I am trying to solve a system of four equations in four variables. I have read a number of threads on similar issues and tried to follow the suggestions. But I think it is a bit messy because of the logs and cross products here. This is the exact system:
7*w = (7*w+5*x+2*y+z) * ( 0.76 + 0.12*Log[w] -0.08*Log[x] -0.03*Log[y] -0.07*Log[7*w+5*x + 2*y + z]),
5*x = (7*w+5*x+2*y+z) * ( 0.84 - 0.08*Log[w] +0.11*Log[x] -0.02*Log[y] -0.08*Log[7*w+5*x + 2*y + z]),
2*y = (7*w+5*x+2*y+z) * (-0.45 - 0.03*Log[w] -0.02*Log[x] +0.05*Log[y] +0.12*Log[7*w+5*x + 2*y + z]),
1*z = (7*w+5*x+2*y+z) * (-0.16 + 0*Log[w] - 0*Log[x] - 0*Log[y] + 0.03*Log[7*w+5*x + 2*y + z])
(FYI-I am a young economist, and this is an extension of a consumer demand system.) Theoretically, we know that there exists a unique solution to this system that is positive.
Trys
Solve & NSolve : As there should be a solution I tried these but neither works. I guess that the system has too many logs to handle.
FindRoot : I started with an initial value of (14,15,10,100) which I get from my data. FindRoot returns the last value (which does not satisfy my system) and the following message.
FindRoot::lstol: The line search decreased the step size to within tolerance specified by AccuracyGoal and PrecisionGoal but was unable.....
I tried different initial values, including the value returned by FindRoot. I tried to analyze the pattern of the solution value at each step. I didn’t see any pattern but noticed that the z values become negative early in the process. So I put bounds on the values. This just stops the code at the minimum value of 0.1. I also tried an exponential system instead of log, same issues.
Reap[FindRoot[{
7*w==(7*w+5*x + 2*y + z)*(0.76 + 0.12*Log[w] -0.08*Log[x] -0.03*Log[y] -0.07*Log[7*w+5*x + 2*y + z]),
5*x==(7*w+5*x + 2*y + z)*(0.84 -0.08*Log[w] +0.11*Log[x] -0.02*Log[y] -0.08*Log[7*w+5*x + 2*y + z]),
2*y==(7*w+5*x + 2*y + z)*(-0.45 - 0.03*Log[w] -0.02*Log[x] +0.05*Log[y] +0.12*Log[7*w+5*x + 2*y + z]),
z==(7*w+5*x + 2*y + z)*(-0.16 + 0*Log[w] -0*Log[x] -0*Log[y] +0.03*Log[7*w+5*x + 2*y + z])},
{{w,14,0.1,500},{x,15,0.1,500},{y,10,0.1,500},
{z,100,0.1,500}},EvaluationMonitor:>Sow[{w,x,y,z}] ]]
FindMinimum : As we can write this problem as a minimization problem, I tried this (following the suggestion here). The value returned did not converge the system or equations to zero. I tried with only the first two equations, and that sort of converged to zero.
Hope this is engaging enough for the experts here! Any ideas how I should find the solution or why can’t I? It’s the first time I am using Mathematica, and unfortunately the first time I am empirically solving a system/optimizing! Thanks a lot.
{g1,g2,g3, g4}={7*w - (7*w+5*x+2*y+z)* (0.76+0.12*Log[w]-0.08*Log[x]-0.03*Log[y] -0.07*Log[7*w+5*x+2*y+z]),5*x - (7*w+5*x+2*y+z)*(0.84-0.08*Log[w]+0.11*Log[x]-0.02*Log[y] -0.08*Log[7*w+5*x+2*y+z]),2*y - (7*w+5*x+2*y+z)*(-0.45-0.03*Log[w]-0.02*Log[x]+0.05*Log[y]+0.12*Log[7*w+5*x+2*y+z]), 1*z - (7*w+5*x+2*y+z)*(-0.16+0*Log[w]-0*Log[x]-0*Log[y]+0.03*Log[7*w+5*x+2*y+z])};subdomain=0<w<100 &&0<x<100 && 0<y<100 && 0<z<100;res=FindMinimum[{Total[{g1,g2,g3, g4}^2],subdomain},{w,x,y,z},AccuracyGoal->5]{g1,g2,g3,g4}/.res[[2]]
I don't have access to Mathematica, I put your equations into AMPL which is free for students. Here is what I did:
var w := 14 >= 0;
var x := 15 >= 0;
var y := 10 >= 0;
var z := 100 >= 0;
eq1: 7*w = (7*w+5*x+2*y+z) * ( 0.76 + 0.12*log(w) -0.08*log(x) -0.03*log(y) -0.07*log(7*w+5*x + 2*y + z));
eq2: 5*x = (7*w+5*x+2*y+z) * ( 0.84 - 0.08*log(w) +0.11*log(x) -0.02*log(y) -0.08*log(7*w+5*x + 2*y + z));
eq3: 2*y = (7*w+5*x+2*y+z) * (-0.45 - 0.03*log(w) -0.02*log(x) +0.05*log(y) +0.12*log(7*w+5*x + 2*y + z));
eq4: 1*z = (7*w+5*x+2*y+z) * (-0.16 + 0*log(w) - 0*log(x) - 0*log(y) +0.03*log(7*w+5*x + 2*y + z));
option show_stats 1;
option presolve 10;
option solver "/home/ali/ampl/ipopt"; # put your path here
option seed 1731;
# Initial solve
solve;
display w, x, y, z;
# Multistart
for {1..10} {
for {j in 1.._snvars}
let _svar[j] := Uniform(1, 50);
solve;
if (solve_result_num < 200) then {
display w, x, y, z;
}
}
If I only require that the variables are nonnegative, I get rubbish, for example
w = 2.39266e-11
x = 6.62678e-11
y = 1.57043e-24
z = 7.0842e-10
or
w = 1.09972e-12
x = 9.77807e-11
y = 3.36229e-21
z = 1.85441e-09
Numerically, these are indeed solutions, they satisfy the equations to a fairly high precision, although I am pretty sure it's not what you are looking for. This indicates issues with your model.
If I increase the lower bounds of the variables a bit:
var w := 14 >= 0.1;
var x := 15 >= 0.1;
var y := 10 >= 0.1;
var z := 100 >= 0.01;
I get, even with multistart, Ipopt 3.11.6: Converged to a locally infeasible point. Problem may be infeasible. This again indicates issues with your model equations.
I am afraid you will have to revise your model.
This won't fix the issues with your model equations but I would introduce new variables: a=log(w), b=log(x), c=log(y), d=log(z). Then w=exp(a) and so on. It has the advantage that the function evaluations won't fail due to negative arguments for the logarithm function.
I would probably also introduce a new variable for (7*w+5*x+2*y+z) just to make the equations more compact.
Neither of these new variables will solve the above issues with your model equations.
If it is really your first time using Mathematica, you might be better off with AMPL and IPOPT; these tools are custom-tailored for solving equations and optimization problems. I suggest you use the AMPL mailing list if you have question and not Stackoverflow; you will get better answers on the mailing list.
This method will often rapidly find approximate solutions, NMinimize the sum of squares with constraints.
In[2]:= NMinimize[{
(7*w - (7*w + 5*x + 2*y + z)*(0.76 + 0.12*Log[w] - 0.08*Log[x] -
0.03*Log[y] - 0.07*Log[7*w + 5*x + 2*y + z]))^2 +
(5*x - (7*w + 5*x + 2*y + z)*(0.84 - 0.08*Log[w] + 0.11*Log[x] -
0.02*Log[y] - 0.08*Log[7*w + 5*x + 2*y + z]))^2 +
(2*y - (7*w + 5*x + 2*y + z)*(-0.45 - 0.03*Log[w] - 0.02*Log[x] +
0.05*Log[y] + 0.12*Log[7*w + 5*x + 2*y + z]))^2 +
(1*z - (7*w + 5*x + 2*y + z)*(-0.16 + 0*Log[w] +
0.03*Log[7*w + 5*x + 2*y + z]))^2,
w > 0 && x > 0 && y > 0 && z > 0}, {w, x, y, z},
Method -> "RandomSearch"]
Out[2]= {9.34024*10^-12, {w->1.86998*10^-8, x->3.83383*10^-8, y->4.59973*10^-8, z->5.29581*10^-7}}
Related
I am trying to numerically solve the Klein-Gordon equation that can be found here. To make sure I solved it correctly, I am comparing it with an analytical solution that can be found on the same link. I am using the finite difference method and Matlab. The initial spatial conditions are known, not the initial time conditions.
I start off by initializing the constants and the space-time coordinate system:
close all
clear
clc
%% Constant parameters
A = 2;
B = 3;
lambda = 2;
mu = 3;
a = 4;
b = - (lambda^2 / a^2) + mu^2;
%% Coordinate system
number_of_discrete_time_steps = 300;
t = linspace(0, 2, number_of_discrete_time_steps);
dt = t(2) - t(1);
number_of_discrete_space_steps = 100;
x = transpose( linspace(0, 1, number_of_discrete_space_steps) );
dx = x(2) - x(1);
Next, I define and plot the analitical solution:
%% Analitical solution
Wa = cos(lambda * x) * ( A * cos(mu * t) + B * sin(mu * t) );
figure('Name', 'Analitical solution');
surface(t, x, Wa, 'edgecolor', 'none');
colormap(jet(256));
colorbar;
xlabel('t');
ylabel('x');
title('Wa(x, t) - analitical solution');
The plot of the analytical solution is shown here.
In the end, I define the initial spatial conditions, execute the finite difference method algorithm and plot the solution:
%% Numerical solution
Wn = zeros(number_of_discrete_space_steps, number_of_discrete_time_steps);
Wn(1, :) = Wa(1, :);
Wn(2, :) = Wa(2, :);
for j = 2 : (number_of_discrete_time_steps - 1)
for i = 2 : (number_of_discrete_space_steps - 1)
Wn(i + 1, j) = dx^2 / a^2 ...
* ( ( Wn(i, j + 1) - 2 * Wn(i, j) + Wn(i, j - 1) ) / dt^2 + b * Wn(i - 1, j - 1) ) ...
+ 2 * Wn(i, j) - Wn(i - 1, j);
end
end
figure('Name', 'Numerical solution');
surface(t, x, Wn, 'edgecolor', 'none');
colormap(jet(256));
colorbar;
xlabel('t');
ylabel('x');
title('Wn(x, t) - numerical solution');
The plot of the numerical solution is shown here.
The two plotted graphs are not the same, which is proof that I did something wrong in the algorithm. The problem is, I can't find the errors. Please help me find them.
To summarize, please help me change the code so that the two plotted graphs become approximately the same. Thank you for your time.
The finite difference discretization of w_tt = a^2 * w_xx - b*w is
( w(i,j+1) - 2*w(i,j) + w(i,j-1) ) / dt^2
= a^2 * ( w(i+1,j) - 2*w(i,j) + w(i-1,j) ) / dx^2 - b*w(i,j)
In your order this gives the recursion equation
w(i,j+1) = dt^2 * ( (a/dx)^2 * ( w(i+1,j) - 2*w(i,j) + w(i-1,j) ) - b*w(i,j) )
+2*w(i,j) - w(i,j-1)
The stability condition is that at least a*dt/dx < 1. For the present parameters this is not satisfied, they give this ratio as 2.6. Increasing the time discretization to 1000 points is sufficient.
Next up is the boundary conditions. Besides the two leading columns for times 0 and dt one also needs to set the values at the boundaries for x=0 and x=1. Copy also them from the exact solution.
Wn(:,1:2) = Wa(:,1:2);
Wn(1,:)=Wa(1,:);
Wn(end,:)=Wa(end,:);
Then also correct the definition (and use) of b to that in the source
b = - (lambda^2 * a^2) + mu^2;
and the resulting numerical image looks identical to the analytical image in the color plot. The difference plot confirms the closeness
How to realize the random walk algorithm with MATLAB? I don't understand this algorithm, because the condition 3 never holds. I post my code in the end.
Choose a time-step h>0 so that M=9/h is a integer.
Set log(L_0) = L_0 = 0.13 and Z(0)=0.
rho_k are independent random variables distributed by the law P(+-1)=0.5
1) log(L_{k+1}) = log(L_k)-0.5*sigma^2*h+sigma*sqrt(h)*rho_{k+1}
2) Z_{k+1} = Z_k - sigma* [(erfc((2*2^(1/2)*(log((25*exp(log L_k))/7) + 9/32))/3)/2 - erfc((2*2^(1/2)*(log(100*exp(log L_k)) + 9/32))/3)/2 + erfc((2*2^(1/2)*(log(7/(25*exp(log L_k))) - 9/32))/3)/56 - erfc((2*2^(1/2)*(log(196/(25*exp(log L_k))) - 9/32))/3)/56 + (exp(log L_k)*((2*2^(1/2)*exp(-(8*(log(7/(25*exp(log L_k))) - 9/32)^2)/9))/(3*exp(log L_k)*pi^(1/2)) - (2*2^(1/2)*exp(-(8*(log(196/(25*exp(log L_k))) - 9/32)^2)/9))/(3*exp(log L_k)*pi^(1/2))))/28 - exp(log L_k)*((2*2^(1/2)*exp(-(8*(log((25*exp(log L_k))/7) + 9/32)^2)/9))/(3*exp(log L_k)*pi^(1/2)) - (2*2^(1/2)*exp(-(8*(log(100*exp(log L_k)) + 9/32)^2)/9))/(3*exp(log L_k)*pi^(1/2))) - (14*2^(1/2)*exp(-(8*(log(7/(25*exp(log L_k))) + 9/32)^2)/9))/(75*exp(log L_k)*pi^(1/2)) + (14*2^(1/2)*exp(-(8*(log(196/(25*exp(log L_k))) + 9/32)^2)/9))/(75*exp(log L_k)*pi^(1/2)) + (2^(1/2)*exp(-(8*(log((25*exp(log L_k))/7) - 9/32)^2)/9))/(150*exp(log L_k)*pi^(1/2)) - (2^(1/2)*exp(-(8*(log(100*exp(log L_k)) - 9/32)^2)/9))/(150*exp(log L_k)*pi^(1/2))))]*sqrt(h)*rho_{k+1};
3) log L_k < log(H)+1/2*sigma^2*h-sigma*sqrt(h)
H=0.28
sigma=0.25
K=0.01
To realize the algorithm we follow the random walk generated by (1),
and at each time t_k; we check whether the condition 3 holds. If it does not, L_k has reached the boundary zone and we stop the chain at log(H). If it does,
we perform (1)-(2) to find log(L_k+1) and Z_{k+1}. If k+1=M; we stop, otherwise we continue with the algorithm.
The outcome of simulating each trajectory is a point (t_kappa, log(L_kappa), Z_kappa).
Evaluate the expectation E[(exp(log(L_{kappa}))-K)^+ * Chi(kappa=M) + Z_{kappa}]=price with the Monte Carlo technique and do 10^6 Monte Carlo runs.
Here is my code. It doesn't work, because I don't get something near 0.0657 (the result).
Y = zeros(1,M+1); %ln L_k = Y(k)
Z = zeros(1,M+1);
sigma = 0.25;
H = 0.28;
K = 0.01;
Y(1) = 0.13;
Z(1) = 0;
M = 90;
h = 0.1;
for k = 1:M+1
vec = [-1 1];
index = random('unid', length(vec),1);
x(1,k) = vec(index);
end %'Rho'
for k = 1:M
if Y(k) < log(H) + 1/2*sigma^2*h - sigma*sqrt(h) %Bedingung
Y(k+1) = Y(k) - (1/2)*(sigma)^2*h + sigma*sqrt(h)*x(k+1);
Z(k+1) = Z(k) + (-sigma*(erfc((2*2^(1/2)*(log((25*exp(Y(k)))/7) + 9/32))/3)/2 - erfc((2*2^(1/2)*(log(100*exp(Y(k))) + 9/32))/3)/2 + erfc((2*2^(1/2)*(log(7/(25*exp(Y(k)))) - 9/32))/3)/56 - erfc((2*2^(1/2)*(log(196/(25*exp(Y(k)))) - 9/32))/3)/56 + (exp(Y(k))*((2*2^(1/2)*exp(-(8*(log(7/(25*exp(Y(k)))) - 9/32)^2)/9))/(3*exp(Y(k))*pi^(1/2)) - (2*2^(1/2)*exp(-(8*(log(196/(25*exp(Y(k)))) - 9/32)^2)/9))/(3*exp(Y(k))*pi^(1/2))))/28 - exp(Y(k))*((2*2^(1/2)*exp(-(8*(log((25*exp(Y(k)))/7) + 9/32)^2)/9))/(3*exp(Y(k))*pi^(1/2)) - (2*2^(1/2)*exp(-(8*(log(100*exp(Y(k))) + 9/32)^2)/9))/(3*exp(Y(k))*pi^(1/2))) - (14*2^(1/2)*exp(-(8*(log(7/(25*exp(Y(k)))) + 9/32)^2)/9))/(75*exp(Y(k))*pi^(1/2)) + (14*2^(1/2)*exp(-(8*(log(196/(25*exp(Y(k)))) + 9/32)^2)/9))/(75*exp(Y(k))*pi^(1/2)) + (2^(1/2)*exp(-(8*(log((25*exp(Y(k)))/7) - 9/32)^2)/9))/(150*exp(Y(k))*pi^(1/2)) - (2^(1/2)*exp(-(8*(log(100*exp(Y(k))) - 9/32)^2)/9))/(150*exp(Y(k))*pi^(1/2))))*sqrt(h)*x(k+1);
else
Y(k)=log(H);
break
end
end
if k == M
end
if (L(M)-K) > 0
(L(M)-K);
else
0;
end
return
exp(Y(M))-K+Z(M)
I have used Solve to find the solution of an equation in Mathematica (The reason I am posting here is that no one could answer my question in mathematica stack.)The solution is called s and it is a function of two variables called v and ro. I want to find imaginary and real part of s and I want to use the information that v and ro are real and they are in the below interval:
$ 0.02 < ro < 1 ,
40
The code I used is as below:
ClearAll["Global`*"]
d = 1; l = 100; k = 0.001; kk = 0.001;ke = 0.0014;dd = 0.5 ; dr = 0.06; dc = 1000; p = Sqrt[8 (ro l /2 - 1)]/l^2;
m = (4 dr + ke^2 (d + dd)/2) (-k^2 + kk^2) (1 - l ro/2) (d - dd)/4 -
I v p k l (4 dr + ke^2 (d + dd)/2)/4 - v^2 ke^2/4 + I v k dr l p/4;
xr = 0.06/n;
tr = d/n;
dp = (x (v I kk/2 (4 dr + ke^2 (d + dd)/2) - I v kk ke^2 (d - dd)/8 - dr l p k kk (d - dd)/4) + y ((xr I kk (ro - 1/l) (4 dr + ke^2 (d + dd)/2)) - I v kk tr ke^2 (1/l - ro/2) + I dr xr 4 kk (1/l - ro/2)))/m;
a = -I v k dp/4 - I xr y kk p/2 + l ke^2 dp p (d + dd)/8 + (-d + dd)/4 k kk x + dr l p dp;
aa = -v I kk dp/4 + xr I y k p/2 - tr y ke^2 (1/l - ro/2) - (d - dd) x kk^2/4 + ke^2 x (d - dd)/8;
ca = CoefficientArrays[{x (s + ke^2 (d + dd)/2) +
dp (v I kk - l (d - dd) k p kk/2) + y (tr ro ke^2) - (d -
dd) ((-kk^2 + k^2) aa - 2 k kk a)/(4 dr + ke^2 (d + dd)/2) == 0, y (s + dc ke^2) + n x == 0}, {x, y}];
mat = Normal[ca];
matt = Last#mat;
sha = Solve[Det[matt] == 0, s];
shaa = Assuming[v < 100 && v > 40 && ro < 1 && ro > 0.03,Simplify[%]];
reals = Re[shaa];
ims = Im[shaa];
Solve[reals == 0, ro]
but it gives no answer. Could anyone help? I really appreciate any solution to this problem.
I run your code down to this point
mat = Normal[ca]
and look at the result.
There are lots of very tiny floating point coefficients, so small that I suspect most of them are just floating point noise now. Mathematica thinks 0.1 is only known to 1 significant digit of precision and your mat result is perhaps nothing more than zero correct digits now.
I continue down to this point
sha = Solve[Det[matt] == 0, s]
If you look at the value of sha you will see it is s->stuff and I don't think that is at all what you think it is. Mathematica returns "rules" from Solve, not just expressions.
If I change that line to
sha = s/.Solve[Det[matt] == 0, s]
then I am guessing that is closer to what you are imagining you want.
I continue to
shaa = Assuming[40<v<100 && .03<ro<1, Simplify[sha]];
reals = Re[shaa]
And I instead use, because you are assuming v and ro to be Real and because ComplexExpand has often been very helpful in getting Re to provide desired results,
reals=Re[ComplexExpand[shaa]]
and I click on Show ALL to see the full expanded value of that. That is about 32 large screens full of your expression.
In that are hundreds of
Arg[-1. + 50. ro]
and if I understand your intention I believe all those simplify to 0. If that is correct then
reals=reals/.Arg[-1. + 50. ro]->0
reduces the size of reals down to about 20 large screen fulls.
But there are still hundreds of examples of Sqrt[(-1.+50. ro)^2] and ((-1.+50. ro)^2)^(1/4) making up your reals. Unfortunately I'm expecting your enormous expression is too large and will take too long for Simplify with assumptions to be able to be practically effective.
Perhaps additional replacements to coax it into dramatically simplifying your reals without making any mistakes about Real versus Complex, but you have to be extremely careful with such things because it is very common for users to make mistakes when dealing with complex numbers and roots and powers and functions and end up with an incorrect result, might get your problem down to the point where it might be feasible for
Solve[reals == 0, ro]
to give you a meaningful answer.
This should give you some ideas of what you need to think carefully about and work on.
I want to select a random number from 0,1,2,3...n, however I want to make it that the chance of selecting k|0<k<n will be lower by multiplication of x from selecting k - 1 so x = (k - 1) / k. As bigger the number as smaller the chances to pick it up.
As an answer I want to see the implementation of the next method:
int pickANumber(n,x)
This is for a game that I am developing, I saw those questions as related but not exactly that same:
How to pick an item by its probability
C Function for picking from a list where each element has a distinct probabili
p1 + p2 + ... + pn = 1
p1 = p2 * x
p2 = p3 * x
...
p_n-1 = pn * x
Solving this gives you:
p1 + p2 + ... + pn = 1
(p2 * x) + (p3 * x) + ... + (pn * x) + pn = 1
((p3*x) * x) + ((p4*x) * x) + ... + ((p_n-1*x) * x) + pn = 1
....
pn* (x^(n-1) + x^(n-2) + ... +x^1 + x^0) = 1
pn*(1-x^n)/(1-x) = 1
pn = (1-x)/(1-x^n)
This gives you the probability you need to set to pn, and from it you can calculate the probabilities for all other p1,p2,...p_n-1
Now, you can use a "black box" RNG that chooses a number with a distribution, like those in the threads you mentioned.
A simple approach to do it is to set an auxillary array:
aux[i] = p1 + p2 + ... + pi
Now, draw a random number with uniform distribution between 0 to aux[n], and using binary search (aux array is sorted), get the first value, which matching value in aux is greater than the random uniform number you got
Original answer, for substraction (before question was editted):
For n items, you need to solve the equation:
p1 + p2 + ... + pn = 1
p1 = p2 + x
p2 = p3 + x
...
p_n-1 = pn + x
Solving this gives you:
p1 + p2 + ... + pn = 1
(p2 + x) + (p3 + x) + ... + (pn + x) + pn = 1
((p3+x) + x) + ((p4+x) + x) + ... + ((p_n-1+x) + x) + pn = 1
....
pn* ((n-1)x + (n-2)x + ... +x + 0) = 1
pn* x = n(n-1)/2
pn = n(n-1)/(2x)
This gives you the probability you need to set to pn, and from it you can calculate the probabilities for all other p1,p2,...p_n-1
Now, you can use a "black box" RNG that chooses a number with a distribution, like those in the threads you mentioned.
Be advised, this is not guaranteed you will have a solution such that 0<p_i<1 for all i, but you cannot guarantee one given from your requirements, and it is going to depend on values of n and x to fit.
Edit This answer was for the OPs original question, which was different in that each probability was supposed to be lower by a fixed amount than the previous one.
Well, let's see what the constraints say. You want to have P(k) = P(k - 1) - x. So we have:
P(0)
P(1) = P(0) - x
P(2) = P(0) - 2x
...
In addition, Sumk P(k) = 1. Summing, we get:
1 = (n + 1)P(0) -x * n / 2 (n + 1),
This gives you an easy constraint between x and P(0). Solve for one in terms of the other.
For this I would use the Mersenne Twister algorithm for a uniform distribution which Boost provides, then have a mapping function to map the results of that random distribution to the actual number select.
Here's a quick example of a potential implementation, although I left out the quadtratic equation implementation since it is well known:
int f_of_xib(int x, int i, int b)
{
return x * i * i / 2 + b * i;
}
int b_of_x(int i, int x)
{
return (r - ( r ) / 2 );
}
int pickANumber(mt19937 gen, int n, int x)
{
// First, determine the range r required where the probability equals i * x
// since probability of each increasing integer is x higher of occuring.
// Let f(i) = r and given f'(i) = x * i then r = ( x * i ^2 ) / 2 + b * i
// where b = ( r - ( x * i ^ 2 ) / 2 ) / i . Since r = x when i = 1 from problem
// definition, this reduces down to b = r - r / 2. therefore to find r_max simply
// plugin x to find b, then plugin n for i, x, and b to get r_max since r_max occurs
// when n == i.
// Find b when
int b = b_of_x(x);
int r_max = f_of_xib(x, n, b);
boost::uniform_int<> range(0, r_max);
boost::variate_generator<boost::mt19937&, boost::uniform_int<> > next(gen, range);
// Now to map random number to desired number, just find the positive value for i
// when r is the return random number which boils down to finding the non-zero root
// when 0 = ( x * i ^ 2 ) / 2 + b * i - r
int random_number = next();
return quadtratic_equation_for_positive_value(1, b, r);
}
int main(int argc, char** argv)
{
mt19937 gen;
gen.seed(time(0));
pickANumber(gen, 10, 1);
system("pause");
}
I am building my first large-scale MATLAB program, and I've managed to write original vectorized code for everything so for until I came to trying to create an image representing vector density in stereographic projection. After a couple failed attempts I went to the Mathworks file exchange site and found an open source program which fits my needs courtesy of Malcolm Mclean. With a test matrix his function produces something like this:
And while this is almost exactly what I wanted, his code relies on a triply nested for-loop. On my workstation a test data matrix of size 25000x2 took 65 seconds in this section of code. This is unacceptable since I will be scaling up to a data matrices of size 500000x2 in my project.
So far I've been able to vectorize the innermost loop (which was the longest/worst loop), but I would like to continue and be rid of the loops entirely if possible. Here is Malcolm's original code that I need to vectorize:
dmap = zeros(height, width); % height, width: scalar with default value = 32
for ii = 0: height - 1 % 32 iterations of this loop
yi = limits(3) + ii * deltay + deltay/2; % limits(3) & deltay: scalars
for jj = 0 : width - 1 % 32 iterations of this loop
xi = limits(1) + jj * deltax + deltax/2; % limits(1) & deltax: scalars
dd = 0;
for kk = 1: length(x) % up to 500,000 iterations in this loop
dist2 = (x(kk) - xi)^2 + (y(kk) - yi)^2;
dd = dd + 1 / ( dist2 + fudge); % fudge is a scalar
end
dmap(ii+1,jj+1) = dd;
end
end
And here it is with the changes I've already made to the innermost loop (which was the biggest drain on efficiency). This cuts the time from 65 seconds down to 12 seconds on my machine for the same test matrix, which is better but still far slower than I would like.
dmap = zeros(height, width);
for ii = 0: height - 1
yi = limits(3) + ii * deltay + deltay/2;
for jj = 0 : width - 1
xi = limits(1) + jj * deltax + deltax/2;
dist2 = (x - xi) .^ 2 + (y - yi) .^ 2;
dmap(ii + 1, jj + 1) = sum(1 ./ (dist2 + fudge));
end
end
So my main question, are there any further changes I can make to optimize this code? Or even an alternative method to approach the problem? I've considered using C++ or F# instead of MATLAB for this section of the program, and I may do so if I cannot get to a reasonable efficiency level with the MATLAB code.
Please also note that at this point I don't have ANY additional toolboxes, if I did then I know this would be trivial (using hist3 from the statistics toolbox for example).
Mem consuming solution
yi = limits(3) + deltay * ( 1:height ) - .5 * deltay;
xi = limits(1) + deltax * ( 1:width ) - .5 * deltax;
dx = bsxfun( #minus, x(:), xi ) .^ 2;
dy = bsxfun( #minus, y(:), yi ) .^ 2;
dist2 = bsxfun( #plus, permute( dy, [2 3 1] ), permute( dx, [3 2 1] ) );
dmap = sum( 1./(dist2 + fudge ) , 3 );
EDIT
handling extremely large x and y by breaking the operation into blocks:
blockSize = 50000; % process up to XX elements at once
dmap = 0;
yi = limits(3) + deltay * ( 1:height ) - .5 * deltay;
xi = limits(1) + deltax * ( 1:width ) - .5 * deltax;
bi = 1;
while bi <= numel(x)
% take a block of x and y
bx = x( bi:min(end, bi + blockSize - 1) );
by = y( bi:min(end, bi + blockSize - 1) );
dx = bsxfun( #minus, bx(:), xi ) .^ 2;
dy = bsxfun( #minus, by(:), yi ) .^ 2;
dist2 = bsxfun( #plus, permute( dy, [2 3 1] ), permute( dx, [3 2 1] ) );
dmap = dmap + sum( 1./(dist2 + fudge ) , 3 );
bi = bi + blockSize;
end
This is a good example of why starting a loop from 1 matters. The only reason that ii and jj are initiated at 0 is to kill the ii * deltay and jj * deltax terms which however introduces sequentiality in the dmap indexing, preventing parallelization.
Now, by rewriting the loops you could use parfor() after opening a matlabpool:
dmap = zeros(height, width);
yi = limits(3) + deltay*(1:height) - .5*deltay;
matlabpool 8
parfor ii = 1: height
for jj = 1: width
xi = limits(1) + (jj-1) * deltax + deltax/2;
dist2 = (x - xi) .^ 2 + (y - yi(ii)) .^ 2;
dmap(ii, jj) = sum(1 ./ (dist2 + fudge));
end
end
matlabpool close
Keep in mind that opening and closing the pool has significant overhead (10 seconds on my Intel Core Duo T9300, vista 32 Matlab 2013a).
PS. I am not sure whether the inner loop instead of the outer one can be meaningfully parallelized. You can try to switch the parfor to the inner one and compare speeds (I would recommend going for the big matrix immediately since you are already running in 12 seconds and the overhead is almost as big).
Alternatively, this problem can be solved in using kernel density estimation techniques. This is part of the Statistics Toolbox, or there's this KDE implementation by Zdravko Botev (no toolboxes required).
For the example code below, I get 0.3 seconds for N = 500000, or 0.7 seconds for N = 1000000.
N = 500000;
data = [randn(N,2); rand(N,1)+3.5, randn(N,1);]; % 2 overlaid distrib
tic; [bandwidth,density,X,Y] = kde2d(data); toc;
imagesc(density);