% For the case of no regrouping, this code is executed. It is also executed
% once before entering the "Reg_Method_1" loop.
% Copyright 2008, 2009, 2010, 2011 George I. Evers (moved from "RegPSO_main.m" on May 3, 2009).
if OnOff_Tricias_NN_training
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% This paragraph copyright 2010, 2011, 2012 Tricia Rambharose.
% Created on: 2010/09/18
% info@tricia-rambharose.com
% Seed the random generator
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%Seed the random generator to fix the non-randomness of initial particle
%positions.
RandStream.setDefaultStream ...
(RandStream('mt19937ar','seed',sum(100*clock)));
end
%$%%%%%%%%%%%%
%Standard PSO%
%%%%%%%%%%%%%%
if OnOff_asymmetric_initialization
vmax_perc = 4*vmax_perc; %When asymmetric initialization is used,
%velocities should be clamped based on the range of the search
%space rather than on the range of the initialization space.
end
vmax(1:np, 1:dim) = vmax_perc.*range_IS; %Define the maximum
%velocity for velocity initialization and, optionally,
%velocity clamping.
x(1:np, 1:dim) = center_IS + range_IS.*rand(np, dim) - range_IS./2; %Randomly initialize the
%particles about "center_IS".
RegPSO_grouping_counter = 1; %stores the current grouping #
if Reg_Method == 1;
if OnOff_MPSO
MPSO_grouping_counter = MPSO_grouping_counter + 1; %counts the
%total # of groupings over the entire MPSO trial (i.e. across
%all MPSO starts): the first increment done here is from 1
%to 2
end
if OnOff_func_evals
OPSO_ghost_FEs_RegPSO = 0;
end
end
if OnOff_lbest %Call "gbest_core" or "lbest_core."
lbest_core
else
gbest_core
end
if OnOff_graphs
if OnOff_swarm_trajectory
RegPSO_k = 0; %Add this variable, which would otherwise
%not apply to the standard case, for use below
%(updated later in case Reg_Method == 1).
if dim == 2
if GraphParams_swarm_traj_snapshot_mode == 2 %For this case, determine the meshgrid now rather than
%recursively within "Swarm_Trajectory."
%Set min & max on each dimension then generate a meshgrid.
Internal_GraphParams_dim1min = min(xhist(:, 1:2:size(xhist, 2))); %2*0+1 = 1st column (iter 0, 1st dim),
Internal_GraphParams_dim1max = max(xhist(:, 1:2:size(xhist, 2))); %2*1+1 = 3rd column (1st iter, 1st dim)
Internal_GraphParams_dim2min = min(xhist(:, 2:2:size(xhist, 2))); %2*0+1 = 1st column (iter 0, 2nd dim),
Internal_GraphParams_dim2max = max(xhist(:, 2:2:size(xhist, 2))); %2*1+1 = 3rd column (1st iter, 2nd dim)
f_vs_2d %Call script to generate points for graphing obj. func. vs 2D meshgrid.
end
Swarm_Trajectory; %Calculate swarm state at various iterations using "xhist."
end
end
end
if Reg_Method == 1;
RegPSO_k = k; %"RegPSO_k" accrues with each grouping the # of PSO iterations within each RegPSO trial.
if OnOff_func_evals
OPSO_ghost_FEs_RegPSO = OPSO_ghost_FEs_counter;
end
if OnOff_graphs
if OnOff_swarm_trajectory
RegPSO_g_after_each_grouping = g(1, :);
% "RegPSO_g_after_each_grouping" is used to determine when to begin decreasing the regrouping factor
% and requires the past 50 global bests to be maintained within each trial rather than simply
% being reconstructed as is the other grouping data after all groupings have occurred.
% "OnOff_num_gs_identical_b4_refinement == 1" can therefore be expected to slow program
% execution a bit.
if OnOff_g_after_each_grouping && OnOff_MPSO && (~OnOff_Autosave_Workspace_Per_Grouping)
%If "OnOff_Autosave_Workspace_Per_Grouping,"
%"MPSO_g_after_each_grouping" will more efficiently be
%reconstructed within "RegPSO_load_grouping_data.m," which loads
%the desired data from the workspace saved at the end of each
%grouping.
MPSO_g_after_each_grouping = [MPSO_g_after_each_grouping; g(1, :)];
end
end
elseif (OnOff_g_after_each_grouping && (OnOff_MPSO || ~OnOff_Autosave_Workspace_Per_Grouping)) || OnOff_num_gs_identical_b4_refinement
RegPSO_g_after_each_grouping = g(1, :);
% "RegPSO_g_after_each_grouping" is used to determine when to begin decreasing the regrouping factor
% and requires the past 50 global bests to be maintained within each trial rather than simply
% being reconstructed as is the other grouping data after all groupings have occurred.
% "OnOff_num_gs_identical_b4_refinement == 1" can therefore be expected to slow program
% execution a bit.
if OnOff_g_after_each_grouping && OnOff_MPSO && (~OnOff_Autosave_Workspace_Per_Grouping)
%If "OnOff_Autosave_Workspace_Per_Grouping,"
%"MPSO_g_after_each_grouping" will more efficiently be
%reconstructed within "RegPSO_load_grouping_data.m," which loads
%the desired data from the workspace saved at the end of each
%grouping.
MPSO_g_after_each_grouping = [MPSO_g_after_each_grouping; g(1, :)];
end
end
if OnOff_range_IS_per_grouping && (OnOff_MPSO || OnOff_see_data_per_grouping || ~OnOff_Autosave_Workspace_Per_Grouping)
%If the user does not wish to see the data per grouping and ~OnOff_MPSO,
%this can be generated more efficiently from the saved
%WS of the groupings. In the case of MPSO, it may
%be beneficial to still generate this.
RegPSO_range_IS_per_grouping = range_IS0.*ones(1, dim); %Store the initial range_IS
%and, if specified as a scalar for convenience, convert to the expected vector
if OnOff_MPSO && (~OnOff_Autosave_Workspace_Per_Grouping) %otherwise, this history...
%will more efficiently be constructed within...
%"RegPSO_load_grouping_data.m" rather than needing to be...
%initialized here.
MPSO_range_IS_per_grouping = [MPSO_range_IS_per_grouping; range_IS(1, :)];
end
end
if OnOff_MPSO || (~OnOff_Autosave_Workspace_Per_Grouping)
%If the workspace is saved per grouping, this data
%will more efficiently be constructed within
%"RegPSO_load_grouping_data.m" rather than needing to be maintained
%here. However, if MPSO is active, it may be beneficial to
%maintain the "RegPSO_k_after_each_grouping" over each RegPSO execution within an
%MPSO trial as well as reconstructing "MPSO_k_after_each_grouping" over all
%MPSO groupings spanning multiple trials.
if OnOff_fg_after_each_grouping
%If the worskpace is saved per grouping, "MPSO_fg_after_each_grouping" can more efficiently be
%constructed at the end of the trial.
if OnOff_MPSO && (~OnOff_Autosave_Workspace_Per_Grouping)
MPSO_fg_after_each_grouping = [MPSO_fg_after_each_grouping, fg];
end
RegPSO_fg_after_each_grouping = fg;
end
if OnOff_k_after_each_grouping
if OnOff_MPSO && (~OnOff_Autosave_Workspace_Per_Grouping)
MPSO_k_after_each_grouping = [MPSO_k_after_each_grouping, MPSO_k + RegPSO_k];
%If the workspace is saved per grouping, this history...
%will more efficiently be constructed within...
%"RegPSO_load_grouping_data.m" rather than needing to be maintained
%here.
end
RegPSO_k_after_each_grouping(RegPSO_grouping_counter) = RegPSO_k; %"RegPSO_k_after_each_grouping" is a matrix of order "num_trials" by #
end %of groupings implemented. It holds the iteration numbers AT which RegPSO regrouped and exists purely
%for user analysis.
end
end
iter_tot_all_trials = iter_tot_all_trials + k + 1;
%The "+ 1" has the effect of counting all random initializations in
%addition to the number of updates counted by "k".
if Reg_Method ~= 0
if OnOff_Autosave_Workspace_Per_Grouping
RegPSO_save_grouping_data
end
end
%(removed for now) tcpu_tot = tcpu_tot + tcpu_elapsed; %%(option to track "cpu time" too)
