pso.zip_PSO_melanoma diagnosis_zip

; + ; NAME: ; RMD_PSO ; ; PURPOSE: ; Function that attempts to find the global minimum of a function ; using a particle swarm optimization (PSO) strategy. ; ; In this algorithm the individual agents (swarm members) "fly" through ; parameter space seeking the global function minimum. At each time ; step in the search every agent determines the value of the function ; at its location in parameter space and compares this value with its ; own previous best value as well as the overall best parameter found ; among all of the agents. The velocity of each agent is determined by ; an attraction between its current position, the best solution for that ; particular agent, and the best solution for all agents found. ; The velocity of each agent is updated at each time step according to ; the equation: ; ; v[t+dt] = b*v[t] + dv[t] where ; dv[t] = w1*r*(pbest[t] - p[t]) + w2*r*(gbest[t] - p[t]) ; ; w1 and w2 are weights (also known as learning factors) with ; typical values ranging between 0 and 4, b is a velocity damping factor ; (also called an inertia weight), r is a uniform random deviate, ; p is the current parameter set, pbest is the best parameter set found ; for this particular agent, and gbest is the best parameter set found ; for all particular agents. The inertia weight decreases from 1 at each ; time step so that after k updates, the value is b^k. One points to ; understand about the learning factors: (1) w1 governs the "cognitive" ; aspect of the motion (that which depends on its own memory) and w2 ; governs the social aspect of the motion (that which depends on the ; discovery of the entire swarm). Note that the scales of the parameters, ; as defined by PRANGE, is absorbed into the learning factors. ; ; The position in parameter space is updated according to the equation: ; p[t+dt] = p[t] + dt*v[t] ; ; Note that "elastic" boundary conditions are implemented so that if an ; agent exceeds the limits of the parameter range (PRANGE) then it will ; be "reflected" (or bounce) from the boundary. ; ; For additional details on this optimization technique please consult ; the following web site: http://www.swarmintelligence.org/ ; ; RETURN_VALUE: ; Best parameters found in the minimization. ; ; CATEGORY: ; OPTIMIZATION, OBJECTS, SWARM INTELLIGENCE ; ; CALLING SEQUENCE: ; RESULT = RMD_PSO(PRANGE = prange,FUNCTION_NAME = function_name,....) ; ; PARAMETERS: ; NONE ; ; KEYWORDS: ; FTOL: Fractional tolerance defined as the average fitness ; divided by the best fitness. FTOL is the termination ; criteria for the quantity defined above (DEF: 0.1) ; PRANGE: Range of parameters (2 by N array where N is the number ; of parameters in the optimization). ; WEIGHTS: Learning factors. (w1 and w2 in equation above) (DEF: [1.,1.]) ; FUNCTION_NAME: Name of the function to be used in the minimization ; FUNCTION_VALUE:Value of FUNCTION_NAME at the result found in the minimization. ; FUNCTARGS: Exogenous inputs required for function evaluation above ; and beyond the parameters in the optimization. ; NUM_PARTICLES: Number of particles in the swarm (DEF: 10) ; ITMAX: Maximum number of iterations to perform if FTOL termination ; criteria not met. ; VEL_FRACTION: The fraction of the parameter range by which the initial ; velocity is created. (DEF: 0.1) ; VEL_DECREMENT: The fractional amount by which the velocity will be decremented ; at each iteration. Also called velocity damping and inertia weight. ; This is the parameter designated b in the equation above. (DEF: 0.95) ; ITERPROC: Name (string) of the procedure to be called at ; each iteration step. An example of such a procedure is ; PSO_ITERPROC, shown below. ; ; The requirements on the intermediate reporting procedure are ; that it be written as a PROCEDURE and it must accept the ; following arguments: FUNC, P (the current best parameter set), ; ITER (the current iteration, and INTERRUPT (a byte variable equal 1B or 0B ; indicating if the algorithm should cease at the next update--note that ; interrupt must be a variable since it is expected to be passed by reference). ; Optional keywords are FUNCTARGS, ITERARGS, and OREF. OREF is provided ; so that you can extract information on the PSO object (its state) ; at the current iteration. This can be useful if you want to get ; all of the current members of the swarm and display them, for instance. ; ; ITERARGS: Structure containing exogenous information necessary ; to perform the intermediate reporting procedure ; ITERPROC. ; NCALLS: Number of function calls over the course of the ; optimization procedure. ; QUIET: If set, no intermediate output will be provided ; (i.e. ITERPROC will not be called). Default: 1B ; ; REQUIREMENTS: ; IDL 6.0 and higher ; ; REQUIRED PROGRAMS: ; NONE ; ; COMMON BLOCKS: ; NONE ; ; SIDE EFFECTS: ; NONE ; ; EXAMPLE: ; Example usage found at end of this source listing namded ; PSO_MIN_EXAMPLE. To run this code, compile this file and ; type PSO_MIN_EXAMPLE at the IDL prompt. ; ; IDL> PSO_MIN_EXAMPLE ; ; AUTHOR: ; Robert Dimeo ; NIST Center for Neutron Research ; National Institute of Standards and Technology ; 100 Bureau Drive-Stop 8562 ; Gaithersburg, MD 20899 ; ; DISCLAIMER ; This software is provided as is without any warranty whatsoever. ; Permission to use, copy, modify, and distribute modified or ; unmodified copies is granted, provided this disclaimer ; is included unchanged. ; ; ACKNOWLEDGMENT: ; This work is part of the DAVE development effort at the NCNR and ; was supported in part by the National Science Foundation ; under Agreement No. DMR-0086210. ; ; MODIFICATION HISTORY: ; Written by RMD (12/20/04) ; Updated header documentation, tightened up some ; of the code, and corrected a few errors in the implementation ; of the intermediate reporting (ITERPROC) -- RMD (12/22/04) ; Replaced the periodic boundary conditions with "elastic" ; boundaries -- RMD (12/22/04) ; Removed the velocity tolerance and redefined FTOL so that ; it is not problem nor scale-dependent -- RMD (12/22/04) ; Removed the scale dependence of the learning factors ; -- RMD (12/22/04) ; ; - ; *************************************************** ; pro pso::cleanup compile_opt idl2,hidden ptr_free,self.v_ptr,self.p_ptr,self.pbest_ptr ptr_free,self.gbest_ptr,self.prange_ptr,self.f_ptr ptr_free,self.fbest_ptr,self.gfbest_ptr,self.iterargs_ptr ptr_free,self.functargs_ptr,self.seed_ptr end ; *************************************************** ; pro pso_iterproc, func, $ p, $ iter, $ interrupt, $ functargs = functargs, $ oref = opso, $ _Extra = iterargs compile_opt hidden,idl2 ; Default intermediate reporting procedure print,'Iteration: '+strtrim(string(iter),2) print,' *********************** ' for i = 0,n_elements(p)-1 do begin strout = 'p['+strtrim(string(i),2)+']='+strtrim(string(p[i]),2) print,strout endfor print end ; *************************************************** ; function pso::evaluate_function compile_opt idl2,hidden p = *self.p_ptr psize = size(p) nparms = psize[1] f = fltarr(self.num_particles) for i = 0,self.num_particles-1 do $ f[i] = call_function(self.func,reform(p[*,i]),_Extra = *self.functargs_ptr) *self.f_ptr = f self.ncalls = self.ncal