|
Size: 1364
Comment:
|
← Revision 4 as of 2014-09-08 19:45:49 ⇥
Size: 1372
Comment:
|
| Deletions are marked like this. | Additions are marked like this. |
| Line 16: | Line 16: |
| * '''distribution_approximation''' = ['''cubature'''], other possible values are '''unscented''' and '''mc''' | * '''distribution_approximation''' = ['''cubature'''], other possible values are '''unscented''' and '''montecarlo''' |
Nonlinear Estimation interface
This page describes the algorithms available to estimate nonlinear models using particle filters. Currently only the estimation based on the solution of the second order approximation of the model is implemented (third order approxiamtion should be implemented in the coming year).
New options of the estimation command
Currently there is no interface (the user has to manipulate the global structure options_), the following options (discussed with Frédéric) needs to be implemented:
number_of_particles = [5000], any integer
resampling = [systematic], other possible values are none and generic
resampling_method = [kitagawa], other possible value is stratified
filter_algorithm = [sis], other possible values are apf, gf, gmf and cpf
proposal_approximation = [cubature], other possible value are unscented and montecarlo
distribution_approximation = [cubature], other possible values are unscented and montecarlo
online_particle_filter, this keyword triggers the estimation of the posterior mode using the online particle filter approach.