I have to find the two intersection points of pdf function of normal distribution.
I have calculated all the point (x,y) for the curves by iy = pdf('normal', ix, mu, sd) and plotted them on the screen which has two intersection points.
I have tried fzero function but it does not work the means and standard deviations are different for both curves so the length of the arrays are different.
I tried simplest logic two for loops but it did not work either.
The brute force approach did not work for me because of the precision in matlab it does not consider 24.000 and 24.001 for example and the resulting values from the gaussian has 15 integers after decimal point which made it impossible for matlab to check for equality.
Only jump to numerical methods if analysis fails. Finding the intersection points of two normal distributions is a fairly simple algebra problem, which I am too lazy now to do properly, but Matlab can do it for me:
>> syms x sig1 sig2 mu1 mu2;
>> solve(1/sig1/sqrt(2*pi) * exp(-1/2*((x-mu1)/sig1)^2) == ...
1/sig2/sqrt(2*pi) * exp(-1/2*((x-mu2)/sig2)^2), x)
ans =
+(mu2*sig1^2 - mu1*sig2^2 + sig1*sig2*(2*sig2^2*log(sig2/sig1) - 2*sig1^2*log(sig2/sig1) - 2*mu1*mu2 + mu1^2 + mu2^2)^(1/2))/(sig1^2 - sig2^2)
-(mu1*sig2^2 - mu2*sig1^2 + sig1*sig2*(2*sig2^2*log(sig2/sig1) - 2*sig1^2*log(sig2/sig1) - 2*mu1*mu2 + mu1^2 + mu2^2)^(1/2))/(sig1^2 - sig2^2)
where sig1, sig2 are the first and second standard deviation, and mu1, mu2 are the first and second mean, respectively.
If you prefer a numerical approach to an analytic one, you can use fzero and the normpdf function.
x_intersect = fzero(#(x) normpdf(x, mu1, std1) - normpdf(x, mu2, std2), x0);
Since the normal distribution is well behaved, and any two distributions must intersect, any initial guess x0 should work.
Trying to improve the answer as this is an accepted answer(Full credit to Eitan T who has explained beautifully in this related answer here about intersection of curves)
You'll have to find the point of intersection (px, py) manually:
idx = find(y1 - y2 < eps, 1); %// Index of coordinate in array
px = x(idx);
py = y1(idx);
Remember that we're comparing two numbers in floating point representation, so instead of y1 == y2 we must set a tolerance. I've chosen it as eps, but it's up to you to decide.
To draw a circle around this point, you can compute its points and then plot them, but a better approach would be to plot one point with a blown-up circle marker (credit to Jonas for this suggestion):
plot(px, py, 'ro', 'MarkerSize', 18)
This way the dimensions of the circle are not affected by the axes and the aspect ratio of the plot.
Related
I need to plot a parabola and ellipse. However the ellipse is giving me trouble. Can anyone help? The equations are: y = -5*x^2 + 2 and (x^2/16) + (y^2/2) = 4
I've tried this code but obviously I feel like like it isn't right.
x = linspace(-5, 5);
y1 = (x.^2/16) + (y.^2/2) - 1;
y2 = -5*x.^2 +2;
figure(1)
plot(x, y1)
hold on
plot(x, y2)
hold off
Firstly, you did not define a range variable x. Secondly, the ellipse won't pass the vertical line test and can't be plotted like a regular function f(x). Thirdly, your equation y1 = (x.^2/16) + (y.^2/2) - 1; is non-sensical because you have y on each side.
You could correct your method by defining a range variable x1 and x2 that each have appropriate ranges for the functions your plotting. What I mean by this is that you probably don't want the same range for each function, because the ellipse is undefined over most of the range that the parabola is defined. To plot the ellipse using f(x) you could observe that there are + and - values that are identical, using this fact you could plot your ellipse by two functions one to represent the top half and one to represent the bottom half, each of these would pass the vertical line test.
OR
You could utilize ezplot and have a nice time with it because it makes your life easier. Here is a solution.
ezplot('x^2/16+y^2/2-4'); axis equal; hold on
ezplot('-5*x^2+2-y')
There are multiple ways to plot an ellipse, e.g. you could also use a parametric representation of the equation.
In your approach though, when plotting functions using plot(x,y) command, you need to express your dependent variable (y) through independent variable (x). You defined the range for x, which is what you substitute into your equations in order to find y's. While for the parabola, the dependency of y from x is obvious, you forgot to derive such a relationship for the ellipse. In this case it will be +-sqrt((1 - x^2/16)*2). So in your approach, you'll have to take into account both negative and positive y's for the same value of x. Also there's discrepancy in your written equation for the ellipse (=4) and the one in Matlab code (=1).
x = linspace(-5, 5);
y1 = sqrt((1 - x.^2/16)*2);
y2 = -5*x.^2 +2;
figure(1)
plot(x, real(y1), 'r', x, -real(y1), 'r')
hold on
plot(x, y2)
hold off
Since the ellipse has real y's not on the whole x domain, if you want to plot only real parts, specify real(y1) or abs(y1) (even though Matlab does it for you, too). You can also dismiss complex numbers for certain x when computing y1, but you'll need a for-loop for that.
In order to make things simpler, you can check the function fimplicit, ezplot is not recommended according to Matlab's documentation. Or if you want to plot the ellipse in a parametric way, fplot will work, too.
Another (more classic) approach for parametric plotting is given here already, then you don't need any other functions than what you already use. I think it is the simplest and most elegant way to plot an ellipse.
You will not be able to generate points for the ellipse using a function f(x) from a Cartesian linspace range. Instead, you can still use linspace but for the angle in a polar notation, from 0 to 2*pi. You should also be able to easily adjust radius and offset on both axis on the cos and sin expressions.
x = linspace(-5, 5);
y2 = -5*x.^2 +2;
figure(1)
clf;
plot(x, y2)
hold on
a = linspace(0,2*pi);
x2 = 4*cos(a);
y2 = sqrt(2)*sin(a);
plot(x2, y2)
xlim([-5,5]);
ylim([-5,5]);
hold off
Say I've got a number of curves with different length (number of points in each curve and points distance are all vary). Could I find a curve in 3D space that fit best for this group of lines?
Code example in Matlab would be appreciated.
example data set:
the 1st curve has 10 points.
18.5860 18.4683 18.3576 18.2491 18.0844 17.9016 17.7709 17.6401 17.4617 17.2726
91.6178 91.5711 91.5580 91.5580 91.5701 91.6130 91.5746 91.5050 91.3993 91.2977
90.6253 91.1090 91.5964 92.0845 92.5565 93.0199 93.5010 93.9785 94.4335 94.8851
the 2nd curve has 8 points.
15.2091 15.0894 14.9765 14.8567 14.7360 14.6144 14.4695 14.3017
90.1138 89.9824 89.8683 89.7716 89.6889 89.6040 89.4928 89.3624
99.4393 99.9066 100.3802 100.8559 101.3340 101.8115 102.2770 102.7296
a desired curve is one that could represent these two exist curves.
I have thinking of make these curves as points scatters and fit a line out of them. But only a straight line can I get from many code snippet online.
So did I missing something or could someone provide some hint. Thanks.
Hard to come up with a bulletproof solution without more details, but here's an approach that works for the sample data provided. I found the line of best fit for all the points, and then parameterized all the points along that line of best fit. Then I did least-squares polynomial fitting for each dimension separately. This produced a three-dimensional parametric curve that seems to fit the data just fine.
Note that curve fitting approaches other than polynomial least-squares might be better suited to some cases---just substitute the preferred fitting function for polyfit and polyval.
Hope this is helpful!
clear;
close all;
pts1=[18.5860 18.4683 18.3576 18.2491 18.0844 17.9016 17.7709 17.6401 17.4617 17.2726;
91.6178 91.5711 91.5580 91.5580 91.5701 91.6130 91.5746 91.5050 91.3993 91.2977;
90.6253 91.1090 91.5964 92.0845 92.5565 93.0199 93.5010 93.9785 94.4335 94.8851]';
pts2=[ 15.2091 15.0894 14.9765 14.8567 14.7360 14.6144 14.4695 14.3017;
90.1138 89.9824 89.8683 89.7716 89.6889 89.6040 89.4928 89.3624;
99.4393 99.9066 100.3802 100.8559 101.3340 101.8115 102.2770 102.7296]';
%Combine all of our curves into a single point cloud
X = [pts1;pts2];
%=======================================================
%We want to first find the line of best fit
%This line will provide a parameterization of the points
%See accepted answer to http://stackoverflow.com/questions/10878167/plot-3d-line-matlab
% calculate centroid
x0 = mean(X)';
% form matrix A of translated points
A = [(X(:, 1) - x0(1)) (X(:, 2) - x0(2)) (X(:, 3) - x0(3))];
% calculate the SVD of A
[~, S, V] = svd(A, 0);
% find the largest singular value in S and extract from V the
% corresponding right singular vector
[s, i] = max(diag(S));
a = V(:, i);
%=======================================================
a=a / norm(a);
%OK now 'a' is a unit vector pointing along the line of best fit.
%Now we need to compute a new variable, 't', for each point in the cloud
%This 't' value will parameterize the curve of best fit.
%Essentially what we're doing here is taking the dot product of each
%shifted point (contained in A) with the normal vector 'a'
t = A * a;
tMin = min(t);
tMax = max(t);
%This variable represents the order of our polynomial fit
%Use the smallest number that produces a satisfactory result
polyOrder = 8;
%Polynomial fit all three dimensions separately against t
pX = polyfit(t,X(:,1),polyOrder);
pY = polyfit(t,X(:,2),polyOrder);
pZ = polyfit(t,X(:,3),polyOrder);
%And that's our curve fit: (pX(t),pY(t),pZ(t))
%Now let's plot it.
tFine = tMin:.01:tMax;
fitXFine = polyval(pX,tFine);
fitYFine = polyval(pY,tFine);
fitZFine = polyval(pZ,tFine);
figure;
scatter3(X(:,1),X(:,2),X(:,3));
hold on;
plot3(fitXFine,fitYFine,fitZFine);
hold off;
I am trying to construct the product of the FFT of a 2D box function and the FFT of a 2D Gaussian function. After, I find the inverse FFT and I am expecting a convolution of the two functions as a result. However, I am getting a weird one-sided result as seen below. The result appears on the bottom right of the subplot.
The Octave code I wrote to reproduce the above subplot as well as the calculations I performed to construct the convolution is shown below. Can anyone tell me what I'm doing wrong?
clear all;
clc;
close all;
% domain on each side is 0-9
L = 10;
% num subdivisions
N = 32;
delta=L/N;
sigma = 0.5;
% get the domain ready
[x,y] = meshgrid((0:N-1)*delta);
% since domain ranges from 0-(N-1) on both sdes
% we need to take the average
xAvg = sum(x(1, :))/length(x(1,:));
yAvg = sum(y(:, 1))/length(x(:,1));
% gaussian
gssn = exp(- ((x - xAvg) .^ 2 + (y - yAvg) .^ 2) ./ (2*sigma^2));
function ret = boxImpulse(a,b)
n = 32;
L=10;
delta = L/n;
nL = ((n-1)/2-3)*delta;
nU = ((n-1)/2+3)*delta;
if ((a >= nL) && (a <= nU) && ( b >= nL) && (b <= nU) )
ret=1;
else
ret=0;
end
ret;
endfunction
boxResponse = arrayfun(#boxImpulse, x, y);
subplot(2,2,1);mesh(x,y,gssn); title("gaussian fun");
subplot(2,2,2);mesh(x,y, abs(fft2(gssn)) .^2); title("fft of gaussian");
subplot(2,2,3);mesh(x,y,boxResponse); title("box fun");
inv_of_product_of_ffts = abs(ifft2(fft2(boxResponse) * fft2(gssn))) .^2 ;
subplot(2,2,4);mesh(x,y,inv_of_product_of_ffts); title("inv of product of fft");
Let's first address your most obvious errors. This is the way you are computing the convolution in frequency domain:
inv_of_product_of_ffts = abs(ifft2(fft2(boxResponse) * fft2(gssn))) .^2 ;
The first problem is that you are using *, which is matrix multiplication. In frequency domain, element-wise multiplication is the equivalent to convolution in the frequency domain, so you need to use .* instead. The second problem is you also don't need the .^2 term and the abs operation. You're performing convolution, then finding the absolute value and squaring each term. This isn't required. Remove the abs and .^2 operations.
Therefore, what you need is:
inv_of_product_of_ffts = ifft2(fft2(boxResponse).*fft2(gssn)));
However, what you're going to get is this result. Let's place this in a new figure instead of a subplot*:
figure;
mesh(x,y,ifft2(fft2(boxResponse).*fft2(gssn)));
title('Convolution... not right though');
You can see that it's the right result... but it's not centred.... why is that?
This is actually one of the most common problems when computing convolution in the frequency domain. In fact, even the most experienced encounter this problem because they don't understand the internals of how the FFT works.
This is a consequence with how MATLAB and Octave compute the 2D FFT. Specifically, MATLAB and Octave defines a meshgrid of coordinates that go from 0,1,...M-1 for the horizontal and 0,1,...N-1 for the vertical, giving us a M x N result. This means that the origin / DC component is at the top-left corner of the matrix, not the centre as we usually define things. Specifically, the traditional 2D FFT defines the coordinates from -(M-1)/2, ..., (M-1)/2 for the horizontal and -(N-1)/2, ..., (N-1)/2 for the vertical.
Referencing the aforementioned, you defined your signal with the centre assuming that it's the origin and not the top-left corner. To compensate for this, you need to add an fftshift (MATLAB doc, Octave doc) so that the output of the FFT is now centred at the origin and not the top-left corner to bring things back to the way they were, and therefore you really need:
figure;
mesh(x,y,fftshift(ifft2(fft2(boxResponse).*fft2(gssn))));
title('Convolution... now it is right');
If you want to double check that we have the right result, you can perform direct convolution in the spatial domain and we can compare the results between the method in frequency domain.
This is how you'd compute the result directly in spatial domain:
figure;
mesh(x,y,conv2(gssn,boxResponse,'same'));
title('Convolution... spatial domain');
conv2 (MATLAB doc, Octave doc) performs 2D convolution between two signals, and the 'same' flag ensures that the output size is the largest of the two signals, which is either the Gaussian or Box filter. You'll see that it's the same curve, and I won't show it here for brevity.
However, we can compare both of the results and see if they're the same element-wise. A method to do this would be to determine if subtracting each element in the result is less than some threshold... say.. 1e-10:
>> out1 = conv2(boxResponse, gssn, 'same');
>> out2 = fftshift(ifft2(fft2(boxResponse).*fft2(gssn)));
>> all(abs(out1(:)-out2(:)) < 1e-10)
ans =
1
This means that indeed both of these are the same.
Hope that makes sense! Remember, since you're defining your signal where the origin is at the centre, once you find the inverse FFT, you must shift things back so that the origin is now at the centre, not the top-left corner.
*: Minor note - All plots produced in this answer were generated with MATLAB R2015a. However, the code fully works in Octave - tested with Octave 4.0.0.
I have a set of data with over 4000 points. I want to exclude grooves from them, ideally from the point from which they start. The data look for example like this:
The problem with this is the noise I get at the top of the plateaus. I have an idea, in which I would take an average value of the most common within some boundaries (again, ideally sth like the red line here:
and then I would construct a temporary matrix, which would fill up one by one with Y if they are less than this average. If the Y(i) would rise above average, the matrix would find its minima and compare it with the global minima. If the temporary matrix's minima wouldn't be sth like 80% of the global minima, it would be discarded as noise.
I've tried using mean(Y), interpolating and fitting it in a polynomial (the green line) - none of those method would cut it to the point I would be satisfied.
I need this to be extremely robust and it doesn't need to be quick. The top and bottom values can vary a lot, as well as the shape of the plateaus. The groove width is more or less the same.
Do you have any ideas? Again, the point is to extract the values that would make the groove.
How about a median filter?
Let's define some noisy data similar to yours, and plot it in blue:
x = .2*sin((0:9999)/1000); %// signal
x(1000:1099) = x(1000:1099) + sin((0:99)/50*pi); %// noise: spike
x(5000:5199) = x(5000:5199) - sin((0:199)/100*pi); %// noise: wider spike
x = x + .05*sin((0:9999)/10); %// noise: high-freq ripple
plot(x)
Now apply the median filter (using medfilt2 from the Image Processing Toolbox) and plot in red. The parameter k controls the filter memory. It should chosen to be large compared to noise variations, and small compared to signal variations:
k = 500; %// filter memory. Choose as needed
y = medfilt2(x,[1 k]);
hold on
plot(y, 'r', 'linewidth', 2)
In case you don't have the image processing toolbox and can't use medfilt2 a method that's more manual. Skip the extreme values, and do a curve fit with sin1 as curve type. Note that this will only work if the signal is in fact a sine wave!
x = linspace(0,3*pi,1000);
y1 = sin(x) + rand()*sin(100*x).*(mod(round(10*x),5)<3);
y2 = 20*(mod(round(5*x),5) == 0).*sin(20*x);
y = y1 + y2; %// A messy sine-wave
yy = y; %// Store the messy sine-wave
[~, idx] = sort(y);
y(idx(1:round(0.15*end))) = y(idx(round(0.15*end))); %// Flatten out the smallest values
y(idx(round(0.85*end):end)) = y(idx(round(0.85*end)));%// Flatten out the largest values
[foo goodness output] = fit(x.',y.', 'sin1'); %// Do a curve fit
plot(foo,x,y) %// Plot it
hold on
plot(x,yy,'black')
Might not be perfect, but it's a step in the right direction.
I've been asked to plot the function
f(x) = (x^2 - 3x + 7) / (sqrt(2x +5))
for -1< x <5
how would I go about starting this?
Thanks all
Matlab works with arrays (matrices actually). First, you need to create an array with the x values:
x = -1: 0.01: 5
where 0.01 is the interval between consecutive values.
Then you need to calculate to y values.
y = (x.^2 - 3*x + 7) ./ (sqrt(2*x + 5));
This is quite straightforward. The only thing you need to notice is the dots. The .*, ./ operators work element-wise (which you need, since you want to calculate the square of the values). If instead you typed x^2, it would mean matrix multiplication of x by itself, which would produce the wrong values in your case.
Finally, to plot it:
plot(x, y, '.b');
x and y are obvious. The last part refers to the color and style of the line. In this case it mean blue dots. For other styles see Matlab line styles.
You can, also, do this by using the symbolic toolbox as follows:
syms x
y = (x^2 - 3*x + 7)/(sqrt(2*x+5));
figure
ezplot(y,[-1,5]) % -1 is the xmin and 5 the xmax
update: I just noticed that patrick mentioned the usage of symbolic toolbox, but it worths seeing the code as a whole