How to plot a smooth curve for Empirical probability density ?
20 vues (au cours des 30 derniers jours)
Afficher commentaires plus anciens
Hi everyone,
I require to plot an empirical probability density curve of a data set (184 rows, 59 columns). Initially, I pick one column to plot the results but, the EPD curve looks so wired. May someone suggest to me how can I get desired result (Attached). Data is also attached for reference.
clear all
clc
X = load('R_0.01T.csv'); % input data
SzX = size(X) % matrix size
r=[X(1,:)]; % first row of the data set
ra = r(~isnan(r)); % remove all nan enteries
[f,x,flo,fhi] = ecdf(ra);
WinLen = 5
dfdxs = smoothdata(gradient(f)./gradient(x), 'movmedian',WinLen); % ... Then, Smooth Them Again ...
aaa = smooth(dfdxs);
plot(x, aaa)
This is what i get
Here is the expected results.
2 commentaires
Réponse acceptée
Mathieu NOE
le 29 Mar 2022
hello
tried a few options , smoothdata and spline fit. In fact, I was thinking you wanted something super smooth so I first opted for the spline fit. But I saw that the expected plot was supposed to have some waviness , so I changed to smoothdata
NB as I don't have the toolbox for ecdf , I found an alternative on FEX for it https://fr.mathworks.com/matlabcentral/fileexchange/32831-homemade-ecdf
but I had to correct a bug (the corrected function is attaced fyi)
hope the code below can help you
clear all
clc
X = load('R_0.01T.csv'); % input data
SzX = size(X); % matrix size
r=X(1,:); % first row of the data set
ra = r(~isnan(r)); % remove all nan enteries
% [f,x,flo,fhi] = ecdf(ra);
[f,x] = homemade_ecdf(ra); % FEX : https://fr.mathworks.com/matlabcentral/fileexchange/32831-homemade-ecdf
% remove duplicates
[x,ia,ic] = unique(x);
f = f(ia);
% resample the data (interpolation) on linearly spaced points
xx = linspace(min(x),max(x),100);
ff = interp1(x,f,xx);
% smoothing
ffs = smoothdata(ff, 'loess',40);
% or spline fit ?
% Breaks interpolated from data (log x scale to get more points in the
% lower x range and less in the upper x range)
breaks = logspace(log10(min(x)),log10(max(x)),5); %
p = splinefit(xx,ff,breaks); %
ff2 = ppval(p,xx);
figure(1),
plot(x,f,'*',xx,ffs,'-',xx,ff2,'-')
legend('raw','smoothed','spline fit');
% compute gradient from spline fitted data (my choice)
dx = mean(diff(xx));
dfdxs = gradient(ffs)./dx;
figure(2),plot(xx, dfdxs)
14 commentaires
Mathieu NOE
le 1 Avr 2022
hello again
1/ seems the expected curves are smoother compare to what we have now. You can probably increase the smoothing (driven by the value in smoothdata parameter)
ffs = smoothdata(ff , 'loess',10); % increase 10 =>20
2/ scale of the colorbar (ticks) : I don't know what are the values you specify in UU coming from ?
Plus de réponses (0)
Voir également
Catégories
En savoir plus sur Preprocessing Data dans Help Center et File Exchange
Produits
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!