Implementation of FitzHugh-Nagumo diffusion model diverging by first iteration - java

I'm trying to implement the model described in this paper, which simulates the equation proposed by Alan Turing of the FitzHugh-Nagumo model in 2D as a model for forming animal skin patterns — in other words: simulating two substances diffusing across a surface, how they interact with each other, and what patterns arise. This is the paper's result:
                                            
I've implemented it (my interpretation of at least) in Processing/Java, but it's not working like it should (values are diverging lots, even by the first iteration), so I'm wondering what's going wrong in my implementation (included at the end of the post).
These are the 3 relevant parts from the paper regarding implementation:
1. U & V
There are two substances, u and v (which can be thought of as an activator and inhibitor respectively)
2. Finite difference equations
A fairly simple pixel convolution is defined for each value (pixel) of u and v. Values for a given pixel on the next generation are calculated using both it and its neighbours' current-iteration values.
The value of u on the n+1 iteration for a given pixel (i,j) is defined as:
The value of v on the n+1 iteration for a given pixel (i,j) is defined as:
3. Constants they used
So the problem I'm getting is that the values of u and v are quickly diverging to infinity/NaN (I expect they should stay within 0...1 although the paper doesn't explicitly mention this). v seems to diverge first, taking u along with it, as can be seen here (for some constant index):
0.94296926 0.77225316 // u, v after random initialisation
0.91600573 -62633.082 // values after first iteration -- v has already diverged massively
63.525314 5.19890688E8 // second iteration -- even more divergence
-520088.38 -2.98866172E14 // ...and so on...
1.40978577E14 1.2764294E19
-Infinity -1.7436987E24
NaN NaN
NaN NaN
Code
//Parallel Simulation of Pattern formation in a reactiondiffusion system of Fitzhugh-Nagumo Using GPU CUDA
// Alfredo Gormantara and Pranowo1
static final float a = 2.8e-4;
static final float b = 5.0e-3;
static final float k = -0.005;
static final float tau = 0.1;
static final float delta_t = 1e-3;
float[][] u; // activator
float[][] v; // inhibitor
void setup() {
size(512, 512);
frameRate(5);
u = new float[height][width];
v = new float[height][width];
for (int i = 0; i < u.length; i++) {
for (int j = 0; j < u[0].length; j++) {
u[i][j] = random(1); // random of max of 1 ?
v[i][j] = random(1); // random of max 1?
}
}
loadPixels();
}
void draw() {
float[][] u_n_1 = new float[height][width]; // array holding the n+1 iteration values of u
float[][] v_n_1 = new float[height][width]; // array holding the n+1 iteration values of v
float denom = 2f / width; // 2/MESH_SIZE -- I think mesh_size is dimension of the grid
denom*=denom; // square for denominator
println(u[34][45], v[34][45]); // print vals of one location to see divergence
for (int y = 0; y < height; y++) {
int negative_y_i = y-1 < 0 ? height-1 : y-1; // wrap around grid
for (int x = 0; x < width; x++) {
final float u_n = u[y][x];
final float v_n = v[y][x];
int negative_x_i = x-1 < 0 ? width-1 : x-1; // wrap around grid
// calculate laplace (12)
float u_n_1_laplace = u[y][(x+1) % (width)] + u[y][negative_x_i] + u[(y+1) % (height)][x] + u[negative_y_i][x]; //n+1th iteration
u_n_1_laplace -= (4 * u_n);
u_n_1_laplace /= denom; // divide by (2/DIM)^2
u_n_1_laplace *= a;
// calculate n+1th iteration u value
u_n_1[y][x] = u_n + delta_t*( u_n_1_laplace + u_n -(u_n*u_n*u_n) - v_n + k );
// calculate laplace (14)
float v_n_1_laplace = v[y][(x+1)% (width)] + v[y][negative_x_i] + v[(y+1)% (height)][x] + v[negative_y_i][x]; //n+1th iteration
v_n_1_laplace -= (4 * u_n);
v_n_1_laplace /= denom; // denom is really small, so value goes huge
v_n_1_laplace *=b;
v_n_1[y][x] = v_n + (tau/delta_t)*( v_n_1_laplace + u_n - v_n);
pixels[y*width + x] = color((int) ((u_n_1[y][x]-v_n_1[y][x])*255));
}
}
u = u_n_1.clone(); // copy over new iteration values
v = v_n_1.clone(); // copy over new iteration values
updatePixels();
}

Related

Print numerals in order in a sine wave

Background:
I've successfully written code that generates a sine wave from 0 to 2pi. Adjusting the constants xPrecision and yPrecision, you can stretch the graph horizontally or vertically.
I gain this neat output (in Eclipse), when xPrecision = yPrecision = 10:
My query:
I now wish to display digits 0 to 9 instead of the stars. So, the leftmost star is replaced by 0, the second left-most star is replaced by 1, and so on. When you reach 9, the next digit is again zero.
I am clueless as to how to do this. I have looked at wave patterns like this, but they are fixed width patterns, while mine is scalable.
The only way I can think of is converting my output to a 2D character array, then scraping the *s manually from left to right, and replacing them with the digits, and then printing it. However, this is extremely memory consuming at bigger values of x/yPrecision.
What is the most optimized way to achieve this output?
Code to print sine wave:
class sine {
static final double xPrecision = 10.0; // (1/xPrecision) is the precision on x-values
static final double yPrecision = 10.0; // (1/yPrecision) is the precision on y-values
static final int PI = (int) (3.1415 * xPrecision);
static final int TPI = 2 * PI; // twice PI
static final int HPI = PI / 2; // half PI
public static void main(String[] args) {
double xd;
for(int start = (int) (1 * yPrecision), y = start; y >= -start; y--){
double x0 = Math.asin(y / yPrecision),
x1 = bringXValueWithinPrecision(x0),
x2 = bringXValueWithinPrecision(x0 + TPI / xPrecision),
x3 = bringXValueWithinPrecision(PI/xPrecision - x0);
// for debug
//System.out.println(y + " " + x0 + " " + x1 + " " + x2 + " " + x3);
for(int x = 0; x <= TPI; x++){
xd = (x / xPrecision);
if(x1 == xd || x2 == xd || x3 == xd)
System.out.print("*");
else System.out.print(" ");
}
System.out.println();
}
}
public static double bringXValueWithinPrecision(double num){
// obviously num has 16 floating points
// we need to get num within our precision
return Math.round(num * xPrecision) / xPrecision;
}
}
"Draw" the graph in memory first, then assign digits to its vertical points, and print them in a separate pass.
01
9 2
8 3
7 4
6 5
5 6
4 7
3 8
2 9
1 0
0 1 2
2 1
3 0
4 9
5 8
6 7
7 6
8 5
9 4
0 3
12
See comments in the code for an explanation of how this works:
static final double xPrecision = 10.0; // (1/xPrecision) is the precision on x-values
static final double yPrecision = 10.0; // (1/yPrecision) is the precision on y-values
static final int PI = (int) (3.1415 * xPrecision);
static final int TPI = 2 * PI; // twice PI
static final int HPI = PI / 2; // half PI
public static void main(String[] args) {
// This part is the same as OP's code, except that instead of printing '*'
// it stores the corresponding row number in the array of rows
double xd;
int[] row = new int[100];
Arrays.fill(row, -1);
int r = 0;
int maxc = 0; // Mark the rightmost column of all iterations
for(int start = (int) (1 * yPrecision), y = start; y >= -start; y--){
double x0 = Math.asin(y / yPrecision),
x1 = bringXValueWithinPrecision(x0),
x2 = bringXValueWithinPrecision(x0 + TPI / xPrecision),
x3 = bringXValueWithinPrecision(PI/xPrecision - x0);
int c = 0;
for(int x = 0; x <= TPI; x++, c++){
xd = (x / xPrecision);
// This is where the asterisk used to go
if(x1 == xd || x2 == xd || x3 == xd)
row[c] = r;
}
maxc = Math.max(c, maxc);
r++;
}
// Walk the assigned rows, and give each one a consecutive digit
int[] digit = new int[100];
int current = 0;
for (int i = 0 ; i != 100 ; i++) {
if (row[i] != -1) {
digit[i] = (current++) % 10;
}
}
// Now walk the rows again, this time printing the pre-assigned digits
for (int i = 0 ; i != r ; i++) {
for (int c = 0 ; c != maxc ; c++) {
if (row[c] == i) {
System.out.print(digit[c]);
} else {
System.out.print(' ');
}
}
System.out.println();
}
}
public static double bringXValueWithinPrecision(double num){
// obviously num has 16 floating points
// we need to get num within our precision
return Math.round(num * xPrecision) / xPrecision;
}
The first part of the code fills row[i] array, which contains row for the asterisk in column i. First few numbers from row[] array would look like this:
10 9 8 7 6 5 4 - 3 2 - 1 - - - 0 0 - - - 1 - 2 3 - 4 5 6 7 8 9 10
- denotes cells with -1, which represents a missing value. The array says that the left-most asterisk is on row 10, the next asterisk is on row 9, then 8, 7, 6, and so on. Asterisks 11 and 12 are on row zero, which is at the top.
The second loop walks rows, skips -1s, and assign consecutive digits to all non-negative positions.
The third loop walks the entire field again going row-by-row, printing values from pre-assigned digit[] array when the current row matches the value in the row[] array.
Demo.
If you replace:
System.out.print("*");
with
System.out.print(""+(x%10));
it seems to nearly work.
56
1 0
9 2
8 3
6 5
5 6
4 7
3 8
2 9
1 0
0 1 2
2 1
3 0
4 9
5 8
6 7
7 6
9 4
0 3
2 1
67
Perhaps some further adjustments might get it perfect.
Doing it in a completely different way produces a different picture but achieves your effect.
Essentially,
for each y
for each x
calculate fx = sin(x)
if fx == y print * else print space
It's very inefficient as it calculates sin(x) x*y times when, if you filled a matrix, you could calculate sin(x) just x times.
static final double xPrecision = 10.0; // (1/xPrecision) is the precision on x-values
static final double yPrecision = 10.0; // (1/yPrecision) is the precision on y-values
private void sine() {
for (double y = 1; y >= -1; y -= 1.0 / yPrecision) {
int n = 0;
for (double x = 0; x < 2.0 * Math.PI; x += 1.0 / xPrecision, n++) {
double fx = Math.sin(x);
boolean star = Math.round(fx*xPrecision) == Math.round(y*yPrecision);
System.out.print((star ? ""+(n%10) : " "));
}
System.out.println();
}
}
public void test(String[] args) {
sine();
}
Gives you:
345678
12 901
90 2
8 34
67 5
5 6
4 7
3 8
2 9
1 0
0 1
2 2
3 1
4 0
56 9
7 8
8 67
9 5
01 34
23 12
4567890
Since this is Java, how about let's actually use some objects as objects rather than just as places to define a couple of functions.
Treat your wavy graph as if it is a composition of several different "branches" of the inverse sine function. (Mathematically, that's how we explain the way your version of the program uses Math.asin to produce multiple coordinates for stars.)
Branch 0 is the initial rising part of the curve,
Branch 1 is the falling part of the curve after Branch 0,
Branch 2 is the rising part of the curve after Branch 1, and so forth.
The branches cross the middle line of the output at x values 0,
PI, 2*PI, 3*PI, and so forth.
Depending on how far you want the graph to extend to the right, it is easy to determine how many branches you need.
For example, to plot from 0 to 8*PI you need nine branches
(Branch 0, Branch 8, and the seven branches between those two).
You can implement each branch using an object of some class,
let's call it ArcSineBranch.
It has a constructor, ArcSineBranch(int), that takes the branch number as a parameter.
Create some sort of ordered list (which could just be an ArcSineBranch[] array) and put these branch objects in it,
making sure the branch numbers go in sequence from 0 up to the largest number needed.
You'll also want to implement some way to tell the first ArcSineBranch where its leftmost end is--in the example in the question, the leftmost end of first branch is at y == 0, whereas for all other rising branches it is at y == -start and for all falling branches it is at y == start.
Now you call a mutator function of the first ArcSineBranch that tells it its leftmost symbol is 0. Treat this as an integer (rather than a string) for now to make the arithmetic easier.
You then query the first ArcSineBranch for the rightmost symbol it will write, which it can compute from the leftmost symbol and the number of lines it will write symbols on.
You also query it for the x coordinate of that rightmost symbol.
(The object computes the x-coordinate of the symbol for any y-coordinate by either adding or subtracting a rounded multiple of Math.asin(y / yPrecision) from a multiple of PI.)
Now for each ArcSineBranch in the list, you pass to it the rightmost symbol and x coordinate written by the previous branch.
This ArcSineBranch uses that information to determine the leftmost symbol it writes and the y coordinate of that symbol.
(I am being careful here about the y coordinate in case you choose a value of xPrecision that causes the rightmost x coordinate of one branch to be the same as the leftmost x coordinate of the next; we should only write one symbol at that place in the output, so we want the later branch to skip its leftmost x coordinate and write its leftmost symbol in the next place, one line up or down. But if the x coordinates are different we want the later branch to write a symbol on the same line.)
Now that the later ArcSineBranch "knows" the leftmost symbol it will print and thata symbol's y coordinate, you can query it for its rightmost symbol and x coordinate, and pass those to the next ArcSineBranch, and so forth.
Once you have traversed all the ArcSineBranch objects in this way,
so that each object knows what symbols need to be written for its branch and where to write them, you can loop for (y = start; y >= -start; y--);
within that loop you loop over the list of ArcSineBranch objects;
for each object you query whether it requires a symbol to be written at
y-coordinate y.
If the object requires a symbol to be written,
you query which symbol to write at which x-coordinate,
then space the output to the right until you reach that x-coordinate and write that symbol there.
But of course, first check that this would not plot a symbol beyond the
right-hand edge of the desired graph.
(This check really only applies to the last ArcSineBranch, so you can optimize the code a bit by looping over the other branches first and then dealing with the last ArcSineBranch separately.)
I've already described this algorithm in more detail than I initially wanted to. There should be enough information here to code this into Java in a relatively straightforward way, though there are still some localized details to be worked out.
Note that the design in this answer is intended to use the same mathematical ideas as the code in the question uses to decide where to plot the points.
Specifically, ArcSineBranch(0) produces the x1 values from the original code, ArcSineBranch(1) produces the x3 values, and ArcSineBranch(2) produces the x2 values.
The implementation of this design should plot a digit at the location of each star plotted by the original code, and should plot no other digits.
Care about a different approach?
3030
28 28
26 26
22 22
18 18
12 12
06 06
00 00 00
06 06
12 12
18 18
22 22
26 26
28 28
3030
Solution:
import static java.lang.Math.sin;
import static java.lang.Math.PI;
import static java.lang.Math.abs;
public class Sine {
static final Integer points = 30; // points on x and y axis
public static void main(String[] args) {
// contains graph points
Boolean[][] graph = new Boolean[points + 1][points + 1];
for (Double x = 0.0; x <= points; x++) {
// x axis pi value
Double pi = (x / points) * 2 * PI;
// sin(x) plot for x
Integer sinx = (int) Math.round((sin(pi) * points / 2) + points / 2);
graph[sinx][x.intValue()] = true;
}
for (Integer i = 0; i <= points; i++) {
for (Integer j = 0; j <= points; j++) {
// space characters on x axis
Integer pt = (int) Math.floor(Math.log10(points) + 1);
String space = String.format("%" + pt + "s", " ");
// padding for p
String p = String.format("%0" + (pt) + "d", abs(i - points / 2) * 2);
System.out.print(graph[i][j] != null ? p : space);
}
System.out.println();
}
}
}
Approach:
points contains the number of characters on x and y axis.
graph contains true or null for each x and y characters.
1st for loop:
Since the value of x in sine graph is from 0 to 2π, we need to convert x accordingly. So pi contains the value of the same range but according to x.
sinx is the Integer value according to x.
No need to explain graph[sinx][x.intValue()] = true;.
2nd for loops:
1st for loop
Execute LOOPLABEL.
Break to next line at the end.
2nd for loop(LOOPLABEL)
pt holds the number for padding on y axis.
space is the space characters to be printed on y axis.
p is the converted range between 0 to points.
Printing graph[i][j]
DEMO
By using the fact that each row has one point (on each slope), you can calculate which digit to display at each point without using extra memory or loops. Here's my example. Note that I only checked that this example only works if xPrecision and yPrecision are integers. You'll have to modify it if you want to use doubles.
class sine {
static final double xPrecision = 10.0; // (1/xPrecision) is the precision on x-values
static final double yPrecision = 10.0; // (1/yPrecision) is the precision on y-values
static final int PI = (int) Math.round(Math.PI * xPrecision);
static final int TPI = 2 * PI; // twice PI
static final int HPI = PI / 2; // half PI
static final int cycles = 2; // prints from x=0 to 2*cycles*pi
public static void main(String[] args) {
double xd;
int cycleoffset, cycleoffset2, topbottomoffset = 1;
for (int start = (int) (1 * yPrecision), y = start; y >= -start; y--) {
double x0 = Math.asin(y / yPrecision), x1 = bringXValueWithinPrecision(x0),
x2 = bringXValueWithinPrecision(x0 + TPI / xPrecision),
x3 = bringXValueWithinPrecision(PI / xPrecision - x0), tmp;
if (y == start) {
if (x1 == x3) // when there is only one point at the top/bottom
topbottomoffset = 0;
else if (x1 > x3) // swap x1 and x3
{
tmp = x1;
x1 = x3;
x3 = tmp;
}
} else if (y == -start) {
// I don't think this is needed, but just for safety make sure there is only one point on the bottom if there was only one point at the top
if (topbottomoffset == 0)
x2 = x3;
else if (x2 < x3) // swap x2 and x3
{
tmp = x2;
x2 = x3;
x3 = tmp;
}
}
cycleoffset = (int) (4 * yPrecision + 2 * topbottomoffset);
cycleoffset2 = -cycleoffset;
int start1 = topbottomoffset + 2 * (int) yPrecision, start2 = 2 * topbottomoffset + 4 * (int) yPrecision;
for (int x = 0, lim = cycles * TPI; x <= lim; x++) {
xd = ((x % TPI) / xPrecision);
if (x % TPI == 0)
cycleoffset2 += cycleoffset;
// x = 0 to pi/2
if (x1 == xd)
System.out.print((cycleoffset2 + y) % 10);
// x = 3pi/2 to 2pi
else if (x2 == xd)
System.out.print((cycleoffset2 + start2 + y) % 10);
// x = pi/2 to 3pi/2
else if (x3 == xd)
System.out.print((cycleoffset2 + start1 - y) % 10);
else
System.out.print(" ");
}
System.out.println();
}
}
public static double bringXValueWithinPrecision(double num) {
// obviously num has 16 floating points
// we need to get num within our precision
return Math.round(num * xPrecision) / xPrecision;
}
}
EDIT
The digits for the different ranges are calculated as follows
0 < x < π/2 : This one is simplest since it is the first range. Since the middle row is y=0 and that is where the sine wave starts, we can just use y to find the digit.
π/2 < x < 3π/2 : The digits here count up as we go down, but y decreases as we go down. So we have to use a -y term. On the top row, y=yPrecision, and the last digit from the previous range was yPrecision. So we use 2*yPrecision - y, because that includes the -y, and is equal to yPrecision at the first term (where y=yPrecision).
3π/2 < x < 2π : The digits here count down as we go down, so we need a +y term, but the tricky part is figuring where to start. Since the sine wave by this point has gone from 0 to yPrecision to 0 to -yPrecision, the bottom point (x=3π/2) should start at 3*yPrecision. Since y=-yPrecision at the bottom point, we use 4*yPrecision + y, since that includes a +y and is equal to 3*yPrecision at the first term (where y=-yPrecision).
The topbottomoffset term : Depending on the values used for xPrecision and yPrecision, there can be one or two points plotted on the top and bottom rows. If there are two points, we need to add one to digits in the π/2 to 3π/2 range, and two to the digits in the 3π/2 to 2π range.
The cycleoffset term : If multiple cycles of the sine wave are plotted, additional cycles need to start from the last digit used in the previous cycle. Each cycle goes from 0 to yPrecision to 0 to -yPrecision to 0, which is equal to 4*yPrecision. So each new cycle needs to start at 4*yPrecision*[the number of previous cycles]. If there are two points on the top and bottom rows, those need to be factored in as well.
Swapping values: When there are two points on the top row, then x1>x3. This happens because when y=yPrecision, we're taking Math.asin(1), which happens to be exactly pi/2=1.5707963267948966 in Java's double system. On lower xPrecision (<100.0), the rounding done by bringXValueWithinPrecision brings x1 up to 1.58 while x3 down to nearly 1.56. Hence, they need to be swapped in order to get the correct numerical order.
Here's my solution, which basically uses the half of the sine in 4 for loops:
from half to 0
from 0 to half
from half to the end
from the end to the half
And in each loop replace only the first asterisk.
class sine {
static final double xPrecision = 14.0; // (1/xPrecision) is the precision on x-values
static final double yPrecision = 14.0; // (1/yPrecision) is the precision on y-values
static final int PI = (int)(3.1415 * xPrecision);
static final int TPI = 2 * PI; // twice PI
static final int HPI = PI / 2; // half PI
public static void main(String[] args) {
double xd;
String str = "";
for (int start = (int)(1 * yPrecision), y = start; y >= -start; y--) {
double x0 = Math.asin(y / yPrecision),
x1 = bringXValueWithinPrecision(x0),
x2 = bringXValueWithinPrecision(x0 + TPI / xPrecision),
x3 = bringXValueWithinPrecision(PI / xPrecision - x0);
// for debug
//System.out.println(y + " " + x0 + " " + x1 + " " + x2 + " " + x3);
for (int x = 0; x <= TPI; x++) {
xd = (x / xPrecision);
if (x1 == xd || x2 == xd || x3 == xd)
str += "*";
else str += " ";
}
str += "\n";
}
String[] rows = str.split("\n");
int half = (int)(1 * yPrecision);
// we use this half in for loops, from half to 0, from 0 to half, from half to the end and from the end to the half, and replace only the first asterisk.
int val = 0;
for (int i = half; i >= 0; i--) {
if (val == 10) val = 0;
rows[i] = rows[i].replaceFirst("\\*", Integer.toString(val++));
}
for (int i = 0; i <= half; i++) {
if (val == 10) val = 0;
rows[i] = rows[i].replaceFirst("\\*", Integer.toString(val++));
}
for (int i = half + 1; i < rows.length; i++) {
if (val == 10) val = 0;
rows[i] = rows[i].replaceFirst("\\*", Integer.toString(val++));
}
for (int i = rows.length - 1; i >= half; i--) {
if (val == 10) val = 0;
rows[i] = rows[i].replaceFirst("\\*", Integer.toString(val++));
}
System.out.println(String.join("\n", rows));
}
public static double bringXValueWithinPrecision(double num) {
// obviously num has 16 floating points
// we need to get num within our precision
return Math.round(num * xPrecision) / xPrecision;
}
}
Result:
01
9 2
8 3
7 4
6 5
5 6
4 7
3 8
2 9
1 0
0 1 2
2 1
3 0
4 9
5 8
6 7
7 6
8 5
9 4
0 3
12
Add a counter in your loop and reset it when 9 is reached:
for(int x = 0, counter = 0; x <= TPI; x++, counter++){
xd = (x / xPrecision);
if(x1 == xd || x2 == xd || x3 == xd) {
System.out.print("" + counter);
if (counter == 9) {
counter = 0;
}
} else {
System.out.print(" ");
}
}

Find longest distance from a certain point (java, 2d diagram)

I'm working on a clustering program in Java. I'm trying to find the point that has the longest distance from another point in an 2-dimentional diagram with x and y axis.
I though I could use pytagoras:
Where the square of the Y-axis of the starting point + the square of the X-axis of the of the other points will determine the distance between them.
What my code does is for this certain point, check all other points to see if it finds a point with a higher distance. The code I have at the moment is the following:
// The points to find longest distance from
x_coord = 2;
y_coord = 4;
// Need to find right size
double var1 = Math.pow(y_coord, 2); // square of Y
double var2 = 0;
double var3 = 0;
int sum = 0;
/* For all coords ( of the cluster clusters)
* coordsX is an array that holds all the X coordinates
* of all other points
*/
for (int k = 0; k < coordsX.length; k++){
// Check which is furthest away from
var2 = Math.pow(coordsX[k], 2); // square of X
var3 = var1 + var2; // Sum of var1 and var2
sum = (int)Math.sqrt(var3); // Square root to find distance
if (sum > longestDistance){
longestDistance = sum;
}
}
Does anyone have any suggestions what might be wrong? Or is this an unsuited method to calculate distances?
So, to find the distance between two points, say A and B located on a xy plane, where A and B are indices, here is what you need to do:
double distanceSquared = Math.pow(coordsX[A] - coordsX(B), 2) + Math.pow(coordsY[A] - coordsY(B), 2);
And if you just want to find the most distant point, there is no need to take a square root, because it is a monotonic function, so it does not change the result of comparison at all.
Simply compare the distance squares
EDIT: code for you.
double longestDistanceSquared = 0;
int mostDistantPointIndex = -1;
for (int k = 0; k < coordsX.length; k++){
double distanceSquared = Math.pow(coordsX[k] - x_coord, 2) + Math.pow(coordsY[k] - y_coord, 2);
if (distanceSquared > longestDistanceSquared){
longestDistanceSquared = distanceSquared;
mostDistantPointIndex = k;
}
}
The first thing that jumps out of the code to me is that you are casting the result of the sqrt() call to an int. That seems like it's going to cause problems.

Reduce treatment time of the FFT

I'm currently working on Java for Android. I try to implement the FFT in order to realize a kind of viewer of the frequencies.
Actually I was able to do it, but the display is not fluid at all.
I added some traces in order to check the treatment time of each part of my code, and the fact is that the FFT takes about 300ms to be applied on my complex array, that owns 4096 elements. And I need it to take less than 100ms, as my thread (that displays the frequencies) is refreshed every 100ms. I reduced the initial array in order that the FFT results own only 1028 elements, and it works, but the result is deprecated.
Does someone have an idea ?
I used the default fft.java and Complex.java classes that can be found on the internet.
For information, my code computing the FFT is the following :
int bytesPerSample = 2;
Complex[] x = new Complex[bufferSize/2] ;
for (int index = 0 ; index < bufferReadResult - bytesPerSample + 1; index += bytesPerSample)
{
// 16BITS = 2BYTES
float asFloat = Float.intBitsToFloat(asInt);
double sample = 0;
for (int b = 0; b < bytesPerSample; b++) {
int v = buffer[index + b];
if (b < bytesPerSample - 1 || bytesPerSample == 1) {
v &= 0xFF;
}
sample += v << (b * 8);
}
double sample32 = 100 * (sample / 32768.0); // don't know the use of this compute...
x[index/bytesPerSample] = new Complex(sample32, 0);
}
Complex[] tx = new Complex[1024]; // size = 2048
///// reduction of the size of the signal in order to improve the fft traitment time
for (int i = 0; i < x.length/4; i++)
{
tx[i] = new Complex(x[i*4].re(), 0);
}
// Signal retrieval thanks to the FFT
fftRes = FFT.fft(tx);
I don't know Java, but you're way of converting between your input data and an array of complex values seems very convoluted. You're building two arrays of complex data where only one is necessary.
Also it smells like your complex real and imaginary values are doubles. That's way over the top for what you need, and ARMs are veeeery slow at double arithmetic anyway. Is there a complex class based on single precision floats?
Thirdly you're performing a complex fft on real data by filling the imaginary part of your complexes with zero. Whilst the result will be correct it is twice as much work straight off (unless the routine is clever enough to spot that, which I doubt). If possible perform a real fft on your data and save half your time.
And then as Simon says there's the whole issue of avoiding garbage collection and memory allocation.
Also it looks like your FFT has no preparatory step. This mean that the routine FFT.fft() is calculating the complex exponentials every time. The longest part of the FFT calculation is working out the complex exponentials, which is a shame because for any given FFT length the exponentials are constants. They don't depend on your input data at all. In the real time world we use FFT routines where we calculate the exponentials once at the start of the program and then the actual fft itself takes that const array as one of its inputs. Don't know if your FFT class can do something similar.
If you do end up going to something like FFTW then you're going to have to get used to calling C code from your Java. Also make sure you get a version that supports (I think) NEON, ARM's answer to SSE, AVX and Altivec. It's worth ploughing through their release notes to check. Also I strongly suspect that FFTW will only be able to offer a significant speed up if you ask it to perform an FFT on single precision floats, not doubles.
Google luck!
--Edit--
I meant of course 'good luck'. Give me a real keyboard quick, these touchscreen ones are unreliable...
First, thanks for all your answers.
I followed them and made two test :
first one, I replace the double used in my Complex class by float. The result is just a bit better, but not enough.
then I've rewroten the fft method in order not to use Complex anymore, but a two-dimensional float array instead. For each row of this array, the first column contains the real part, and the second one the imaginary part.
I also changed my code in order to instanciate the float array only once, on the onCreate method.
And the result... is worst !! Now it takes a little bit more than 500ms instead of 300ms.
I don't know what to do now.
You can find below the initial fft fonction, and then the one I've re-wroten.
Thanks for your help.
// compute the FFT of x[], assuming its length is a power of 2
public static Complex[] fft(Complex[] x) {
int N = x.length;
// base case
if (N == 1) return new Complex[] { x[0] };
// radix 2 Cooley-Tukey FFT
if (N % 2 != 0) { throw new RuntimeException("N is not a power of 2 : " + N); }
// fft of even terms
Complex[] even = new Complex[N/2];
for (int k = 0; k < N/2; k++) {
even[k] = x[2*k];
}
Complex[] q = fft(even);
// fft of odd terms
Complex[] odd = even; // reuse the array
for (int k = 0; k < N/2; k++) {
odd[k] = x[2*k + 1];
}
Complex[] r = fft(odd);
// combine
Complex[] y = new Complex[N];
for (int k = 0; k < N/2; k++) {
double kth = -2 * k * Math.PI / N;
Complex wk = new Complex(Math.cos(kth), Math.sin(kth));
y[k] = q[k].plus(wk.times(r[k]));
y[k + N/2] = q[k].minus(wk.times(r[k]));
}
return y;
}
public static float[][] fftf(float[][] x) {
/**
* x[][0] = real part
* x[][1] = imaginary part
*/
int N = x.length;
// base case
if (N == 1) return new float[][] { x[0] };
// radix 2 Cooley-Tukey FFT
if (N % 2 != 0) { throw new RuntimeException("N is not a power of 2 : " + N); }
// fft of even terms
float[][] even = new float[N/2][2];
for (int k = 0; k < N/2; k++) {
even[k] = x[2*k];
}
float[][] q = fftf(even);
// fft of odd terms
float[][] odd = even; // reuse the array
for (int k = 0; k < N/2; k++) {
odd[k] = x[2*k + 1];
}
float[][] r = fftf(odd);
// combine
float[][] y = new float[N][2];
double kth, wkcos, wksin ;
for (int k = 0; k < N/2; k++) {
kth = -2 * k * Math.PI / N;
//Complex wk = new Complex(Math.cos(kth), Math.sin(kth));
wkcos = Math.cos(kth) ; // real part
wksin = Math.sin(kth) ; // imaginary part
// y[k] = q[k].plus(wk.times(r[k]));
y[k][0] = (float) (q[k][0] + wkcos * r[k][0] - wksin * r[k][1]);
y[k][1] = (float) (q[k][1] + wkcos * r[k][1] + wksin * r[k][0]);
// y[k + N/2] = q[k].minus(wk.times(r[k]));
y[k + N/2][0] = (float) (q[k][0] - (wkcos * r[k][0] - wksin * r[k][1]));
y[k + N/2][1] = (float) (q[k][1] - (wkcos * r[k][1] + wksin * r[k][0]));
}
return y;
}
actually I think I don't understand everything.
First, about Math.cos and Math.sin : how do you want me not to compute it each time ? Do you mean that I should instanciate the whole values only once (e.g store it in an array) and use them for each compute ?
Second, about the N % 2, indeed it's not very useful, I could make the test before the call of the function.
Third, about Simon's advice : I mixed what he said and what you said, that's why I've replaced the Complex by a two-dimensional float[][]. If that was not what he suggested, then what was it ?
At least, I'm not a FFT expert, so what do you mean by making a "real FFT" ? Do you mean that my imaginary part is useless ? If so, I'm not sure, because later in my code, I compute the magnitude of each frequence, so sqrt(real[i]*real[i] + imag[i]*imag[i]). And I think that my imaginary part is not equal to zero...
thanks !

How to re-implement sin() method in Java ? (to have results close to Math.sin() )

I know Math.sin() can work but I need to implement it myself using factorial(int) I have a factorial method already below are my sin method but I can't get the same result as Math.sin():
public static double factorial(double n) {
if (n <= 1) // base case
return 1;
else
return n * factorial(n - 1);
}
public static double sin(int n) {
double sum = 0.0;
for (int i = 1; i <= n; i++) {
if (i % 2 == 0) {
sum += Math.pow(1, i) / factorial(2 * i + 1);
} else {
sum += Math.pow(-1, i) / factorial(2 * i + 1);
}
}
return sum;
}
You should use the Taylor series. A great tutorial here
I can see that you've tried but your sin method is incorrect
public static sin(int n) {
// angle to radians
double rad = n*1./180.*Math.PI;
// the first element of the taylor series
double sum = rad;
// add them up until a certain precision (eg. 10)
for (int i = 1; i <= PRECISION; i++) {
if (i % 2 == 0)
sum += Math.pow(rad, 2*i+1) / factorial(2 * i + 1);
else
sum -= Math.pow(rad, 2*i+1) / factorial(2 * i + 1);
}
return sum;
}
A working example of calculating the sin function. Sorry I've jotted it down in C++, but hope you get the picture. It's not that different :)
Your formula is wrong and you are getting a rough result of sin(1) and all you're doing by changing n is changing the accuracy of this calculation. You should look the formula up in Wikipedia and there you'll see that your n is in the wrong place and shouldn't be used as the limit of the for loop but rather in the numerator of the fraction, in the Math.pow(...) method. Check out Taylor Series
It looks like you are trying to use the taylor series expansion for sin, but have not included the term for x. Therefore, your method will always attempt to approximate sin(1) regardless of argument.
The method parameter only controls accuracy. In a good implementation, a reasonable value for that parameter is auto-detected, preventing the caller from passing to low a value, which can result in highly inaccurate results for large x. Moreover, to assist fast convergence (and prevent unnecessary loss of significance) of the series, implementations usually use that sin(x + k * 2 * PI) = sin(x) to first move x into the range [-PI, PI].
Also, your method is not very efficient, due to the repeated evaluations of factorials. (To evaluate factorial(5) you compute factorial(3), which you have already computed in the previous iteration of the for-loop).
Finally, note that your factorial implementation accepts an argument of type double, but is only correct for integers, and your sin method should probably receive the angle as double.
Sin (x) can be represented as Taylor series:
Sin (x) = (x/1!) – (x3/3!) + (x5/5!) - (x7/7!) + …
So you can write your code like this:
public static double getSine(double x) {
double result = 0;
for (int i = 0, j = 1, k = 1; i < 100; i++, j = j + 2, k = k * -1) {
result = result + ((Math.pow(x, j) / factorial (j)) * k);
}
return result;
}
Here we have run our loop only 100 times. If you want to run more than that you need to change your base equation (otherwise infinity value will occur).
I have learned a very good trick from the book “How to solve it by computer” by R.G.Dromey. He explain it like this way:
(x3/3! ) = (x X x X x)/(3 X 2 X 1) = (x2/(3 X 2)) X (x1/1!) i = 3
(x5/5! ) = (x X x X x X x X x)/(5 X 4 X 3 X 2 X 1) = (x2/(5 X 4)) X (x3/3!) i = 5
(x7/7! ) = (x X x X x X x X x X x X x)/(7 X 6 X 5 X 4 X 3 X 2 X 1) = (x2/(7 X 6)) X (x5/5!) i = 7
So the terms (x2/(3 X 2)) , (x2/(5 X 4)), (x2/(7 X 6)) can be expressed as x2/(i X (i - 1)) for i = 3,5,7,…
Therefore to generate consecutive terms of the sine series we can write:
current ith term = (x2 / ( i X (i - 1)) ) X (previous term)
The code is following:
public static double getSine(double x) {
double result = 0;
double term = x;
result = x;
for (int i = 3, j = -1; i < 100000000; i = i + 2, j = j * -1) {
term = x * x * term / (i * (i - 1));
result = result + term * j;
}
return result;
}
Note that j variable used to alternate the sign of the term .

Convolution Filter - Float Precision C Vs Java

I'm porting a library of image manipulation routines into C from Java and I'm getting some very small differences when I compare the results. Is it reasonable that these differences are in the different languages' handling of float values or do I still have work to do!
The routine is Convolution with a 3 x 3 kernel, it's operated on a bitmap represented by a linear array of pixels, a width and a depth. You need not understand this code exactly to answer my question, it's just here for reference.
Java code;
for (int x = 0; x < width; x++){
for (int y = 0; y < height; y++){
int offset = (y*width)+x;
if(x % (width-1) == 0 || y % (height-1) == 0){
input.setPixel(x, y, 0xFF000000); // Alpha channel only for border
} else {
float r = 0;
float g = 0;
float b = 0;
for(int kx = -1 ; kx <= 1; kx++ ){
for(int ky = -1 ; ky <= 1; ky++ ){
int pixel = pix[offset+(width*ky)+kx];
int t1 = Color.red(pixel);
int t2 = Color.green(pixel);
int t3 = Color.blue(pixel);
float m = kernel[((ky+1)*3)+kx+1];
r += Color.red(pixel) * m;
g += Color.green(pixel) * m;
b += Color.blue(pixel) * m;
}
}
input.setPixel(x, y, Color.rgb(clamp((int)r), clamp((int)g), clamp((int)b)));
}
}
}
return input;
Clamp restricts the bands' values to the range [0..255] and Color.red is equivalent to (pixel & 0x00FF0000) >> 16.
The C code goes like this;
for(x=1;x<width-1;x++){
for(y=1; y<height-1; y++){
offset = x + (y*width);
rAcc=0;
gAcc=0;
bAcc=0;
for(z=0;z<kernelLength;z++){
xk = x + xOffsets[z];
yk = y + yOffsets[z];
kOffset = xk + (yk * width);
rAcc += kernel[z] * ((b1[kOffset] & rMask)>>16);
gAcc += kernel[z] * ((b1[kOffset] & gMask)>>8);
bAcc += kernel[z] * (b1[kOffset] & bMask);
}
// Clamp values
rAcc = rAcc > 255 ? 255 : rAcc < 0 ? 0 : rAcc;
gAcc = gAcc > 255 ? 255 : gAcc < 0 ? 0 : gAcc;
bAcc = bAcc > 255 ? 255 : bAcc < 0 ? 0 : bAcc;
// Round the floats
r = (int)(rAcc + 0.5);
g = (int)(gAcc + 0.5);
b = (int)(bAcc + 0.5);
output[offset] = (a|r<<16|g<<8|b) ;
}
}
It's a little different xOffsets provides the xOffset for the kernel element for example.
The main point is that my results are out by at most one bit. The following are pixel values;
FF205448 expected
FF215449 returned
44 wrong
FF56977E expected
FF56977F returned
45 wrong
FF4A9A7D expected
FF4B9B7E returned
54 wrong
FF3F9478 expected
FF3F9578 returned
74 wrong
FF004A12 expected
FF004A13 returned
Do you believe this is a problem with my code or rather a difference in the language?
Kind regards,
Gav
After a quick look:
do you realize that (int)r will floor the r value instead of rounding it normally?
in the c code, you seem to use (int)(r + 0.5)
Further to Fortega's answer, try the roundf() function from the C math library.
Java's floating point behaviour is quite precise. What I expect to be happening here is that the value as being kept in registers with extended precision. IIRC, Java requires that the precision is rounded to that of the appropriate type. This is to try to make sure you always get the same result (full details in the JLS). C compilers will tend to leave any extra precision there, until the result in stored into main memory.
I would suggest you use double instead of float. Float is almost never the best choice.
This might be due to different default round in the two languages. I'm not saying they have (you need to read up to determine that), but it's an idea.

Categories