancefilter< StateVar, MeasVar >.

virtual bool Update ( SystemModel< StateVar > *const  sysmodel,
MeasurementModel< MeasVar , StateVar > *const  measmodel,
const MeasVar &  z,
const StateVar &  s 
) [virtual, inherited]

Full Update (system without inputs, with sensing params).

Parameters:
sysmodel pointer to the system model to use for update
measmodel pointer to the measurement model to use for update
z measurement
s "sensing parameter"

Reimplemented in Optimalimportancefilter< StateVar, MeasVar >.

virtual bool Update ( SystemModel< StateVar > *const  sysmodel,
const StateVar &  u,
MeasurementModel< MeasVar , StateVar > *const  measmodel,
const MeasVar &  z,
const StateVar &  s 
) [virtual, inherited]

Full Update (system with inputs/sensing params).

Parameters:
sysmodel pointer to the system model to use for update
u input to the system
measmodel pointer to the measurement model to use for update
z measurement
s "sensing parameter"

Reimplemented in Optimalimportancefilter< StateVar, MeasVar >.

virtual bool UpdateWeightsInternal ( SystemModel< StateVar > *const  sysmodel,
const StateVar &  u,
MeasurementModel< MeasVar , StateVar > *const  measmodel,
const MeasVar &  z,
const StateVar &  s 
) [protected, virtual, inherited]

Update Weights.

Parameters:
sysmodel pointer to the used system model
u input param for proposal density
measmodel pointer to the used measurementmodel
z measurement param for proposal density
s sensor param for proposal density


Member Data Documentation

Pdf<StateVar >* _post [protected, inherited]

Pointer to the Posterior Pdf.

The Posterior Pdf represents the subjective belief of the person applying the filter AFTER processing inputs and measurements. A filter does not maintain the beliefs at all timesteps t, since this leads to non-constant (or ever growing if you prefer) memory requirements. However, it is possible, to copy the Posterior density at all timesteps in your application by means of the PostGet() member function

See also:
PostGet()

Definition at line 95 of file filter.h.

ConditionalPdf<StateVar ,StateVar >* _proposal [protected, inherited]

Pointer to the Proposal Density.

Every particle filter (or more correct: every Sequential Importance Sampling method) uses a proposal density to do the forward sampling step

Definition at line 174 of file particlefilter.h.

int _resamplePeriod [protected, inherited]

Number of timestep between resampling from the Posterior Pdf.

By choosing this period, one can avoid numerical instability (aka Degeneration of the particle filter

Definition at line 193 of file particlefilter.h.

int _timestep [protected, inherited]

Represents the current timestep of the filter.

Todo:
Check wether this really belongs here

Definition at line 100 of file filter.h.


The documentation for this class was generated from the following file:

Generated on Thu Mar 24 16:57:09 2011 for Bayesian Filtering Library by  doxygen 1.5.9