p147685/CRE_RB_and_CCRE
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Big modification in the way the histogram is calculated.

Browse Source 
Now first the edges are calculated based on the two paramters: EqualSizeBinFlag and UseHistProxy.
Once edges for bins are known (calculated) creating the cdf is the same for all cases.
This commit is contained in:
R.J. Renken
2022-03-26 16:10:39 +01:00
parent 3998755f40
commit ce95974c7b
2 changed files with 99 additions and 66 deletions
Download Patch File Download Diff File Expand all files Collapse all files

86
TestCREClass.m
View File

@ -4,33 +4,54 @@
%% set some constants
A=randn(100);
%% create instance of CRE class
S=CREClass;
%% check histogram vs "new" technique
% NB I come to the conclusion that the new method is not always working.
% Why?
S.Data=A;% set gaussian data
S.nBin=(numel(S.Data)); % set nbin to npoints; This is fine as we use the cummulative distribution and the formulas as defined in Zografos
% use histogram
S.UseHistProxy=true;
S.Calc;
S
% Use my aproximation for the histogram
% only do this if the number of data points is not too big.
S.UseHistProxy=false;
S.Calc;
S
S1=CREClass;
%% Calculate by setting labda equal to each of the (unique) values in the data.
S1.UseHistProxy=false;
S1.EqualSizeBinFlag=false;
S1.Data=A;% set gaussian data
S1.Calc;
% use histogram aproximation with unequal bin sizes in the histogram
S2=CREClass;
S2.UseHistProxy=true;
S2.EqualSizeBinFlag=false;
S2.Data=A;% set gaussian data
S2.nBin=numel(A); % set nbin to npoints; This is fine as we use the cummulative distribution and the formulas as defined in Zografos
S2.Calc
% use histogram aproximation with equal bin sizes in the histogram
S3=CREClass;
S3.UseHistProxy=true;
S3.EqualSizeBinFlag=false;
S3.Data=A;% set gaussian data
S3.nBin=numel(A); % set nbin to npoints; This is fine as we use the cummulative distribution and the formulas as defined in Zografos
S3.Calc
%Check recalculation of edges
S4=CREClass;
S4.UseHistProxy=true;
S4.EqualSizeBinFlag=true;
S4.Data=A;% set gaussian data
S4.nBin=numel(A); % set nbin to npoints; This is fine as we use the cummulative distribution and the formulas as defined in Zografos
S4.Calc
disp(S4)
S4.nBin=numel(A)/10;
S4.CalcEdges ; % force a recalculation of the edges
S4.Calc;
disp(S4)
return;
%% Show effect of scaling or ofsett the data
% Scale
S.Data=100*A;
S.UseHistProxy=true;
S.Calc;
S
disp(S);
% Offset
S.Data=A+10;
S.UseHistProxy=true;
S.Calc;
S
disp(S);
%%
%% Test CCRE class
@ -39,21 +60,30 @@ figure(2);clf;
figure(3);clf;hold on;
A=randn(100);
B=A;
B(:)=A(randperm(numel(A)))
for k=-1:0.1:1
B(:)=A(randperm(numel(A)));
w=-1:0.1:1;
out(numel(w),1)=struct('w',[],'CRE',[],'CCRE',[],'R',[]);
for k=1:numel(w)
out(k).w=w(k);
CS=CCREClass;
CS.Data=1*(1-abs(k))*A+k*B;
CS.EquidistantBinFlag=false;
CS.Data=1*(1-abs(w(k)))*A+w(k)*B;
CS.nBin=50;
CS.DataRef=B;
CS.nBinRef=500;
CS.Calc;
figure(1);
plot(k,CS.CCRE,'o')
plot(k,CS.CRE,'x');
out(k).CRE=CS.CRE;
out(k).CCRE=CS.CCRE;
out(k).R=CS.CCRE./CS.CRE;
figure(2);
scatter(CS.DataRef(:),CS.Data(:));
figure(3);
plot(k,CS.CCRE./CS.CRE,'d');
title(sprintf('w: %d',w(k)));
scatter(CS.DataRef(:),CS.Data(:));
snapnow
end
CS
snapnow
disp(CS);
figure(1);
clf;hold on;
plot([out(:).w],[out(:).CCRE],'o');
plot([out(:).w],[out(:).CRE],'x');
figure(3);clf
plot([out(:).w],[out(:).R],'d')