I have a set of 3D points (x,y,z) and I would like to fit a straight line using Least absolute deviation method to those data.
I found a function from the internet which works pretty well with 2D data, how could I modify this to adapt 3D data points?
function B = L1LinearRegression(X,Y)
% Determine size of predictor data
[n m] = size(X);
% Initialize with least-squares fit
B = [ones(n,1) X] \ Y;
% Least squares regression
BOld = B;
BOld(1) = BOld(1) + 1e-5;
% Force divergence
% Repeat until convergence
while (max(abs(B - BOld)) > 1e-6) % Move old coefficients
BOld = B; % Calculate new observation weights (based on residuals from old coefficients)
W = sqrt(1 ./ max(abs((BOld(1) + (X * BOld(2:end))) - Y),1e-6)); % Floor to avoid division by zero
% Calculate new coefficients
B = (repmat(W,[1 m+1]) .* [ones(n,1) X]) \ (W .* Y);
end
Thank you very much!
I know that this is not answer to the question but rather to different problem leading to the question.
We can use fit function several times.
% XYZ=[x(:),y(:),z(:)]; % suppose we have data in this format
M=size(XYZ,1); % read size of our data
t=((0:M-1)/(M-1))'; % create arbitrary parameter t
% fit all coordinates as function x_i=a_i*t+b_i
fitX=fit(t,XYZ(:,1),'poly1');
fitY=fit(t,XYZ(:,2),'poly1');
fitZ=fit(t,XYZ(:,3),'poly1');
temp=[0;1]; % define the interval where the line shall be plotted
%Evaluate and plot the line coordinates
Line=[feval(fitX(temp)),feval(fitY(temp)),feval(fitZ(temp))];
plot(Line)
The advantage is that this work for any cloud, even if it is parallel to any axis. another advantage is that you are not limitted only to polynomes of 1st order, you can choose any function for different axis and fit any 3D curve.
Related
I'm trying to generate a 1D mesh with unequal step length, and with a fixed number of elements [same as the initial mesh].
The length should be proportional to a node density. In the example, this density is inversely proportional to the gradient of a function.
[imagine for example that you have a distribution of the temperature in a 1D mesh, and you want to make the mesh denser in the parts of the mesh where the temperature gradient is higher]
This is what I coded so far:
% % % Initial fixed-step 1D mesh
X=(0:.01:2)';
% % % Values of a function at each mesh node [in this example, f(x)=5*sin(2*pi*x)*x ]
Y=5*sin(2*pi*X).*X;
% % % Calculate density of mesh points based on the Y function gradient
rho=[1e-9; abs(diff(Y))];
% % % Calculate x-steps from the original mesh
h = diff(X);
% % % Rescale the steps based on the inverse of the density
F = cumsum([0; h]./rho);
% % % Make sure F is between 0 and 1
F = F/F(end);
% % % Calculate the new mesh with scaled steps
X2 = X(1) + F * (X(end)-X(1));
% % % Interpolate the function Y at the new positions
Y2 = interp1(X,Y,X2);
% % % Plot
figure
subplot(2,1,1)
hold on
plot(X,Y,'ko-')
plot(X2,Y2,'r.-')
xlabel('x')
ylabel('y')
subplot(2,1,2)
plot(X,rho,'ko-')
xlabel('x')
ylabel('rho')
There are a few problems with this approach:
1. as you see from this example, there are big jumps when the density is very low (gradient almost zero). How could I implement a minimum/maximum time step size?
2. the node density is calculated correctly, but after "integrating" the unequal steps it can happen that the imposed large time step when the gradient is small causes to skip a high-gradient region that should have finer time-steps. [for example, please take a look at the region 1.8-1.9 in the example below, which should have small time step because it has high node density, but the large step size at ~1.75 causes to skip a large section of X]
Any suggestion to improve my code will be appreciated!
Calculate the cumulative sum (CDF) of rho. Take equally spaced samples from the CDF. Map from CDF to X to get new X3. Map from X3 to Y to get Y3:
CDF = cumsum(rho);
eq_smpl = linspace(CDF(1), CDF(end), numel(CDF)+1).';
eq_smpl = eq_smpl(1:end-1) + diff(eq_smpl)/2; %use center points
X3 = interp1(CDF, X, eq_smpl);
Y3 = interp1(X, Y, X3);
plot(X3,Y3,'ro-')
hold on
plot(X,Y,'k.')
The third subplot shows the the result.
rahnema1's answer gave me a huge help, but there were still two remaining issues:
1- the first element of the new mesh is not identical to the first element of the original mesh
2- in case the gradient is zero at some point, the interp1 function will give error ["The grid vectors are not strictly monotonic increasing."]
For the 1st point, I replaced the two lines defining eq_smpl with the following line:
eq_smpl = linspace(CDF(1), CDF(end), numel(CDF))';
[taking as many elements as CDF, and not centering the points]
For the 2nd point, I added a line after the calculation of rho to replace eventual 0 with small non-zero values:
rho(rho==0)=1e-12;
The final code that does what I want is the following:
% % % Initial fixed-step 1D mesh
X=(0:.01:2)';
% % % Values of a function at each mesh node [in this example, f(x)=5*sin(2*pi*x)*x ]
Y=5*sin(2*pi*X).*X;
% % % Calculate density of mesh points based on the Y function gradient
rho=[0; abs(diff(Y)./abs(diff(X)))];
% % % Replace eventual 0 with small non-zero values
rho(rho==0)=1e-12;
CDF = cumsum(rho);
eq_smpl = linspace(CDF(1), CDF(end), numel(CDF))';
% % % Calculate new mesh
X3 = interp1(CDF, X, eq_smpl);
% % % Interpolate the function Y at the new positions
Y3 = interp1(X, Y, X3);
% % % Plot
figure
subplot(2,1,1)
hold on
plot(X,Y,'ko-')
plot(X3,Y3,'r.-')
xlabel('x')
ylabel('y')
subplot(2,1,2)
plot(X,rho,'ko-')
xlabel('x')
ylabel('rho')
Thank you again to rahnema1 for providing 90% of the answer [probably I didn't explain very well the original request]!
Interpolate the Runge function of Example 10.6 at Chebyshev points for n from 10 to 170
in increments of 10. Calculate the maximum interpolation error on the uniform evaluation
mesh x = -1:.001:1 and plot the error vs. polynomial degree as in Figure 10.8 using
semilogy. Observe spectral accuracy.
The runge function is given by: f(x) = 1 / (1 + 25x^2)
My code so far:
x = -1:0.001:1;
n = 170;
i = 10:10:170;
cx = cos(((2*i + 1)/(2*(n+1)))*pi); %chebyshev pts
y = 1 ./ (1 + 25*x.^2); %true fct
%chebyshev polynomial, don't know how to construct using matlab
yc = polyval(c, x); %graph of approx polynomial fct
plot(x, yc);
mErr = (1 / ((2.^n).*(n+1)!))*%n+1 derivative of f evaluated at max x in [-1,1], not sure how to do this
%plotting stuff
I know very little matlab, so I am struggling on creating the interpolating polynomial. I did some google work, but I was confused with the current functions as I didn't find one that just simply took in points and the polynomial to be interpolated. I am also a bit confused in this case of whether I should be doing i = 0:1:n and n=10:10:170 or if n is fixed here. Any help is appreciated, thank you
Since you know very little about MATLAB, I will try explain everything step by step:
First, to visualize the Runge function, you can type:
f = #(x) 1./(1+25*x.^2); % Runge function
% plot Runge function over [-1,1];
x = -1:1e-3:1;
y = f(x);
figure;
plot(x,y); title('Runge function)'); xlabel('x');ylabel('y');
The #(x) part of the code is a function handle, a very useful feature of MATLAB. Notice the function is properly vecotrized, so it can receive as an argument a variable or an array. The plot function is straightforward.
To understand the Runge phenomenon, consider a linearly spaced vector of [-1,1] of 10 elements and use these points to obtain the interpolating (Lagrange) polynomial. You get the following:
% 10 linearly spaced points
xc = linspace(-1,1,10);
yc = f(xc);
p = polyfit(xc,yc,9); % gives the coefficients of the polynomial of degree 10
hold on; plot(xc,yc,'o',x,polyval(p,x));
The polyfit function does a polynomial curve fitting - it obtains the coefficients of the interpolating polynomial, given the poins x,y and the degree of the polynomial n. You can easily evaluate the polynomial at other points with the polyval function.
Obseve that, close to the end domains, you get an oscilatting polynomial and the interpolation is not a good approximation of the function. As a matter of fact, you can plot the absolute error, comparing the value of the function f(x) and the interpolating polynomial p(x):
plot(x,abs(y-polyval(p,x))); xlabel('x');ylabel('|f(x)-p(x)|');title('Error');
This error can be reduced if, instead of using a linearly space vector, you use other points to do the interpolation. A good choice is to use the Chebyshev nodes, which should reduce the error. As a matter of fact, notice that:
% find 10 Chebyshev nodes and mark them on the plot
n = 10;
k = 1:10; % iterator
xc = cos((2*k-1)/2/n*pi); % Chebyshev nodes
yc = f(xc); % function evaluated at Chebyshev nodes
hold on;
plot(xc,yc,'o')
% find polynomial to interpolate data using the Chebyshev nodes
p = polyfit(xc,yc,n-1); % gives the coefficients of the polynomial of degree 10
plot(x,polyval(p,x),'--'); % plot polynomial
legend('Runge function','Chebyshev nodes','interpolating polynomial','location','best')
Notice how the error is reduced close to the end domains. You don't get now that high oscillatory behaviour of the interpolating polynomial. If you plot the error, you will observe:
plot(x,abs(y-polyval(p,x))); xlabel('x');ylabel('|f(x)-p(x)|');title('Error');
If, now, you change the number of Chebyshev nodes, you will get an even better approximation. A little modification on the code lets you run it again for different numbers of nodes. You can store the maximum error and plot it as a function of the number of nodes:
n=1:20; % number of nodes
% pre-allocation for speed
e_ln = zeros(1,length(n)); % error for the linearly spaced interpolation
e_cn = zeros(1,length(n)); % error for the chebyshev nodes interpolation
for ii=1:length(n)
% linearly spaced vector
x_ln = linspace(-1,1,n(ii)); y_ln = f(x_ln);
p_ln = polyfit(x_ln,y_ln,n(ii)-1);
e_ln(ii) = max( abs( y-polyval(p_ln,x) ) );
% Chebyshev nodes
k = 1:n(ii); x_cn = cos((2*k-1)/2/n(ii)*pi); y_cn = f(x_cn);
p_cn = polyfit(x_cn,y_cn,n(ii)-1);
e_cn(ii) = max( abs( y-polyval(p_cn,x) ) );
end
figure
plot(n,e_ln,n,e_cn);
xlabel('no of points'); ylabel('maximum absolute error');
legend('linearly space','chebyshev nodes','location','best')
My code is not plotting the correct contour plot for a plane perpendicular to a ring of charge running through its center properly. My problem is that the contour plot is not filling 2D space.
I've made two versions of code, one uses a for loop to calculate a
Riemann sum and the other simply uses the sum command. Both rely on the
'subs' command for substituting values from a meshgrid into my expression for V (electric potential).
Version 1 (using for loop):
%% Computing a symbolic expression for V for anywhere in space
syms x y z % phiprime is angle that an elemental dq of the circular
charge is located at, x,y and z are arbitrary points in space outside the
charge distribution
N = 200; % number of increments to sum
R = 2; % radius of circle is 2 meters
dphi = 2*pi/N; % discretizing the circular line of charge which spans 2pi
integrand = 0;
for phiprime = 0:dphi:2*pi
% phiprime ranges from 0 to 2pi in increments of dphi
integrand = integrand + dphi./(sqrt(((x - R.*cos(phiprime) )).^2 + ((y -
R.*sin(phiprime) ).^2) + z.^2));
end
intgrl = sum(integrand);
% uncessary but harmless step that I leave to show that I am using the
sum of the above expression for each dphi
eps0 = 8.854e-12;
kC = 1/(4*pi*eps0);
rhol = 1*10^-9; % linear charge density
Vtot = kC*rhol*R.*intgrl; % symbolic expression for Vtot
%% Graphing V & E in plane perpedicular to ring & passing through center
[Y1, Z1] = meshgrid(-4:.5:4, -4:.5:4);
Vcont1 = subs(Vtot, [x,y,z], {0,Y1,Z1}); % Vcont1 stands for V contour; 1
is because I do the plane of the ring next
contour(Y1,Z1,Vcont1)
xlabel('y - axis [m]')
ylabel('z - axis [m]')
title('V in a plane perpendicular to a ring of charge (N = 200)')
str = {'Red line is side view', 'of ring of charge'};
text(-1,2,str)
hold on
% visually displaying line of charge on plot
circle = rectangle('Position',[-2 0 4 .1],'Curvature',[1,1]);
set(circle,'FaceColor',[1, 0, 0],'EdgeColor',[1, 0, 0]);
% taking negative gradient of V and finding symbolic equations for Ex, Ey
and Ez
g = gradient(-1.*(kC*rhol*R.*intgrl),[x,y,z]);
%% now substituting all the values of the 2D coordinate system for the
symbolic x and y variables to get numeric values for Ex and Ey
Ey1 = subs(g(2), [x y z], {0,Y1,Z1});
Ez1 = subs(g(3), [x y z], {0,Y1,Z1});
E1 = sqrt(Ey1.^2 + Ez1.^2); % full numeric magnitude of E in y-z plane
Eynorm1 = Ey1./E1; % This normalizes the electric field lines
Eznorm1 = Ez1./E1;
quiver(Y1,Z1,Eynorm1,Eznorm1);
hold off
Version 2 (using sum command):
syms x y z
R = 2; % radius of circle is 2 meters
N=100;
dphi = 2*pi/N; % discretizing the circular line of charge which spans 2pi
phiprime = 0:dphi:2*pi; %phiprime ranges from 0 to 2pi in increments of
dphi
integrand = dphi./(sqrt(((x - R.*cos(phiprime) )).^2 + ((y -
R.*sin(phiprime) ).^2) + z.^2));
phiprime = 0:dphi:2*pi;
intgrl = sum(integrand); % Reimann sum performed here
eps0 = 8.854e-12;
kC = 1/(4*pi*eps0);
rhol = 1*10^-9; % linear charge density
Vtot = kC*rhol*R.*intgrl; % symbolic expression for Vtot
Everything else after that point for version 2 is the same as version 1 (substituting for the symbols x,y,z etc)
I would post images of what the code produces but apparently you need 10 reputation for that. Thanks stackoverflow. This will be much more confusing to understand without the images.
The vector field produced by my code is correct while the contour plot seems to use only a few points around the ends of the ring and connect them with straight lines in a strange diamond shape. I can't get it to fill space.
I receive no error messages. The contour lines accumulate around the ends of the ring (where the potential would approach infinity) in a strange diamond shape but aren't graphed anywhere else. I need the contour plot to fill the 2D grid
I received a solution to this question from MATLAB's community and posted about it here:
https://scicomp.stackexchange.com/questions/32834/graphing-electric-potential-of-a-ring-of-charge-using-matlab-help/32842#32842
I would post here but this "you can't post images because you don't have enough reputation" thing would make my explanation too abstract and difficult to understand so go take a look if you are having MATLAB contour plot issues and want to see my problem and solution
Edit: Some time after I asked this question, an R package called MonoPoly (available here) came out that does exactly what I want. I highly recommend it.
I have a set of points I want to fit a curve to. The curve must be monotonic (never decreasing in value) i.e. the curve can only go upward or stay flat.
I originally had been polyfitting my results and this had been working great until I found a particular dataset. The polyfit for data in this dataset was non-monotonic.
I did some research and found a possible solution in this post:
Use lsqlin. Constrain the first derivative to be non-negative at both
ends of the domain of interest.
I'm coming from a programming rather than math background so this is a little beyond me. I don't know how to constrain the first derivative to be non-negative as he said. Also, I think in my case I need a curve so I should use lsqcurvefit but I don't know how to constrain it to produce monotonic curves.
Further research turned up this post recommending lsqcurvefit but I can't figure out how to use the important part:
Try this non-linear function F(x) also. You use it together with
lsqcurvefit but it require a start guess on the parameters. But it is
a nice analytic expression to give as a semi-empirical formula in a
paper or a report.
%Monotone function F(x), with c0,c1,c2,c3 varitional constants F(x)=
c3 + exp(c0 - c1^2/(4*c2))sqrt(pi)...
Erfi((c1 + 2*c2*x)/(2*sqrt(c2))))/(2*sqrt(c2))
%Erfi(x)=erf(i*x) (look mathematica) but the function %looks much like
x^3 %derivative f(x), probability density f(x)>=0
f(x)=dF/dx=exp(c0+c1*x+c2*x.^2)
I must have a monotonic curve but I'm not sure how to do it, even with all of this information. Would a random number be enough for a "start guess". Is lsqcurvefit best? How can I use it to produce a best fitting monotonic curve?
Thanks
Here is a simple solution using lsqlin. The derivative constrain is enforced in each data point, this could be easily modified if needed.
Two coefficient matrices are needed, one (C) for least square error calculation and one (A) for derivatives in the data points.
% Following lsqlin's notations
%--------------------------------------------------------------------------
% PRE-PROCESSING
%--------------------------------------------------------------------------
% for reproducibility
rng(125)
degree = 3;
n_data = 10;
% dummy data
x = rand(n_data,1);
d = rand(n_data,1) + linspace(0,1,n_data).';
% limit on derivative - in each data point
b = zeros(n_data,1);
% coefficient matrix
C = nan(n_data, degree+1);
% derivative coefficient matrix
A = nan(n_data, degree);
% loop over polynomial terms
for ii = 1:degree+1
C(:,ii) = x.^(ii-1);
A(:,ii) = (ii-1)*x.^(ii-2);
end
%--------------------------------------------------------------------------
% FIT - LSQ
%--------------------------------------------------------------------------
% Unconstrained
% p1 = pinv(C)*y
p1 = fliplr((C\d).')
p2 = polyfit(x,d,degree)
% Constrained
p3 = fliplr(lsqlin(C,d,-A,b).')
%--------------------------------------------------------------------------
% PLOT
%--------------------------------------------------------------------------
xx = linspace(0,1,100);
plot(x, d, 'x')
hold on
plot(xx, polyval(p1, xx))
plot(xx, polyval(p2, xx),'--')
plot(xx, polyval(p3, xx))
legend('data', 'lsq-pseudo-inv', 'lsq-polyfit', 'lsq-constrained', 'Location', 'southoutside')
xlabel('X')
ylabel('Y')
For the specified input the fitted curves:
Actually this code is more general than what you requested, since the degree of polynomial can be changed as well.
EDIT: enforce derivative constrain in additional points
The issue pointed out in the comments is due to that the derivative checks are enforced only in the data points. Between those no checks are performed. Below is a solution to alleviate this problem. The idea: convert the problem to an unconstrained optimization by using a penalty term.
Note that it is using a term pen to penalize the violation of the derivative check, thus the result is not a true least square error solution. Additionally, the result is dependent on the penalty function.
function lsqfit_constr
% Following lsqlin's notations
%--------------------------------------------------------------------------
% PRE-PROCESSING
%--------------------------------------------------------------------------
% for reproducibility
rng(125)
degree = 3;
% data from comment
x = [0.2096 -3.5761 -0.6252 -3.7951 -3.3525 -3.7001 -3.7086 -3.5907].';
d = [95.7750 94.9917 90.8417 62.6917 95.4250 89.2417 89.4333 82.0250].';
n_data = length(d);
% number of equally spaced points to enforce the derivative
n_deriv = 20;
xd = linspace(min(x), max(x), n_deriv);
% limit on derivative - in each data point
b = zeros(n_deriv,1);
% coefficient matrix
C = nan(n_data, degree+1);
% derivative coefficient matrix
A = nan(n_deriv, degree);
% loop over polynom terms
for ii = 1:degree+1
C(:,ii) = x.^(ii-1);
A(:,ii) = (ii-1)*xd.^(ii-2);
end
%--------------------------------------------------------------------------
% FIT - LSQ
%--------------------------------------------------------------------------
% Unconstrained
% p1 = pinv(C)*y
p1 = (C\d);
lsqe = sum((C*p1 - d).^2);
p2 = polyfit(x,d,degree);
% Constrained
[p3, fval] = fminunc(#error_fun, p1);
% correct format for polyval
p1 = fliplr(p1.')
p2
p3 = fliplr(p3.')
fval
%--------------------------------------------------------------------------
% PLOT
%--------------------------------------------------------------------------
xx = linspace(-4,1,100);
plot(x, d, 'x')
hold on
plot(xx, polyval(p1, xx))
plot(xx, polyval(p2, xx),'--')
plot(xx, polyval(p3, xx))
% legend('data', 'lsq-pseudo-inv', 'lsq-polyfit', 'lsq-constrained', 'Location', 'southoutside')
xlabel('X')
ylabel('Y')
%--------------------------------------------------------------------------
% NESTED FUNCTION
%--------------------------------------------------------------------------
function e = error_fun(p)
% squared error
sqe = sum((C*p - d).^2);
der = A*p;
% penalty term - it is crucial to fine tune it
pen = -sum(der(der<0))*10*lsqe;
e = sqe + pen;
end
end
Gradient free methods might be used to solve the problem by exactly enforcing the derivative constrain, for example:
[p3, fval] = fminsearch(#error_fun, p_ini);
%--------------------------------------------------------------------------
% NESTED FUNCTION
%--------------------------------------------------------------------------
function e = error_fun(p)
% squared error
sqe = sum((C*p - d).^2);
der = A*p;
if any(der<0)
pen = Inf;
else
pen = 0;
end
e = sqe + pen;
end
fmincon with non-linear constraint might be a better choice.
I let you to work out the details and to tune the algorithms. I hope that it is sufficient.
After constructing the point cloud I want to get the normal of each point and I used the built-in matlab function surfnorm but its takes a lot of processing time. So if anyone could assist me do this a better and more efficient way.
I wonder if the following code would help you. There are three steps here.
Create 500 randomly spaced points (x,y), and compute a corresponding value z (the height of the surface) for which I chose a sinc like function
Resample the random points using the TriScatteredInterp function - this permits me to obtain points on an evenly sampled grid that "roughly correspond" to the initial surface
Compute the normal to "some points" on that grid (since there are 480x640 points, computing the normal at every point would just create an impossibly dense "forest of vectors"; by sampling "every 10th point" you can actually see what you are doing
The code I used was as follows:
randomX = rand(1,500);
randomY = rand(1,500);
r = 5*sqrt(randomX.^2 + randomY.^2);
randomZ = sin(r) ./ r;
% resample the data:
[xx yy] = meshgrid(linspace(0,1,640), linspace(0,1,480));
F = TriScatteredInterp(randomX(:), randomY(:), randomZ(:));
zz = F(xx, yy);
%% at each point, the normal is cross product of vectors to neighbors
xyz=reshape([xx yy zz],[size(xx) 3]);
xv = 10:30:479; yv = 10:30:639; % points at which to compute normals
dx = xyz(xv, yv+1, :) - xyz(xv, yv, :);
dy = xyz(xv+1, yv, :) - xyz(xv, yv, :);
normVecs = cross(dx, dy); % here we compute the normals.
normVecs = normVecs ./ repmat(sqrt(sum(normVecs.^2, 3)), [1 1 3]);
figure;
quiver3(xx(xv, yv), yy(xv, yv), zz(xv, yv), ...
normVecs(:,:,1), normVecs(:,:,2), normVecs(:,:,3));
axis equal
view([56 22]);
And the resulting plot: