RACT
PmWiki
Hosted by

Ract /
DraftPreUsageAdd /ract directory in Matlab path Directory tree and files @UBASE definition and few auxilary routines for a new type of variable: uncertain parameter  PRIVATE routines for random matrix generation (NO EDIT NEED)  display.M nice way to display uncertain param. variable in Matlab (EDIT ONLY IN CASE of new random distributions)  distrtype.M auxilary (NO EDIT NEED)  ubase.M create a new uncertain variable, for more details type 'help ubase' (EDITABLE)  ubasesample.M return some number of samples of an uncertain variable (EDITABLE)  ADDONS generators of random in some case matrices (obsolete, saved for historical reasons and in reserve) most of the code used in @UBASE/PRIVATE (NO EDIT NEED)  PRIVATE two main subroutines of the toolbox  plusproj.M projection of symmetric matrix to the cone of positive matrices (EDITABLE)  usample.M sampling of uncertain variables, but can accept cell arrays of uncertain variables! (for the first glance, NO EDIT NEED)  RACTDEMOS  hurwitz_demo.M a demonstration of the toolbox. The only one yet. :) Usage: make RACT\RACTDEMOS as current dir, and run it (ABSOLUTELY EDITABLE)  RACTTEMPLATES templates for users and developers  tmpl_function.M template for userdefined function, that 1  should define a set of uncertain variables 2  should return an object by given samples of uncertain variables (EDITABLE)  tmpl_help1.M inline help lines structure ex_lqr.M script of an example to test. Unfortunately, I've failed to make use of it, so (SKIP & EDITABLE) ex1_a.M same as previous (SKIP & EDITABLE) ex_robhurwitz.M example of userdefined polynomial, the task is to check probability of stability of the uncertain polynomial. Used by HURWITZ_DEMO.M (ABSOLUTELY EDITABLE) lqr_grad.M a try to implement construction of LQRcontroller. Failed (seems to be error somewhere) (SKIP & EDITABLE) !!ok, now the issue files!! uevaluate.M return array of uncertain object samples, by evaluating userdefined uncertain function on a number of samples allows vectorization (seems GOOD, by still EDITABLE) uprobperf.M check if a property of an uncertain object performs with accuracy Eps and confidence 1Delta both the property and uncertain object given by links to userdefined functions. Chernoff bound used for the number of samples needed. (EDITABLE) It is the most thing possible in the toolbox. utest.M return a portion of samples of uncertain object, for which a property satisfied (property and object are defined by user) Used to estimate of probabilty if perfomance. Few properties, namely 'unit', 'is_hurwitz', 'is_schur', 'is_positive_definite' are already implemented as subfunctions (ABSOLUTELY EDITABLE) The mechanism/usage.Keep in mind that only the perfomance probability estimation works. Step 1. User defines an uncertain object by creating and naming a function using the template TMPL_FUNCTION.M, this function must return an object itself or the value of a property (0  if not satisfied, 1  if satisfied) Step 2. User calls prob_est = UPROBPERF(user_function_name, 'hurwitz', Eps, Delta) where 'schur' or 'posdef' as second argument are possible to check corresponding property, or 'unit' if user function already returns the property of an uncertain object; Eps is accuracy and 1Delta is confidence level. BackgroundVectorization and parameter calling conventions. To discuss. ErrataFound an incompatability with Matlab 6. Latest version at . 